•
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
be treated to a considerable degree of satisfaction perturbatively. in Feynman diagrams in quantum electrodynamics). The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. the aforementioned exactly solvable cases. whereas for the calculation of discrete eigenvalues the Schrodinger equation. again point the way: For scattering problems the path integral seems particularly convenient. With various techniques and deeper studies. but also about complex energies that he encounters in a parallel course on nuclear physics. Thus this first part . To what extent the two methods are actually equivalent.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. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. 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. With the growing importance of models in statistical mechanics and in field theory. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. However. These basic cases will be dealt with in detail by both methods in this text. 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. has not always been understood well. In the first part. the path integral method of Feynman was soon recognized to offer frequently a more general procedure of enforcing first quantization instead of the Schrodinger equation. we recapitulate the origin of quantum mechanics. basic postulates and standard applications. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. even though the expansions are mostly asymptotic. numerous problems could. one problem being that there are few nontrivial models which permit a deeper insight into their connection.e. in fact. but arise also in recently studied models of large spins. that is the Coulomb potential and the harmonic oscillator. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. screened Coulomb potentials and maybe singular potentials. with degenerate vacua) were first and more easily derived from the Schrodinger equation.g. Chapters 1 to 14. like periodic potentials. Excluding spin. its mathematical foundations.
In the second part. 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). Thereafter the concepts of instantons. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states.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. we deal mostly with applications depending on perturbation theory. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. 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. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. 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. and the elliptic or Lame potential — here introduced earlier as a generaliza . We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators.e. 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. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. After a treatment of power potentials. 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. which is presumably the only such singular case permitting complete solution and was achieved only recently. the chapter thereafter deals with Yukawa potentials. i. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). The most prominent examples here are the double well potential and its inverted form. and their eigenvalues. Chapters 15 to 20. periodic instantons. The following chapters deal with the derivation of level splitting formulas (including excited states) for periodic potentials and anharmonic oscillators and — in the oneloop approximation considered — are shown to agree with those obtained by perturbation theory with associated boundary conditions. The potentials with degenerate minima will be seen to reappear throughout the text.
and whenever available also with the results of WKB calculations. an approximation whose higher . 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. 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. quartic. 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. such as limiting procedures. This method is therefore more complicated. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. for instance. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. In fact. This puts Schrodinger equations with e.g. the Mathieu equation. With a fully systematic perturbation method and with applied boundary conditions. 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. however. from our precise reference to unavoidable elliptic integrals taken from Tables). this comparison on a transparent level being one of the main aims of this text. We then illustrate the relevance of this in the method of collective coordinates. except that these are no longer of hypergeometric type. All results are compared with those obtained by perturbation theory. we can make the following observations. it is. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. 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. cubic). The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. for instance. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics.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. anharmonic oscillator potentials on a comparable level with.
XIX order contributions are difficult to obtain. T. Liang (Taiyuan). . Since this comparison was the guideline in writing the text.g. not the least for permitting a more detailed and hopefully comprehensible presentation here. at the end of a chapter (after troublesome turning of pages). J. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. Throughout the text some calculations which require special attention. K. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. N. Park (Masan). B. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. Andrews).g. like E. MiillerKirsten *In the running text references are cited like e. For ease of reading. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. As a rule.* H. The author is deeply indebted to his onetime supervisor Professor R. Dingle (then University of Western Australia. e. Dingle for paving him the way into this field which — though not always at the forefront of current research (including the author's) — repeatedly triggered recurring interest to return to it. the references referred to are never cited by mere numbers which have to be identified e.g. Instead a glance at a nearby footnote provides the reader immediately the names of authors.Q. H. Nonetheless. The line of thinking underlying this text grew out of the author's association with Professor R. 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. shows him that each such formula here has been properly looked up). Tchrakian (Dublin) and Jianzu Zhang (Shanghai). which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. we also consider at various points of the text comparisons with WKB approximations. as well as applications and illustrations. D. Their deep involvement in the attempt described here is evident from the cited bibliography. B. Whittaker and Watson [283]. Watson [283]. in particular Professors J. are relegated to separate subsections which — lacking a better name — we refer to as Examples. D. thereafter University of St. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). W. with the source given in the bibliography at the end. In writing this text the author considered it of interest to demonstrate the parallel application of both the Schrodinger equation and the path integral to a selection of basic problems. also for the verification of results. Whittaker and G.
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. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). Kirchhoff's law in thermodynamics says that in the case of equilibrium. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it.Chapter 1 Introduction 1. very few explain in this context what he really did in view of involvement with statistical mechanics. Instead.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). 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. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. We do not enter here into detailed considerations of Planck. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators.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.
T) were known and tested experimentally.1) c is the velocity of light with c = u\.T) = dv3eC2U/T. the formula (to be explained) u(u. radiators.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. kT kXT (1. h = 6. T)du is the mean energy density (i.e.e. Black bodies as good absorbers are therefore also good emitters. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v.626 x 10 . i. These expressions are: (1) Wien's law. and then measuring the increase in temperature of the heat bath. Fig. The parameters k and h are the constants of Boltzmann and Planck: k = 1.1) ' y Here u(v. and the other in the region of large A (or AT).3 4 J s. Introduction emits. How did Planck arrive at the expression (1.1 Absorption in a cavity. v + dv in equilibrium with the black body at temperature T..2 CHAPTER 1. 1. (1.38 x 1(T 23 J K'1. y J c 3 \ex .2) and the . u(u. It was found that one expression agreed well with observations in the region of small A (or AT). (1. A being the wavelength of the radiation.e. energy per unit volume) of the radiation (i.l ) where x = ^ = ^ .e. In Eq. i.T) = 2*?£(?)kT. 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. Let us look at the final result of Planck.
C2. 1. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. in the first place Planck had tried to find an expression linking both. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. (1.3) Ci. Indeed. and he had succeeded in finding such an expression of the form u(v. he searched for a derivation. that the formulas (1.2) and (1. T) u(i/.T) 2^^kT. "small" (i. 3 (1. discretization begins to enter. but — here the discreteness appears properly — only in elements (quanta) e. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom.3) are contained in Eq.47TZ/ 2 {x small). When Planck had found this expression.1. which are indistinguishable) among the N indistinguishable oscillators at . C3 being constants. e xhv. (x large).2 Distributing quanta (dots) among oscillators (boxes).1) as approximations. Considering Eq.1) in regions of a. Here W is a number which determines the distribution of the energy among a discrete number of objects. where the Fig. and thus over a discrete number of admissible states.e. (1. .T) = av e6"/Ti' where a and b are constants. This is the point. exp(x) ~ 1+x) and "large" (exp(—x) < 1). We see.T) = 2^C3T. we obtain: u(i/. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". To this end he considered Boltzmann's formula S — klnW for the entropy S.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>.
2. M. W. i. as for small values of e). p. . Oberhettinger [181]. (1. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. as in most other Tables. (1.T)dv.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. there not called Stirling's formula. 1.4) IniV! ~ JVlniViV + O(0). Agreement with Eq. formula 8. Ryzhik [122]. Gradshteyn and I.e. We now obtain the energy density of the radiation. N * oo. one obtains (cf. u(i>. 940. . with the quanta represented schematically by dots as indicated in Fig.3. S.4 CHAPTER 1. 1. v + dv. The Stirling formula or approximation will appear frequently in later chapters. e.2) requires that e ex is. p. with riydu — 2 x —w—dv. h = const. I. Magnus and F.7) *See e.g. We visualize the iV oscillators as boxes separated by N — 1 walls.1)!P! (1. = 1/T). Introduction temperature T.e.3 Comparing the polarization modes with those of a 2dimensional oscillator.6) Fig.343(2). and the second law of thermodynamics ((dS/dU)v Example 1. Then W is given by w = (N With the help of Stirling's formula* {N + piy. U{T) being the average energy emitted by one oscillator. i. (1. e = his.g.
where^ 2 [2KUY . so that the derivative of u implies (x as in Eq. (1.2.1). The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^. i. We obtain the expression (1. In terms of A we have u(X.i = 1. '''From the equation I \ JW . is given by . as in electrodynamics.2. where we have for the electric field E oc elwt \ J eK sin KI^I sin K2X2 sin K3X3 K with the boundary condition that at the walls E = 0 at Xi = 0. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.3.3. v + dv. as indicated in Fig.. 2 2 2 r2 L K — 7T n .1. so that .T) has a maximum which follows from du/dX = 0 (with c = vX). We obtain therefore u^T) = Unv = 2^fJ^—i. The number with frequency v in the interval v. so that (lvL\A 0 I I = rr. .. L for i = 1. (1.1)) The solutions of this equation are ^max = 4. nvdv per unit volume.T)dX = ^ehc/*kT_idX. KT = I J . .2.7) for instance.e.8) This is Planck's formula (1. Then L^j = nrii. dj\l dM ..V 2 ) E = 0.7).965 and xmin = 0.UJ = 2KV.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.. rii = 1..3.3 (as for ideal conductors).UJ2/C2 + K? = 0. We observe that u(v..1. (1.
Since temperature originates through contact with other oscillators. However.10) Thus here the socalled zero point energy appears. . Later it was realized by H. of course.6 The first value yields AmaxT = CHAPTER 1. and from which the constant h can be determined from the known value of k.8) could be derived much more easily in the context of statistical mechanics.1. the behaviour of the system at zero absolute temperature. but one can expect an analogy. which can not be eliminated without a different approach. . i. such a procedure leads to contradictions. A. x — oo) the mean energy vanishes (0 < U < * > oo). We are not dealing with the linear harmonic oscillator familiar from mechanics here. One expects.e. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. One might suppose now. that of an oscillation system at absolute temperature T ^ 0. Thus we have a rather complicated system here. . ra = 0. which can assume the discrete energies en — nhv. then (with x = hv/kT) its mean energy is En=0e = nX dx ^0 — /ii/— In = huf r%e dx 1 — e_x (1 — e~x)z hv . that it is easier to consider first the case of T = 0. We therefore examine such contradictions next. .—v <L9) We observe that for T — 0 (i. 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=—. Lorentz and Planck that Eq. which had also been known before Planck's discovery. (1. n = 0. Introduction he 4.965K = Const. . 2 . which did not arise in Planck's consideration of 1900.e. we then have at T = 0 independent oscillators.10). This is Wien's displacement law. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1.2 (1.l.
with Heisenberg's discovery of the uncertainty relation. d .k dU 1+U\ f ln(l T + U\  U U k ( e .In .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > . 1. kT/2.e.2 Contradicting Discretization: Uncertainties 7 Example 1.= . i.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.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. we obtain T V S =fc[ln(TV+ P ..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". {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 the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente"). for 2 degrees of freedom.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.i n .1.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom. 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.ln(TV . so that 1 f ds\ T — \dUjy .1 ' u~ e . .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. around 1926.1)! .
8
CHAPTER 1. Introduction
shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = mx2 +  w V .
ZJ
(1.11)
Zi
The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A
A A
^ =\[Ef+l
(i i2)

Thus the amplitude can assume only these definite values. We now perform the following thought experiment. We give the oscillator initially an amplitude which is not contained in the set (1.12), i.e. for instance an amplitude A with An <A<An+l. Energy conservation then requires that the oscillator has to oscillate all the time with this (according to Eq. (1.12) nonpermissible) amplitude. In order to be able to perform this experiment, the difference AA = An+1  An must not be too small, i.e. the difference AA =
V mu>
n~V tfAA
n+
£HV
1 2h 2 [l + 0 ( l / n ) ] . mui 4^/n
For m = 2kg, h = i x 10 3 4 J s , u = I s  1 , we obtain lO" 17 [l + 0 ( l / n ) ] meter. 2^
+ H. Koppe [152].
1.2 Contradicting Discretization:
Uncertainties
9
This distance is even less than what one would consider as a certain "diameter" of the electron (~ 10~ 15 meter). Thus it is even experimentally impossible to fix the amplitude A of the oscillator with the required precision. Since A is the largest value of x, where x = 0, we have the problem that for a given definite value of mx, i.e. zero, the value of x = A can not be determined, i.e. given the energy of Eq. (1.10), it is not possible to give the oscillator at the same time at a definite position a definite momentum. The above expression (1.10) for the energy of the harmonic oscillator, which we have not established so far, has the further characteristic of possessing the "zeropoint energy" Hu>/2, the smallest energy the oscillator can assume, according to the formula. Let us now consider the oscillator as a pendulum with frequency u in the gravitational field of the Earth. * Then
"2 = f,
(113)
where I is the length of the pendulum. Thus we can vary the frequency cv by varying the length I. This can be achieved with the help of a pivot, attached to a movable frame as indicated in Fig. 1.4. The resultant of the tension in the string of the pendulum, R, always has a nonnegative vertical component. If the pivot is moved downward, work is done against this vertical component of R; in other words, the system receives additional energy. However, there is one case, in which for a very short interval of time, 8t, the pendulum is at angle 0 = 0. Reducing in this short interval of time the length of the pendulum (by an appropriately quick shift of the pivot) by a factor of 4, the frequency of the oscillator is doubled, without supplying it with additional energy. Thus the energy En= ( n +  ) fojj becomes I n +  IH2co,
without giving it additional energy. This is a selfevident contradiction. This means — if the quantum mechanical expression (1.10) is valid — we cannot simultaneously fix the energy (with energy conservation), as well as time t to an interval 8t —• 0.§ The source of our difficulties in the considerations of these two examples is that in both cases we try to incorporate the discrete energies (1.10) into the framework of classical mechanics without any changes in the latter. Thus the theory with discrete energies must be very different from classical mechanics with its continuously variable energies.
H. Koppe [152]. See also Example 1.3.
10
CHAPTER 1. Introduction
It is illuminating in this context to consider the linear oscillator in phase space (q,p) with
P2
1
29
E —
1—mco q = const. 2m 2 *
(1.14)
Fig. 1.4 The pendulum with variable length. This equation is that of an ellipse as a comparison with the Cartesian form a2
+
'" b2
reveals immediately. Evidently the ellipses in the (g,p)plane have semiaxes of lengths 2E b= V2mE. (1.15) a = mw' Inserting here (1.10), we obtain 2(n + l/2)fr^ (1.16) hn = ^2m{n + l/2)^. mar We see that for n — 0,1, 2 , . . . only certain ellipses are allowed. The area enclosed by such an ellipse is (note A earlier amplitude, now means area) An = nanbri or ,( pdq — 2irh I n + ' ]. (1.17b) 2irEn
UJ
2Tih{n+

(1.17a)
In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at
1.2 Contradicting Discretization:
Uncertainties
11
any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 Any'=
(1.17a) 2TT/L
since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)plane, we can write more precisely ApAq > 2irh. (1.18)
This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.
Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by
12 the limits 0 < q < q + Aq, 0<p<p
CHAPTER 1. Introduction
+ Ap,
what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is reinforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.
1.3 ParticleWave Dualism
13
the number of such photoelectrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particlelike, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few pointlike spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particlelike nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum
The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f
L. de Broglie [39].
14
CHAPTER
1.
Introduction
1.4
ParticleWave Dualism and Uncertainties
We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a doubleslit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.
^y
Ax
Fig. 1.7 Schematic arrangement of the singleslit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].
1.4 ParticleWave Dualism and Uncertainties
15
figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.
Si
t
Ax
\
»v
"^^J
B ^ ^
*
w
p3
Q
Fig. 1.8 The wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)
Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — rexp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T
5 = 2a =
lim NSN = — A a ; s i n 0 .
(1.21)
If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.
16
The intensity at the point Q is therefore
CHAPTER
1. Introduction
h = h
where from Eq. (1.21)
•K ,
.
a = flisinB = Aisint A 2
.
k
Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is
Ie=Io
sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2
(1.23)
The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 )  , i.e. for A x sin0 = (2n + 1 )  = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.
A
(1.24a)
(1.24b)
The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the singleslit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave Dualism and Uncertainties
17
Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or
1.4.1
Further thought experiments
Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light
Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it
18
CHAPTER 1. Introduction
is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~±. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have L = rp' = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.
Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.
1.5 The Complementarity
Primciple
19
quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by
AE « T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax
vG
.
It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of Ai£.
1.5
Bohr's Complementarity Principle
Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The socalled Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a doubleslit (this being the apparatus in this case) the photons or
20
CHAPTER 1. Introduction
electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wavelike or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*
1.6
Further Examples
Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with timedependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —yEo, w o (1.28)
where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui
u> = u>o a,t t = 0, w = ui at t = T,
.
_
i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w mx • 2\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt
dt The energy of the oscillator is l „ E — mx1 2
l 0 , z l 29 + mu} (t)x , y 2 '
so that
dE 1 — = mxz dt 2
9
dJ1 > 0, dt ~
(1.29) v '
where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain  [mx + mur (t)x\ = 0, i.e.
1 1 1 — —mx 2 H—mx 2 dt u22 2
1 mx 2
2d
— . 2 dt\u)
"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].
1.6 Farther
Examples
21
d ( E\ 1 , d / 1 \ — — ) = mx2 — ( — = dt\uJ2J 2 dt\u>2)
mx2 dw < 0, UJ3 dt ~
(1.30) v
where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, ie. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E
E dw2 < 0, UJ4 dt ~
i.e.
1 dE 1 dw2 < —2 . E dt ~ u) dt
(1.31)
JEo
or
E  Jui
u, '
> °
"o
E0 ~ UJ22
' ^ u'2 — < —
E'
<
^EQ.
Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with
e + wgeo, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2  =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < 5Bo.
E0<E'
Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and
22
CHAPTER
1.
Introduction
the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average
(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence
(L32)
E i.e. in
If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then lmw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (mx2\ =
If we now insert in the equation for dE/dt,
/I 1£ 2\ (  mx ) = , \ 2 / 2u2' we obtain dE_/l
~dt ~ \2mX
and hence
2\dw
2
_Edw2
dE _ 1 dw2 _ du
/ ~dT ~ 2w2~dT'
E — = const. w
°r
~E ~ ~iU> ~~ ~u7'
In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by
~~ p
2R\2nh)
~ AKR'
See e.g. L. Schiff [243], pp. 25  27.
Chapter 2
Hamiltonian Mechanics
2.1 Introductory Remarks
In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and reexpress Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the cnumber valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.
2.2
The Hamilton Formalism
In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
in analogy with Eqs. i. the entire timedependence of observables u(qi.pi) of qi. x^fdu. the equation of motion of the observable u. du \ x^/dudH du dH\ . whereas the velocity requires observations of space coordinates at different times.p% are therefore again observables. which can be observed directly at time t.24 CHAPTER 2.t).n n. All functions u(qi. (2. one obtains the Hamilton equations* • OH . dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. (2. A. chapter VII. Pi has to be considered as an independent quantity.) = £ [wm + WiK) = £ [WiWi . One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. Goldstein [114] remarks at the end of his chapter VIII.4) contains as special cases the Hamilton Eqs.e.Pi) is contained implicitly in the canonical variables q^ and pi.2) as This equation is.P.1).. chapter VIII.Pi.4) as the generalization of Eqs. Rather. The following properties can be verified: *See e. . (2.^ ^ j .g. Hamiltonian Mechanics is wellknown. (2. We can therefore consider Eq. One can verify readily that Eq. It suggests itself therefore to consider more closely the properties of the symbols (2. qi(t + 6t) . (2. since . as mass times velocity. Goldstein [114].qi(t) qi = hm f .1). M. (2. which all together describe the state of the system. S »(«. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. (2.1). The total time derivative of u can therefore be rewritten with the help of Eqs. this expression is simply a functional determinant. Dirac [75].3). ^As H. the standard reference for the application of Poisson brackets is the book of P. H. A system consisting of several mass points is therefore described by a number of such variables. . Compared with an arbitrary function f(qi. (2.1) as d .2) If we have only one degree of freedom (i = 1). 6t>0 5t Real quantities which are directly observable are called observables.
A}} + {C.B} = {B. B}} = 0. then the ordering is taken as in (2. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. {qi.Pk} = 5ik. Property (2.A}. (2.6) give the values of these.5d) is useful in calculations.Pi. for example. If. As an example we consider a case we shall encounter again and again. that of the linear harmonic oscillator. i.Qk} = 0. {pi. a i S i + a2B2} = ax{A. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. Bx} + a2{A. (2. (2. (2. Since Eqs.B*}.2.5c) The first three properties are readily seen to hold. {A. (2) linearity: {A.5d) above. 25 (2.e. it is irrelevant whether we write {A.6) We can now show. B} is completely evaluated.B}C + B{A.4).B}* (4) product formation: {A. The original definition of the Poisson bracket will not . we wish to evaluate {A. where A and B are arbitrary observables.5a) (2.5b) (3) complex conjugation (note: observables are real. These are {Qi.2 The Hamilton Formalism (1) Antisymmetry: {A.BC} (5) Jacobi identity: {A.B}. (2.5e) = {A. that the very general Eq. If we evaluate the Poisson brackets for qi.6).5d) = {A*. Later we shall consider noncommuting quantities. but could be multiplied by a complex number): {A. B2}. (2. which combines the Hamilton equations. C}} + {B. B}. in other words without any reference to the original definition (2. {B. As long as we are concerned with commuting quantities.Pk} = 0.B}C or C{A. {C.3) of the Poisson bracket. like here.C}. the Poisson bracket {A. we obtain the fundamental Poisson brackets.
and P={p. (2.p. In the evaluation one should also note that the fact that g. (2. q = {q. "4' = q.p.7) into (2.9) and (2...8b) p=q.p) = ±(p2 + q2). and Pi are ordinary real number variables and that H(q. Hamiltonian Mechanics be used at all.q2}) = = 2i{q. and so q = q. According to Eq.t).7) q = [q.6).8b) We insert (2. q + q = o.26 CHAPTER 2. From Eqs. or q(t) = qo+ Pot .9) Similarly we obtain from Eq. from which we infer that q(t) — qocost + posint.p. These are transformations qi—>Qi = Qi(q.P2} + {q. we consider as Hamiltonian the function H(q.p) is an ordinary function is also irrelevant..\(p2 + q2)} = l({q.10) In classical mechanics one studies also canonical transformations. (2. (2.12) .11b) (2. (2.10) we deduce q = p = q.8a) and use the properties of the Poisson bracket and Eqs. (2. Pi—> Pi = Pi(q.H}..8a) (2. Since constants are also irrelevant in this context. (2.4) we have for u = q.p\) P.11a) 'q' = q.t). Then we obtain: (2.H}.p}p + p{q.qot2 . (2.yPot3 + •••• (2.
_ax p__dK_ (2.1 we verify only Eq. Pi^Pi=Pi(Q. chapter VIII. Example 2.15a) °raS {AiB}Q. and Example 2.3 deals with the relativistic extension.e. as {A. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.14) With the help of the definition (2. those at a time t — 0. P) exists.e. i.e. {QhPk} = Sik. B} of two observables A and B in terms of either set of canonical variables.e.Qk} = 0.* Hence in Example 2.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P).15a). which means that a Hamilton function K(Q. Eq.B}Q>P. {qi. verify that the Poisson bracket of two observables A and B is invariant. for which Hamilton's equations hold. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.2. which transforms qi.pk} = 5ik.t). i. {Qi.3) we can now express the Poisson bracket {^4.P One can then show that {A. {PhPk} = 0.15b) The proof of the latter invariance is too long to be reproduced in detail here but can be found in the book of Goldstein.p <> Q.Pi back to their constant initial values. Goldstein [114].P is canonical in the sense defined above.t).P provided the transformation q. (2.12) Qi—>qi = qi{Q. {qi.p = {A. (2.15a). i. Of course. (2. Example 2.2 below contains a further illustration of the use of Poisson brackets. . .qk} = 0. (2. i. *H.e.13) We write the reversal of the transformation (2.B}q. i.2 The Hamilton Formalism 27 for which the new coordinates are also canonical.P. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.Pk} = 0. that (dropping the subscripts therefore) {Pi. (2.P.B}q.
A}q. we obtain dA {Qk.q/c) and (E. we obtain analogously dA dQk' Inserting both of these results into the first equation. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u. 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 „ „•.p) . OQk Pk}q.P^pr.p = J2 d~Fk Replacing in the above A by P and B by A.15a) r„ ™ v .Pj}q.s F 5 7 r ={A.( 9A = {Pk.pc). we obtain as claimed by Eq.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. Qk}q.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential). .15b) dA {Qk.A}q. Replacing here A by Q and B by A.p = dB dA dB \ rA 1 {A'B^ E b r s r . 236. Mittelstaedt [197].p(0) = poSolution: The solution can be looked up in the literature. we have —Et.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. {A.B}q.p^ "Pk. dA {Qk.28 CHAPTER 2. § See P. (2. Solution: Relativistically we have to treat space and time on an equal footing. Thus we extend q and p to spacetime vectors (t.3: Relativistic Poisson brackets By extending qi.Qj}q. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. V V °Qk dPk dPk dQk J E x a m p l e 2.B}QIP.p^v dP.E(t).p^— + {A. p. § E x a m p l e 2.+ dQ.F}r ~di d~E ~ d~E~8t Consider F = H(q. Thus whenever q and p are multiplied. (2. their product Et — qp being relativistically invariant.
H(q. i. . Instead we assume a case in which we know only that the system is located in some particular domain of phase space.p).p) around some point qo. to . dE . 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. g oc AqAp. I .1 Liouville Equation. and E as a function of t).y.t).dt y/l  qZ/c2' 2. We distinguish in the following between two kinds of probabilities. numerically zero. so that Go(qo. of course.to. if qo.E(t)}r = Hence at Relativistically we really should have clu/dr. since H is expressed as a function of q and p. Probabilities 29 (This is. We consider first the a priori weighting or a priori probability.po. (1. rrr .p) . 2TTE UJ and hence g oc — .17a).e.. one obtains Gi with q = q((lo.t0. g. in the case of the linear oscillator with energy E given by Eq. also boundary points of one domain are mapped into boundary domains of the other. Then . Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.Po. .po is a point on Go.3.3 2.2. du q\ P = —. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du .po) = Gi(q.' ±=j1V c \dtj' u^. Of course. but partial derivatives of F do not vanish. p= p(qo. This probability is evidently proportional to AqAp. c dt 2 2 w^ du du dr . i.16) For example.p).. we have A = (p apaq — J .Po.p)._.. But now we consider a system whose phase space coordinates are not known precisely..3 Liouville Equation.t). Let us assume that at some initial time to the system may be found in a domain Go(q. and at time t > to in a domain G\(q.Pi are the coordinates of some mass point m. dudH s l u .e. du . (2.14) and area A of the phase space ellipse given by Eq. and the space spanned by the entire set of canonical coordinates is described as its phase space. (1.. it is the equations of motion which lead from Go(q.)•'=(*)**£.p) to G\{q. (*. Since Hamilton's equations give a continuous map of one domain onto another.
Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. J0=0 Hence g oc 8n2IdE.4. the (pg.30 CHAPTER 2.p^. dt dq dq dp dp d2H dqdp d2H = 0. 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. and hence g oc %ir2IdE..5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . dt dq and with with Hamilton's equations (2. Hence from the difference: — . which means.14) t o (1. this has the same value at a time to. — dp A A — In(AqAp)s = — + — = Example 2. in view of this independence it can be expressed in terms of the conserved energy E. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . (1./. = 2TrIEsm6. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin.)curve for constant E and $ is — as may be seen by comparison with Eqs. as is demonstrated by Liouville's theorem in Example 2.eded<t> = 8TT2IE. Example 2. Example 2.15) — an ellipse of area § dpgdp. Je=0 .— = — Aq. Pi 2 / \. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates.1): d . dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E.4: Liouville's theorem Show that A q A p is independent of time i. Show that the total volume of phase space covered for constant E is 8n2IE. dpdq and similarly dt = — Ap. Thus g must be independent of t.
t). leaving the basis of classical mechanics! Hence we set / «w)^ =. (2. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space.*). Since W has the dimension of a reciprocal action. by jr = P(9»P. however.2.t)dqdp = N. The total number of systems N is obtained by integrating over the whole of phase space.p. Thus dn is that number of systems which at time t are contained in the domain q.17) F ^GT^^ 0 Fig.e. (2.1 The system moving from domain Go to domain G\.p. 2.20) .3. i.t)dqdp= 1.Thus we assume a large number of identical sytems. (2.q + dq. i.19) Thus W is the probability to find the system at time t at q. W=p(q.e. whose positions in phase space are characterized by points. Probabilities 2.3 Liouville Equation. (2. dn — p(q.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.p.p.p. it is suggestive to introduce a factor 2KK with every pair dpdq without.p + dp. d d 1 P = JJdqidpi.Po.18) With a suitable normalization we can write this / W(q.
In doing this. i.32 CHAPTER 2. The equation of motion for n or W is the socalled Liouville equation. p(q. 2.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. i.P dt p(q + dq.p.p. * q Fig. t)dqdp p(q.p.2 The ensemble in phase space.p / . we consider the domain G in Fig.p.3 The region G.p) denote the velocities in directions q and p — p{q. — if vq(q.p) and vp(q.e.p. 2. how the system moves about in phase space. p + dp G P O q qndq Fig. 2.a t dt + q+dq.e. we take into account.t)dp{ jt Q. t + dt)dqdp — p(q.e. that in our consideration no additional points are created or destroyed. In order to derive this equation. i. We are now interested in how n ov W changes in time. The number of points at time t + dt in domain G.t)dp( — I . t + dt)dqdp. 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.
t)vq(q + dq.21 ) This is the Liouville equation which describes the motion of the ensemble or.20) and (2. vp(q.p + dp. dH Mq.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . t)dqdp = p(q. the probable motion of the system under consideration.t) dt {H(q. (2.p(q + dq.p) = q = g^. put differently. With Eqs.p)dtdq . . .p) . Comparison of Eq.p(q + dq.p)=p = 3H —.p.p(q. p.21) we can also write dW(q.p + dp. .W(q.21) with Eq.t)vp(q. = K(9. .p)dtdp +p(q.p). 33 Dividing both sides by dqdpdt this becomes p(q.p.p.p.t + dt)dqdp — p(q. Probabilities and thus p(q.p. (2.p + dp)dtdq.P)].t) dt p(q. t)vq(q + dq. (2.t)vp(q.p. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .t)^ = 1.p.p(q.p.p) .p.19).3 Liouville Equation.t)vp(q.p) dq p(q.2.t)vq(q.t)} with JW(q.p)dtdp . p.p(q.P.p + dp) dp = or  However.t + dt) . (2.4) shows that p and u satisfy very similar equations.p.P *' % = iH>P} ( 2 .t)vp(q.t)vq{q. (2.P)]K(Q.
po are the initial values of q. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. With Eq. and hence that equal phase space volumes contain the same number of systems. We shall see that we have two possibilities for this.24) implies that p is a constant in time. (. t) depends explicitly on time t (if determined at a fixed .e. since no systems are created or destroyed. We now inquire about the time variation of the expectation value (it) of u. i.p) be an observable. that dtj\y v if qo. p. 2. i.24) since the total derivative is made up of precisely the partial derivatives contained in Eq.4 Expectation Values of Observables Let u = u(q.34 CHAPTER 2.26) we described the time variation of the observable u(q.p). for i<«>!/«<*p>"w>(^)". (2.4).t)(^J. The first and most immediate possibility is — as indicated . We deduce from the Liouville equation the important consequence that ^ M = 0.p) the following expression: (u)=Ju(q.24). (2. (2.1 the area Go is equal to the area G\. Equation (2.e. Thus in Fig. Example 2. (2.p. p+dp.4). 2.p (cf.that the density or probability W(q.p)W(q. We define as expectation value of u(q. q+dq.p. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.
po.t)}(^y. and hence u(q.32) In this expression the time t is contained explicitly in the observable u(q. we can also employ a more complicated consideration.t).Po.0) =u(g(qo.po. Then Eq.30) p = f(qo.t). .Po.W(q. we have Qo = g(q. where we used Eq. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.31) i.p).29). (2.p. Observables 35 point in phase space). (2.p.p) = uo{qo.p.Po.po.f(q0.p. (2.8. Reversing Eq. (2.t).Po. Goldstein [114].t) = = W(g(q0. We expect. we can express these in terms of their initial values qo.p0.24): W(q. ie. (2. However.0) = W0(q0. (2. Thus we can write.p. so that Q = g(qo.t).34) S e e also H.t).0) = u0{qo. 1 (2.t).po.t) W(q0.29) The distribution of the canonical variables is given by W(q. since W oc p is constant in time according to Eq. (2.p) = u(q.Po.22). Sec.p) assumes certain values) that depends explicitly on time.4 Expectation Values.po.p0)(^^J.t)W0(q0.t).28) = Ju(q.2.p.t).f(qo. We verify this claim as follows.i).33) po = f(q. (2.Po) at time t = 0. that (u) = (u)0. (2.p.e.^ Solving the equations of motion for q.27) becomes !<»>  /«(*P>!"W>(^)" (2. at t = 0.p){H(q. of course.t). and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q. 8.t). W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.
Po.P0.\„.t). Eq.t) = (2 30) (2.37) and (2.p. p = f(<j(q.p.Po.t). Taking now the total time derivative of (it)o.t) .39) Substituting this into Eq. (2.t).M 27Th i .0) u(q.po. (2. we obtain an expression which is different from that in Eq.p.P.32). (2. Hamiltonian Mechanics so that on the other hand with Eq.p).36 CHAPTER 2. Inserting Eqs.t).f({H(q.t).P)}). Eq.t) dB__ (du\ °P \°PS P=f(qo.t)J(q.t)W "(q0.po.PO.30): duo(qQ.t).P0.36) and (2.f(g(q.po)(~ .p)W{q.30) Moreover.) Wo(qo.28).Po. (2.0)=u(q. (2.t).p. [ du0{q0. (2.p.p)})q=g(QO.(2.t) M f 9u\ V °P / P=f(q0. / ^fdq0dpc .u(q.p0) = W(q.30) u0(g(q.p.p.35) Inserting these expressions into no we obtain uo(qo. we obtain («)o = / u{q.f(q. (2.t)\—^\ = (u)." V " " ™0' > • • » « » 0 .p. (2.29) q = g(g(q.p. (2.t).Po.t) v{g(<j(q.e.f(q.t).t). (cf.=9(.p. (2.31)) W0(q0.t) d t _ (du\ \ d<i dq / P=f(qo.f(q.25) into Eq.t).36) = (2.) f / u (q0.o. .p). (240) .u{q. i.38) we obtain £<«)„ = .po.p. d dt {U)0 = dt.p).t).POit)i p=f(qo. (2.P.32).t).35).t) 0H_ 0 q =  {{H(q.t).t) (2.Po)(^^)n.t) dp M du\ ° q 'P=f(qo.37) as had to be shown.(dq0dp0\n We deal with the partial derivative with the help of Eq.f(q.
t) dqdp\n 2irhJ (2. ^ „.W(q.p. Observables 37 Here we perform the transformation (2.45) the Liouville equation.34) and use (2. described as "Schrodinger picture".t) of u assuming certain values q and p.t) dt {H(q.2. / d q d p \ n 2irh J (2.p).p){H(q.25) and (2. The considerations we just performed demonstrate that we have two ways of treating the timedependence: The explicit timedependence can either be contained in the probability W or in the (transformed) observables. and the time variation of (u(q.p)}W(q.. u}W instead of u{H. uW} = {H. x r „.42) the relation d .28).p).uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2. This timedependence is described by the Liouville equation dW(q. In the first case.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity).p). However.p.u(q. we obtain zero after partial integration of I and hence from Eqs.41) This expression contains {H. .p)) is attributed to the probability W(q. so that ~d~t <«>0 = J{H(q. (2.28) and (2.t)} .31).W(q. W}. from the properties of the Poisson bracket.45) in agreement with Eq. dt <«)o v f u(q.p). we obtain {H. (2.. (2. The phasespace integral of a Poisson bracket like I (2.43) ——uW = 0..p.41) and (2. Alternatively we could deduce from Eqs.W} in Eq. .p. the observable u is treated as a function u(q.(2.t)}.p.28).4 Expectation Values. Consider {H. lim ——uW — 0. (2.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. u}W + u{H.
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.e. the probability of the initial values Wo(qo..p). and probabilities W. does not have to be commutative. i.t) dt __ ~ (2J30) ( 4) du0(qo.p0.po) is assumed.H(q. Hamiltonian Mechanics In the other case.po are constant initial values — we have du0(q0.38 CHAPTER 2. " T h e author learned this approach from lectures of H. The equation of motion is then that of an observable.p)}.4) of an observable on the other." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.po. but does satisfy the usual associative and distributive laws. Eq. . and the equation of motion (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.po.F(«. The time dependence of the expectation values can be dealt with in two different ways. (2. (c) A multiplication of the quantities among themselves is defined. Koppe [152] at the university of Munich around 1964.46) We thus also recognize the connection between the Liouville equation.p)}. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations.t) dt du(q. (cf. (b) As usual a muliplication by a complex number is defined.p) It = {u(q. 2.rt. This formulation deals with observables u representing physical quantities.4)) ^ = {«(. as we observed. called u Heisenberg picture". (2.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. which describe the state of a system. which. as the equation of motion of an ensemble or of a probability distribution on the one hand.t). the reason being that — since qo. " This means: Only in the Heisenberg picture dW/dt = 0. for which the axioms of a commutative group apply.
26)) (u) = Tr(wW). ox In view of our later considerations. that its use here implies the essential properties it has in matrix theory. (c) TV (u*u) > 0. and to be able to correlate these with the above classical considerations.p} = l {q.p)(^y. Thus we can write the expectation value of an observable u (cf. (2. for a phasespace integral we define the word or symbol "trace". (b) TV(cm + j3u) = oTV u + /3Tr v.48): (a) TV u* = (Tr «)* = TV^.q}=0.48) In matrix theory the symbol "trace" has a welldefined meaning. Eq.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2.q} = 0 {p. Introducing it here assumes.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. That this is the case. (d) Tr(ut. if u ^ 0. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. (2. (2. px > in—. ~[q. it is helpful to introduce already at this stage some additional terminology.p} = 0 • — ^ ~[q. also called "density matrix". therefore.) = TV (vu).2.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible.p] = l. can be deduced from the following characteristic properties of a trace. which all apply to (2. B] := AB — BA in the following way: 'i {q. Moreover.6) a socalled "commutator''' [A. also written "TV". y:\p. by Traceu:=Ju(q. In Chapter 9 we attempt a corresponding approach for classical systems with . so that we consider this next.
These aspects will be considered in Chapter 27. Hamiltonian Mechanics a wavelike nature. since gauge fixing (i.e. . obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. the Poisson brackets require modification to Dirac brackets.e. However. it will be shown that in electrodynamics. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. i. electrodynamics. a constraint) has to be taken into account. and.40 CHAPTER 2. excepting the Poisson brackets.
quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. we can define the Hilbert space as the space of states of a physical system. 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. which turn out to be those of the theory today known as quantum mechanics. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. of measurable quantities. We found that the Poisson algebra permits extensions to noncnumber formulations.e.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. i. and selfadjoint operators in this space as representatives of observables. if the elements ipi of M satisfy the usual axioms of addition and K 41 . Building upon this. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.e. A set M. 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.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. 3. i.
5a) (3. (3. n exist (not all zero). n is called the dimension of . A + tpj = tpj + i'i. —»• IK.M is said to be a metric vector space or a preHilbert space. i. 2 .ip2). ^ i = 0 if ^ = 0. . so that n 5 > ^ = 0.~ip2) '• M./?eK).  . n. E (3. • • • . fa. if every vector ip E M can be associated with numbers Q . . (ipi + ipj) + ijjk = ipi + (ipj + ipk).e.ipn are said to be linearly dependent. . are linearly dependent.3) If all on = 0.tp2. (3 5c) + a2{ip. (3. x M.^i) . .2) Vectors ip\.ip2. .^1) = (01. (ipi. i=l (3.aiV>i + a 2 ^ 2 ) = ai(ip.i = 1. f > 0 if V T ^ O .42 CHAPTER 3.5b) . Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. i = 1. and n is the smallest such number.ipn are said to be linearly independent. .4) The vector space . (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M).02)* (hermiticity). . In each case n linearly independent vectors are said to form a basis. If n + 1 elements ^ € M.1) aipi + aipj. ip2 of this space can be associated with a complex number (V'l.. (V>. (aptyi. . if numbers eti. . . (a.^) called inner product. if any two elements ip\. ( n (3. such that n ^ = YJCi^i. 2 .M. with the properties (a* G IK): (^2. the vectors il)i.
which we can write 0 < ( ^ i .9b) = 0 (Pythagoras theorem). The distance between two vectors if>i. WA+Ml2 = ll^il 2 + ll^2 2 . linearity in the second component.9a) (3. antilinearity in the first component (also described as sesquilinearity. (3. ^ i ) + A ( ^ i .rh) (3.e.^) = ai*(V'i.2 Hilbert Spaces 43 where the asterix * means complex conjugation. ^ i ) + A * ( ^ .9a) we start from if) = ipi + \if>2 6 M. In order to verify Eq. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).V>i + A</>2) > 0 . The first two properties imply ( a i ^ i + a2ip2.2.^ 2 ) =   ^   2 + 2 ^ i . The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2. A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. (310) (^2. ^ 2 )  .V') +"2*(V.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).9c) HV'i + tp2\\2 + \\1p1 . (3.^2)1 < H^ill • 11^211 (Schwarz inequality).9d)   V i   = sup  ( ^ i .is defined by d(ipi. for ip\. tp2 £ M: 1(^1.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations. if Wi. meaning oneandahalffold linearity). ^ ) + A 2 (V2. for arbitrary A and ip2 ^ 0: (V>i + AV>2.^1) 2 h 11 0 <%«e + M .3.tp2 £ M.*h)=\\AH\(38) (3.V.. (3.)> (36) i. (3.7) In addition the following relations hold in a metric space M.. »2 2 .
9a).  ^ l l = (HlMI +   « 2 .^)1 <IIV>il. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X .13) the vector is said to be normalized.9d).11). (3. . HIHI (3. In verifying the triangle inequality we use this result (3. ^ e M are said to be orthogonal if (</>!.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. so that   ^ l + ^ 2   <   ^ l   + ^2.^1)1 Vi 2 + ^2 2 + 2   ^   . for which CO 2 := V^l^l 2 < 00.V2) IIV'l2 + V'22 + 21(^2.* Two vectors i f i . (3. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. (3. (3.44 or CHAPTER 3.9e). beginning with A = 1 in the second line of Eq. Mathematical Foundations of Quantum Mechanics 1(^2. V 2 ) = 0 . a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. Examples of metric vector spaces: (1) Let M. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. o ^ V i = o. ll^ill + llVdl. (3.\\ih\\. (3.9c). hence verification in each case is necessary.12) We omit here the verification of the remaining relations (3.ii) thus verifying Eq.
[$]):= / JM /(x)* 5 (x)d 3 *. But this applies also in the case of any function which is nonzero only on a set of measure zero. with inner product oo 45 (v. which satisfies relations (3. etc. = 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. 2 and so on.e. (3. */ x } = / /o for x = 0. (3. Js the space C? is not yet a metric vector space although for (cf.g)= [ /(x)^(x)d3x.3. and one defines addition and multiplication by complex numbers with respect to these classes. are combined to an equivalence class [/] (with space L2).5c)) / ( x ) = 0 = > ( / . all squareintegrable functions / which are "almost everywhere equal". Then L2 is the space of all these equivalence classes. In order to avoid this difficulty. (2) Let M. . is defined by ([/]. Eq. i. i.5a).5b). •'^ \ 0 otherwise. Elements of the classes are then called representatives of these classes.w) :=J2v*Wi.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi).5c).e. / ) = 0. (3. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. With the scalar product (f. which differ solely on a set of measure zero. for which the scalar product.
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.
V \u) (in the case of £) with infinite norm!).and BraFormalism Furthermore we have: ( x  = 0. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*.e.3 Linear Operators. Two ketvectors \u). \v) E V are said to be orthogonal. (3) (u\u) = 0. Dirac's Ket. that the vector \u) E V is associated with a vector in V.4b) (4. 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.9a): \{u\v)\2 < (u\u)(v\v). which we write (u\. if \u) = 0 (see (3. (4. This expression is linear with respect to \u) and antilinear with respect to \v). i. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. if x\u) = 0 V kets \u) E V. the norm squared.5) 4.\u) = (u\0*.4a) \v) = J\(m<% and (4. The dimension of V is the same as the dimension of V.5c)). or (4. (2) (u\u) > 0.4c) (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).(t. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. It follows that we can now write the Schwarzian inequality (3. if (u\v) — 0.62 CHAPTER 4. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV .
4. i. the space Vi <S> V2 has the dimension .\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). if \v) — A\u) i.e. Furthermore linearity applies. The product vectors l?^ 1 )^ 2 )) span a new vector space. Furthermore.. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. The inverse of A is written A . also acts on the dual space V. and one writes: (V\ = (X\A. Finally we define the tensor. ltt) = \u). we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. B. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. Let the following bravector be given: (x\ £ V. let (771 6 V be the bravector defined by this functional.e. denned as a linear operator on V. This product is commutative. In this way the operations of operators A. BA = 1. It follows that (r. If Vi has the dimension A/i.. on bravectors are similar to those on ketvectors. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). i.3 Linear Operators. Then one says: (771 results from the action of A on (x.e. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). 1 and \u) = B\v).or Kronecker.. The operator A. Two operators A and B are said to be inverse to each other. as we can see as follows. The product of such vectors is written: \u^u^) = \u)^\u)^. Multiplication by unity. Then the following rules apply: A + B = B + A. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . The inverse of a product is (AB)'1 =B1A~1. if AB = 1.or direct product of vector spaces. This does not always exist. AB^BA. defines the unit or identity operator. the space Vi <S> V2.
in the product space.e.7b) (4.3). by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. \t) G V.e. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. i. AW\U)W = \V)W. As a consequence we arrive at the following properties (e. also Sec.e.7a) (4. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). the operator A^ is called hermitian conjugate or adjoint operator of A (cf. Let A^> be an operator in the space Vi and A^ space V2.64 CHAPTER 4.7C) («)(u)t = \v){u\. Every operator A^> commutes with every operator A(2\ i. and (v\ = (u\A. (v\t) = (u\A\t). we obtain the conjugate relation <tAt«> = (u\A\t)* (4. [A^. here designated with the same symbol. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. then \v) is a linear function of \u).g.A^] = 0. AW\uW)\uW) = \vWUW). Since we demanded the scalar products to have the property (t\v) = (v\ty. Every such operator is then also associated with an operator. Next we construct the following scalar products: (t\v) = (t\Ai\u). 3. i.and BraFormalism an operator in the Mi A/2.7d) .e. If (v\ = (u\A. i.6) for all \u). we write \v) = A*\u). Dirac's Ket.
if H is hermitian and (u\H\u) > 0 for all ketvectors \u).e.K\. K] = 0. but only if these commutators commute. for we had (\u){v\)i = \v){u\. An hermitian operator A has the properties (a) that all its eigenvalues are real. so that (a)(a)t = a)(a. Verification: Since A — A^ and A\u) = a\u). .K}. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A].K]^ = = [HKKH]* KHHK = = K^H^ [H. i.3 Linear Operators. It follows that a is real. An important theorem is the following. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . 65 The linear operator H is called hermitian. Obviously the quantity \a)(a\ is an hermitian operator. Analogous considerations apply in the case of bravectors (u. hermitian The product of two hermitian operators is not necessarily hermitian. so that (n^4n) is real.e. it follows that (u\A\u) = a(u\u). [H.4. as is also (u\u). i. or (u\A\u)* — {u\A^\u) = (u\A\u). and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. The commutator of two hermitian operators is antihermitian: [H. if H — H' and antihermitian. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. The operator H is said to be positive definite.
Ps\ua. (4. which includes vectors of infinite norm. (u\Ps = {us\.and BraFormalism Moreover. since a is real. (u\u') = 5{v . it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. If these form a complete system. a ^ b. so that (u\Ps = (us\.Au) = a(v\u) — b(v\u) — (a — b)(v\u). we have 0 = (uArt) — (u. i. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). (v\A = b(v\. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us).e.e.66 CHAPTER 4. then the hermitian operator defined on this space is an observable. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. ( n  z / ) = 0 . as we can see as follows.e. so that (v\u) = 0.) = 0. {us\us*) = 0. i. and let S* be its complement.u').e. i. Dirac's Ket. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. (n\n')=5nnl.9a) . Next we introduce the very useful concept of projection operators. The above theorem remains unchanged. Let S be a subspace of the Hilbert space !K. i. A continuous spectrum can immediately be included by passing to the extended Hilbert space.9b) (4. since a ^ b. but the continuum vectors are normalized to a delta function. Ps is idempotent. Assuming A\u)=a\u). Ps2 = Ps(4. \u8*) E S*. The entire set of ketvectors spans the extended Hilbert space.8) The projection operator has the following important properties: Ps is hermitian.
Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). p 2 .4. i.3 Linear Operators. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. 67 (4.e..9c) This property can be seen as follows. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm.e. Let a continuum state be written £).p = 0.. Then 0 = (P2P)\p) and hence p = 0. where £ is a continuous index. . Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. (1 — Ps) is projector onto S*. i. The quantity \a)(a\ is called elementary projector. Ps is projector onto S.9d) Thus the vectors P\u). P\p) =p\p). as can be seen as follows. 2). so that \ua) = (a\u)\a) = \a)(a\u). The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). \ua) = c\a). c = (a\u). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. (4. i. i. (1 — P)\u) are orthogonal.e. = (p2p)\p). iV) is a set of orthonormalized states.1. if the set of vectors 1). so that (Ps2 ~ Ps)\u) = 0.. \u) arbitrary. . Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. Set \u) = \ua) + \ua*) with (a\ua*) = 0. Let p be an eigenvalue of the projection operator P.e. Very similarly we can construct operators which project onto the subspace spanned by the continuum states.
(11. P = £n)(n. 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. one says the degree of degeneracy is r .r\ "An example is provided by the case of the hydrogen atom. If there are r such vectors. see Eq. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. the operator P is correspondingly P2\u)= f2\0m\u). Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). In general it is possible.114c)..6. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. that one and the same eigenvalue Aj is associated with several eigenfunctions.6. However. Degeneracy will be discussed at various points in this text. Dime's Ket. are linearly independent). .. Coulomb potential. 4.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. which are orthogonal to each other (i.e. 8. (11.1 and 11. Let \u\).r)(ui.e. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.68 CHAPTER 4. Examples 8. Sec.. be a system of basis vectors in this space. which we introduced earlier.g.114b). i. See e.4 Observables Operators which play a particular role in quantum mechanics are those called observables. u2). and Eq. in this case the spectrum also has a continuous part. Observables are representatives of measurable quantities.
is always implied).10) expresses the completeness of the orthonormal system.e.10) follows from the fact that i Applying the operator A to the projector (4. PA = y£pi i = Yt\ui.11) i. the projection of this operator is the entire space.r. Together with the orthogonality condition.10) This expression is known as completeness relation or closure relation. Let us set {A . and APi = XiPi.r) (the convergence.r\u). The operators Pi are linearly independent.r)(ui. The uniqueness of the expression (4.r (4. 69 i If A is an observable and if the spectrum is purely discrete. then PiPi' = 0. Applied to an arbitrary vector \u) G "K Eq.r\u)\2. or also as subdivision of unity or of the unit operator. i.r \(ui.4.r \ui.r)(ui. ^ A i P i = A = ^Aiui. which we do not enter into here.r\ = l.4 Observables The dimension of this subspace is that of the degree of degeneracy.r>(ui. i. the operator A is completely determined by specification of the eigenvalues A. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r\u) i. (4.10).r =^ i.12) .e. the relation (4. in the case of degeneracy with (ui. we obtain i i i i i.e. (4.r'} = Sii>6rr>.e.10) gives the linear combination it) = ^ i. and the eigenvectors \ui.\i)Pi = 0. i. i i (4.r\ui/. If Aj ^ Aj'.r)(ui.
In a similar way we can handle functions f(A) of an observable A.e. as explained earlier.14) .v'). Let v be the continuous parameter which characterizes the continuum. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). i.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). i.17) previously. for instance exponential functions. we have the corresponding generalizations. written in Dirac's notation. One defines as action of f(A) on. now however.70 CHAPTER 4. Dime's Ket. the eigenvector \ui): f(A)\ui) = f{\i)\ui). (4. (4. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). 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. (3. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. The ketvectors \uv) are orthonormalized to a delta function. for instance. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. (u„zv) = 8{v . If the spectrum of an observable A consists of discrete as well as continuous parts.and BraFormalism This is the expression called Parseval equation which we encountered with Eq.e.
(4. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. (3. where the symbol ip is already indicative of the Schrodinger wave function ij. In the onedimensional case we write {x\x') = S(xx'). / dx\x){x\ = 1.. Therefore we want to reexpress the Fourier transform (3...e. again later.17) The Fourier transform provides the transition from one representation and basis to the other. Since both expressions (4. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies).4. and (b) if they possess a uniquely determined system of basis vectors. Extending this we can say: A sequence of observables A. as a formal orthonormality condition. For many practical purposes the use of the Fourier transform is unavoidable. their commutator vanishes. 4. which all commute with one another. (4.59).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. (4.B.C. if (a) they all commute pairwise.e.15) and (4. First of all we can rewrite the integral representation of the delta function. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. with energy E. i.16) represent subdi . We shall return to this theorem. x) € Fx.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. for which the completeness relation is f dk\k)(k\=l.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^.58) in terms of ket.and bravectors. B]. if and only if they possess at least one common system of basis vectors. here presented without proof. Eq. In the following we consider more generally ketvectors \ip) as representatives of the physical states. i.. \k)€Fk. form a complete set of commuting observables. [A.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.Fk. il>{x) := (x\i/}) : "K * C.15): 5(x .2a) etc.72 CHAPTER 4. called wave function. Obviously we obtain this by inserting a complete system of basis vectors of Fk. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4.e.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. Dime's Ket. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. <a#) = [ dk{x\k)(k\ip). (4. ${k) := (k\if>). The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K.L e " * * ' = {x'\k)\ V27T (4. V 27T (k\x') = . (x\k) = ^=eikx.21) provides the ketvector \ij)) in the A. (3.space representation. which are representatives of the states of our physical system. (4. i.1a). . Rather \x) and \k) serve as basis vectors in the representation spaces Fx.18) According to Eq.e. this expression has to be identified with — / dkeikxeikx\ i.59) or Eq. 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).x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').and BraFormalism visions of the unit operator.19) Comparison with the orthonormalized system of trigonometric functions (4. (4. (4.e. we can rewrite Eq. i.
which is the quantum mechanical analogue of the classical probability density p(q. the calculation of which for specific cases is the subject of Chapter 7.2 The Density Matrix We shall establish a matrix p.p. if we do not take into account the complementary part of the universe.t). This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.1 Introductory Remarks In this chapter we remind ourselves of the measuring process briefly alluded to in Chapter 1 and of the necessity to recognize the abstraction of a physical situation that we perform. 73 .e. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). i. 5.5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. 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. 1. Schrodinger [244]. We mentioned earlier (in Sec.Chapter 5 Schrodinger Equation and Liouville Equation 5. with the complementary system or *E.
. where T means transpose: p = CTC*.. \a). The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. where i represents collectively all quantum numbers of the state of the system under consideration. \a). The states \i) £ "K. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. that is.. T . p" ( C C * ) t = CTC* = p. \i).. i. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij..i. The actual and real state of a system is then a superposition of two types of states.i.^ Since p is hermitian. we have to deal with a socalled "mixed state". The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. For an exact treatment we would have to express the actual state \ip) of the system as a supersposition not only of the states \i). called the density matrix.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. which in Dirac's notation we can write as = J]C a »i) a. of an operator that acts only on the states of the system we are interested in. i. a (5.e. but also of the states of the complementary set of the universe.b. (a\i)=0 = {j\b). (52) (5. are called "pure states".3a) and N J  a ) ( o  = l in the complementary subspace..74 CHAPTER 5.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. Schrodinger Equation and Liouville Equation the measuring instruments.. These are the quantum mechanical states which are also referred to as microstates. we have (a\b) = 5ab.e.3b) The quantum mechanical expectation value of an observable A.b.
2 The Density Matrix 75 \i') 6 "K.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. The ensemble can be represented as a set of points in phase space. in the position state representation or xrepresentation. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).ji(x / )i*(x).4) we had the expression (A) = Y/Pij(j\A\i) i. In the following we write simply \i) instead of \i'). (5. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities.j(x'i)(ia.5.e. but occupy in general different microscopic. (5.6) (i'\p\i") = Wi'Si'vi = real. as we saw earlier. i. seen macroscopically of Z copies of the system).3a). (see below) eM.) = YJc t .e. This ensemble is interpreted as a large set of identical systems. or (5. a projection operator.\p\i")(i"\ = Y.e.9) is diagonal. This transition to a new set of basis vectors in which the matrix becomes diagonal. i. which are all subjected to the same macroscopic boundary and subsidiary conditions. i i (511) In Eq. where x represents collectively the entire set of position coordinates of the (particle) system. P ^ ) = i)(#) = (#)i) (5. as we discussed previously. e. (5. quantum states.e. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i.. In this sense we have p(x'.2) and (5. which today is a fundamental term in statistical mechanics. . The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). If the ensemble corresponding to the system under consideration consists of Z elements (i. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. Gibbs introduced the concept of ensemble. Thus the real system is in the pure state \i) with probability uii/ J2i Ui. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann. a complete orthonormal system with properties (5.10) and is therefore.3 with Pi^YtCaiCaj*.* We can specify the state of our system.g..7) p = ^2\ii)(i. in a specific representation.x) := (x'\p\x) = yju..
Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf.18).16) (5. (5.17) Hence the sum of the classical probabilities is 1. Then the expectation value of the observable A is (cf. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). 3 (5.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.14) which we can write (A) = Tr(pA).76 CHAPTER 5.4)) (5. (5.) := (mm = (#') W> = l(#')2 > 0. as claimed by Eq. (5.4).3 = pvv = uv.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. (5. (5. i t (5. (A. (5. On the other hand. (5. Eq. We can also show that Ui > 0.5) above) (i\p\j)=Pij. according to Eq. In particular for A = 1. we have Tr(p) = l. Thus Tr(j>ii)<i)=l. Hence Wj/ > 0. Eq.18) Substituting this into Eq. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .
represents a quantum mechanical probability. i. The effect of the interaction of the system under consideration with the surroundings.t). that — different from u>i — P. not in a mixture of states.20) expresses that the system can be found with probability 1 in the state \i).t) = \^{x. one says. (5. Thus we have a system in a pure state. (5. We observed above. In the following we are mostly concerned with considerations of systems in pure states.5. We see this more clearly by going to the position space representation. the system is in a pure state. but we see here. t) p= i The expression (5. (5.e.t) = {x\i).3 The Probability Density p(x. so that p{x. in all other cases the state of the system is a mixed state. If all ui but one are nought.24) s (x\x') = 5{x . ^2ui\i){i\. the only remaining one is 1. We now consider the latter expression. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. defines the statistical operator.t)\2.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l.t) and instead of (x\Pi\x) we now write p(x. i.t) we now write tp(x. . thus enters these states.22) In the following (x\i) is always to be understood as (x\i(t)}. by representing the vector \i) by the wave function i(x.x').21) p=\i){i\=Pi. i. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. Equation (5. (Question: Is the state of the universe a pure state?) 5. since they do not take into account the (interaction with the) rest of the universe. and in the other states with probability zero.8).23) Instead of i(x.3 The Probability Density 77 We see therefore that p defines the mixed states.e. that these are actually not sufficiently general. For the case n we have TJ=J (520) (5.e. which is contained in the coefficients Cai.
^ The equation which is the conjugate of Eq. 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. the Schrodinger equation is always postulated in some way.27) is iHJ\ jt= m . (5.78 CHAPTER 5. We now want to obtain the quantum mechanical analogue.22) can also be written (xi) 2 = (i\x)(x\i). Equation (5. .t) = / dx{i\x)(x\i) = (i\i) = 1. To this end we have to differentiate the density matrix p. can be found with probability 1 in the space R1.e.27) is known as the Schrodinger equation. and hence the integral over all space dxp(x.t)\2.4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec. We therefore postulate first an equation for the time dependence of a state vector. (5. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. but the linear operator states are moving — even if the system is observably stationary). (5.8) and multiplying by ih. t). We shall convince ourselves later that the postulated equation is sensible (in fact. (5.28) Differentiating Eq.26) 5.t) = \rp(x. the expression (5. i. (5.3. with respect to time. or the wave function tp(x.27) where H is the Hamilton operator of the system. it is not "derived"). Schrodinger Equation and Liouville Equation The relation (5. The generalization to a threedimensional position space is selfevident: JK 1 / p(x.25) J Thus the particle whose state is described by \i).8). t h e equation of motion with states represented in a system for which the q's or here x's are diagonal). (5.e. which is chosen in such a way that the desired analogy is obtained. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j). 2. This differentiation requires the time derivative of a state vector \i) E "K.
(2. We can now consider the Schrodinger equation (5. p also contains "quantum mechanical probabilities" (i. With Eqs.5.31) The eigenvalues u>i of p determine the fraction of the total number of systems of the ensemble describing the actual system which occupy the state \i).37)) (5.p}.34) . not the partial derivative. (5.24)). (3.27).Xi)\j). we observe that here on the left hand side we have the total time derivative. both uji and i)(i).e. The reason for this is that in addition to the "classical probabilities". M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.e.27) and (5. Eq. i. dp/dt ^ 0.32) with its classical counterpart (2. (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi. We thus obtain the operator form of the Liouville equation. Eq. (cf. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. i.32) Comparing Eq.p} + ihJ2^\J)(J\.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. (5. (5. such a formulation is described as Schrodinger picture. the relation ih^ = [H.e. (5.28) we obtain (see also Eq.28) the states are considered as timedependent. (5.21).
4. a (5. In expression (5. . T meaning temperature.Xi){x\En) = En(x\En). we can replace these states by corresponding timeindependent states. but with respect to the parameter /3 = 1/kT. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H.4. Feynman [94].e. in Sec.** In view of the close connection between time. with respect to time t. the state \i) is still timedependent. (5. (5.1 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. 0) or \il))t = \il>)t=o exp (5.t) = Hld . this is an application of the BakerCampbellHausdorff formula which we deal with in "Later.80 CHAPTER 5.. in p = ^2i^i\i}(i\. appearing in the Boltzmann distribution.35) Here we can look at the Hamilton operator as a "time generator3''. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. i. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.36) however. 10. it is plausible to refer to this equation at this stage.8) for p...t): > ih—</>(x.x 1 ^ ( x . operator" the exponentiated oper . to) = exp[—iH(t — to)/H\.38) (x\j)t=o = Yt(x\E^(En\J)t=o. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. **R.and temperaturedependent Green's functions. t ) . however. With the help of Eq. we shall describe as "time development ator given by U(t. since the solution of Eq. „ / ith—. we write V>(x.e. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. (5. P. i. or H\En) = En\En). n (537) This equation is described as the timeindependent Schrodinger equation.t) = exp ip(x.0): where H(ih—. which we shall need later.35) can be written'! ip(x. 5.36) in in The initial wave function or timeindependent wave function ^(x.
44) r (TrePH)' (5.42) Inserting here Eq.40). i ( 5 . (1. we obtain p(x. (5.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). Eq. x') = ^ i LUiMxWix'). i (5.39) 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.^ n tne position space representation Eq.5.40) 1.e. i. (5. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. we can rewrite p.x'(3) = Si e ~^W*V).45) since on the one hand •PEi i i i and on the other hand .4 Schrodinger Equation and Liouville Equation 81 Example 5. (5.1). or ^ = ^ e 0Ei _m. (5. (543) Since H(f>i(x) = Ei<f>i(x).41 ) or with 4>i(x) := (x\Ei) this is p(x.x')^p(x.
(5.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5. We now rewrite the factor exp(—(3H) in Eq. (5.51) where the subscript x indicates that H acts on x of PN(X.tt tt For a more extensive discussion see R. (5. Tr(pN) = / dxpN(x.x'. B] + L [A. M.(3).82 CHAPTER 5. This solution is the Green's function or kernel K(x.52) Hence this expectation value can now be evaluated with a knowledge of pN.48) = 1.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. Differentiating Eq. [A. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.4> = ^ g ^ . With Eqs.x'.50) Differentiating this equation with respect to j3. (5. Wilcox [284].45) without normalization as PN(J3) := e~pH.55)). (5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3). (3).46) we can write the expectation value of an observable <^Tr<„.46) is pN(x.45)./3).51) is seen to be very similar to the Schrodinger equation (5. (5.35).L [B. (5.x. (7.x'.47) with respect to (3.(3) := (x\pN(P)\x') = (x\e~0H\x'). we obtain = HxPN{x. A}} + • Solution: The relation can be verified by expansion of the left hand side.X'. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5.49) In the position or configuration space representation Eq. Equation (5. B}} .51).47) (5./?). [B. Example 5. (5. (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij. Schrodinger Equation and Liouville Equation so that p is given by Eq.45) and (5. whose calculation for specific cases is a topic of Chapter 7 (see Eq. . (5.
This is a fundamental topic since the harmonic oscillator. Kim and M. E. Noz [149]. 83 . that the oscillator. like the Coulomb potential. (8. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator." Quite apart from this basic significance of the harmonic oscillator. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". can — in fact — be quantized in quite a few different ways.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. 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). An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs.. its 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.51) to (8. * Comment of Y.53)). S. However. Since the spectrum of the harmonic oscillator is entirely discrete. we shall see on the way..the present form of quantum mechanics is largely a physics of harmonic oscillators.
However.to = 1 reduces to the parabolic cylinder equation. . 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. Quantum Mechanics of the Harmonic Oscillator 6. For instance we can go to the special configuration or position space representation given by q^x. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . so that there is no ambiguity due to commutation.1) for the hermitian operators q. which is — in the first place — representationindependent. This can be done in a number of ways. of which the only nontrivial one here is [q. like taking half and half.2 The OneDimensional Linear Oscillator In the case of the onedimensional.' ^mgLOh by setting and *~.p. When such a term arises one has to resort to some definition. in the present case + [EmooJ2x2)ip = 0. p^ih—.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. hence we do not use a distinguishing notation. h = l.2) which for mo = 1/2.^ The first problem is to determine the eigenvalues and eigenfunctions of H. To this end we introduce first of all the dimensionless quantities rriQUi fl q.84 CHAPTER 6. which is called Weyl ordering. 2mo dx \ 2 2 (6. we can also proceed in a different way. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. This equation would then have to be solved as a second order differential equation.p]=ih (6.
C]. We now use the relation (3. (6. (6. (6.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. (6. Prom these relations we obtain ( ^ = ^ . A^] = A\ (6.8) The eigenstates of H are therefore essentially those of N := ^ f A (6. i.12) in order to obtain the following expressions: [tfA.5) Reexpressing q and p in terms of A.B}C + B[A. . With the help of Eq. (6.1) we obtain immediately the commutation relation [A. C] = A[B.13) ( ^ A ) i t = A\A^A From Eq.A^] = 1. C] + [A. (6. (6. for ^ A  a ) = a\a).BC] = [A.14b) [A* A. + 1).6) and p = y/motoh AAi =^.1 ) .6. i. A*} = A^A. (6.34d): [A. A\ we obtain h A + A^ rriQLU y^ (6. q are hermitian). i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .e.e.14a) we deduce for an eigenvector \a) of A^A.10) then a = (a\A^A\a) = \\A\a)\\2>0. A] = [A\ A]A = A. if A*A\a) = a\a).14a) (6. C}B.11) Thus the eigenvalues are real and nonnegative.7) (6. and [AB.2 The OneDimensional Linear Oscillator 85 (p.
l)A\a). or \\A\a)\\ = Va. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). However.11) this is not possible. since this equation implies that the eigenvalues cannot be negative. in view of Eq. If we continue like this and consider the vectors An\a) ^ 0 for all n.2A)\a) = A(Aa . (6.21) .14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). (6. In this case we have (A*A)A2\a) (6 (6.l)\a) = A(a . (6. unless A^\a) = 0.19) i.86 the relation CHAPTER 6.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). Thus for a certain value of n > 0.A)\a) A{AAU .e.16) This means. A 2 a) is eigenvector of A^A with eigenvalue (a — 2).l)A\a) = A(A^AA .  a) is eigenvector of A^A with eigenvalue (a + 1).. (6. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. or Ata) = v ^ T T .18) = '= a ) A(A^A . but (b) An\n>\ llA An+1a)=0. (6. unless A\a) = 0. Similarly we obtain from Eq. unless A 2 a) = 0.2A)\a) 2 (a . (6. (6. we must have (a) Let \an):= Ana)^0. This would mean that for sufficiently large values of n the eigenvalue would be negative.l)\a) = (a .17) (6. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. Next we consider the vector A 2 a).20) .2)A a).
i. (6.22) With relation (b) of Eq (6. (6. Moreover. (6..I n U .17) a by (a — n).27) According to (b) of (6.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n). 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.20) we have for a = n = 1: A2\1) = 0. From relation (6. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. In the following manipulations we always use the commutator (6. For a — n = 0 we deduce from (b) of (6.5) to shift operators A to the right which then acting on 0) give zero.' " /. (6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.18) we obtain for a = n: p t  n )   2 = n + l.e.25) (6. \ I 2 2 x 87 Replacing in Eq. v. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a .20) that A0) = 0. The state 0) is called ground state or vacuum state. \ 9 \ I / II 4 T ) .28) .. (6.n  a . I „ .n ) = " A»a>=» ' = A«a)F = 1 .26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.24) This is a very important relation which we can use as definition of the state vector 0). that an = 0.6. or a = n > 0.
(A^)3] = [A. The states \n) thus defined are orthonormal.31) (6. (6. and in view of Eq. (6. But using Eqs. (0^ 2 (At) 2 0) = 2.26) the equality l) = A t 0). (6. and in general [A.(6.(A*)n] = n(Ai)n\ (6.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.29) Hence we obtain and in view of Eq.88 and CHAPTER 6.e. A^A* + A*[A. as we can see as follows. we have 2> = ^=dAi\0). (At)2]A+ + (A*)2[A. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0. Similarly we find 2)oc^Ut0).11) we have [A. In general we have \n) = ^(A^n\0) (6. A^} = 2A\ and [A. According to Eq.^ ( O l A ^ n o ) . i.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).33) . (A*)2} = [A. (6.32) (6. (6.34) (6.5) and again (6.32) we have (nm) = . l>ocAt0).29).11).
38) (6. In view of the properties (6. we obtain (n\m) = 5nm We also deduce from Eq. however.6. .8) we obtain therefore H\n) = hu (A^A + ]. •(» (6.35) Inserting this result into Eq. With Eq.33).35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. (6. in quantum mechanics.1) (6.40) (6. (6.1 0).39) A^A\n) = y/nA^\n .J \n) = hw (n + i J \n).e. .34) and (6. (6.1 (A + ) m . chosen in close analogy to the postulated "second quantization relations" of field theory. 2 .36) + 1) (6.41) The contribution hu/2 is called zero point energy. 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 .34) A\n) = ±=A{Al)n\0) . . (6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.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 ) .32): rf\n) = .'n\ and = v n ![ ( A t ) M i= + (6. . n = 0 . we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). ldnm = n\5nm. . . .37) n^)711}^) = v n  n .1) = n\n). the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . Here. (6. i. The terminology is. whose number is given by the integer eigenvalue of the number operator N = A^A. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv. l .
39)).7) the energy representation of the operators q and p. (5. ( 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.'\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 .42) We can deduce the energy representation of the operators A.1. En = hu(n + ). (6..2. also called Heisenberg representation..3 The Energy Representation of the Oscillator The energy representation.6) and (6. i. Eq..46) v^ .43) = = y/n(n'\n — 1) = y/n5n^ni. Vi 0 0 0 2mQUJ 0 V2 0 y/3 0 0 0 V3 0 y/l 0 \/2 0 V3 0 \ V • / P = mohu) VT 0 0 0 ° VT 0 V2 0 0 / 0 0 V3 0 (6. and — in fact — we have used this already previously (see e..45) Correspondingly we obtain with Eqs. Quantum Mechanics of the Harmonic Oscillator 6.38) and (6. n = 0..90 CHAPTER 6.g. A^ from the relations (6. other elements zero.44) yfn. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. other elements zero. (6.39): {n\A. (6. (6. is defined by projection of operators on eigenfunctions of energy. Vn + 1.e.
Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). by (cf. 2 . mow / (6.1) and (6.47) LP 0) = 0 . (6. (6. (6. Eq. The ground state wave function (/>o(x) is defined by Eq. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.e. A\Q) = 0.5) are also satisfied as matrix equations.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .24).4 The Configuration Space Representation We saw that the eigenstates are given by Eq.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. i.3)) ^ (q 2n \ + (6. (6.32).6. 6. This is a simple differential equation of the first order with solution (x0) = The normalization constant C is determined by the condition* OO /C 1 = (00) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / OO J —< emou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero.
(6. it suffices according to Eq.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 . c/>n(x) := (x\n) = .54) (6 55) i/4 rzri7r 1 / 4 vo. These polynomials are real for real arguments and have the symmetry property # n ( . (6.92 CHAPTER 6. ) e"^2. (6.e. they are called Hermite polynomials.56) < ^ = We have as an operator identity the relation .l ) > n ( x ) . i.l ) n f f „ ( 0 . so that <t>n(x) = ( . Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator.0 = ( .L ( x  ( A t ) »  0 ) . Vn! Now {XlA (6.2 n n! The functions Hn(£) are obviously polynomials of the nth degree.32) to apply the appropriate number of creation operators A^ to the vacuum state 0). In order to obtain the wave functions of higher states.51) \l 2h {X q \ m0uP)\l 2h \X m0u.
12.12(20 .en_ dt. p. Oberhettinger [181]. Magnus and F.58a) ' T h i s definition — apart from the usual one — is also cited e.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. 2£.g. 93 . (6.6. e.e. d£" (6. (6. (2036(20.58b) by W. ( 2 0 4 . gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt. 81. (202"2.2)m = (±t)m for some function VKO.57b) • Generalizing this result and inserting it into Eq.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.4 The Position Space Representation i. .54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .
(6.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . £. pp. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. = ffen+itO+ntfeniCO.6 £ 2 + 3. 81.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) .59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.^ + n) He n(0 = 0 (6. (6. (6.58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists. (6. £2l. Magnus and F.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. . £ 4 .59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. Oberhettinger [181]. = nHe^iO.He'n{£). 80.(e^/2) or Hn(0 = 2n/2Hen(V2ti). Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1.60b) J—< "See for instance W.94 CHAPTER 6. £33£. (6.59b) = (l)^2/2^.
. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0. p. W.61) e?l*2nl2Hn{Z).62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S..^ = e^/42~n/2Hn(aV2). F(S.64) n\ See e.s) = J2 n=0 (6. Oberhettinger [181]. and so on.6.4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( . 93.g.0) we obtain + s fdFs 1! V ds s=0 + OF ds 2 dF ds2 = s=0 2(f d S )^«)a s=0 = 2£ = ffi(0. Thus the generating function can be written F(Z. Magnus and F. Their normalization is therefore given by OO /"OO / OO J —OO z (6.s) = eM?(Zs)2}(6. (6. s) as a Taylor series.l ) ^ e^HeniO 2 / 4  .s)=F(£. .63) The meaning is that if we expand F(£.
E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. We also see from the eigenvalue t h a t the energy of the lowest (i. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig.e. i. Quantum Mechanics of the Harmonic Oscillator Fig.1.96 CHAPTER 6. 2 .1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. i. In classical mechanics the energy of the oscillator can be zero. 6. In fact we can use the uncertainty relation AxApx ~ H/2. in order to estimate the lowest energy of the harmonic oscillator. mow' Apx = ^m0Luh/2. We see t h a t the lowest state is symmetric.e. for x(t) = 0 and p(t) = mox(t) — 0. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . has parity + 1 . Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. ground) state is fko/2. 6.e. and we see t h a t even and odd states alternate.
6.e. /classical orbit Here q goes from 0 to a. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q).1/a O l/a Rez all at infinity Fig. back to 0. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways.plane Imz z . n*h. to —a.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". Thus.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. 14. This observability can be taken as direct proof of the uncertainty relation.. But it has a pole at infinity. Example 6. Sec. in corrected version instead with n* = n + 1/2. original path q .2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. In the plane of complex q the integral has no poles for finite values of q.plane pole at z=0 ^ . classical orbit i. E . The kinetic energy is always positive or zero.: 27 T mov2a2. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. to 0. mass mo.4) to obtain by contour integration the eigenvalues of the harmonic oscillator. 6. pdq. 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)).1.2 The original path and the transformed path at infinity. The analogous Example 6. setting q = 1/z. 27rmoi/ <f> V'a 2 — q2dq = n*h. In Example 6. like twoatomic molecules.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. and is directly observable. According to "old quantum theory" with integer n. which is the complete orbit. we obtain the integral .
2 . z = 0 Vi 2Kmov(Ka2) = n*h = E/v. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o . With a2 = E. Example 6. for instance — without deriving all properties of the solutions.1)! (6.e. are the most basic and exactly solvable cases in quantum mechanics. Eq.e.) hv. .c2 = (I + 1/2) 2 (natural units. « qdq + I2 = f dq . 6. but here we leave this . since these two cases. 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). / " ^ .98 CHAPTER 6. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . harmonic oscillator and Coulomb interaction.1. In the following we investigate in more detail the important differential equation of the harmonic oscillator.e. (11.2TTI a ^ = .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.5. a J \d Solution: The solution proceeds as in Example 6. (m . Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i..1. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z .2: Contour integration along a classical orbit Setting q = 1/z. i. £ = n*/w = ( n + .V62 a2). mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). we start with a Laplace transform ansatz for the remaining part of the solution. E = Z2e4/4(l + n+ l ) 2 (cf. After removal of a Gaussian or W K B exponential. hence we consider the following differential equation in which n = 0 . 1 .5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials.( 6 . i. 6.2b = Ze2.3.5 .114a)).65) f ^Va2z2 . in q at infinity). . .a) m = 2 . and obtain with this all properties of the solution.
p t dp = \ps(p)ePt} . and / dpf(p)pn the Mellin transform of / ( p ) .t*ij& + ny{t) = O. so that the integrand of the remaining integral vanishes.69) **See e. (6. E.6.e. (6.i ^  (ps(p)) P n P Integrating this expression. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). 80.g. . (this determines the limits). p.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. W. (6. fp{ps{p)) = {v2 + n)s{p).67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0.111)) the Mellin transform of e~p is n!. . Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. i. (cf.g. (11.66) First we have to remove the £2term.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . (6. Then (with partial integration) *dt = ~*/Ps(p)e. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) . Magnus and F. Eq.J ±(ps(p))e*dp. We demand that C(p) := — \ps(p)e~pt] = 0. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. Substituting these expressions into Eq. / dpf(p)e~px the Laplace transform of f(p). This has to be true for all values of t.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp.68) § = J p2s(p)e*dp. Oberhettinger [181].
For singlevaluedness. (6. as indicated in Fig. so that y(t) ~ exp(i 2 /2). n must then be an integer. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. i. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. 6. Therefore we take ptp2/2 y(t) .100 and hence CHAPTER 6. Hence a possible path of integration > is from — oo to oo.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. We now investigate possibilities to satisfy the boundary condition C(p) = 0. (3) We can choose a contour once around p = 0 in the complex pplane. This can be attained if p is allowed to be large (t can be of either sign). since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. We can show that if \p\ is allowed to be large.e. (2) If n < 0. we can choose a path from zero to infinity (the condition will vanish at p = 0).71) . The quantum mechanical wave function is then (cf. consider the exponential in the integrand. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6.3 The path when n is not an integer.3.70) = 0 P. 6. = / • Im p Rep n not an integer Fig. This exponential assumes a maximum value when pt \—p2 = minimal. so that p cannot go to infinity. when p = —t (which is possible. of course). To see this. (1) This will be satisfied if \p\ — ±oo. The value of the exponential at this point is ~ exp(£ 2 /2).. Eq.n+l between the two limits of integration. then for large t we have y(t) ~ exp(i 2 /2).
(676) .5. pt p2/2 (6. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity..5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. 2 .l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.if n is odd. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( .. 4 .P 2 / 2 00 C—tY ( . . . We can also obtain the derivative form of this Hermite polynomial.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. if n is even and v = 1.. this becomes jfj.3. . In y(t) above.„y2 v l { ' ^ r (^) \ 2 ' where v = 0. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). We have 27T! Setting q — p + t. .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.
)} 2 /2dt = ePde(P 2 W)/2^^} J —CO .5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Solution: We have to evaluate °° 2 .n— 1) = n.102 CHAPTER 6.3 and 6. n + l) = l. 8. (6. one can construct the tower of polynomials Example 6. Example 6.59a).3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.75) as Hermite polynomials. and elsewhere. (6.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). (n. 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+<.n + l ) f f e n ^ i ( £ ) + (n. in radiation problems. Hen(t).7 that we use throughout this text) and their orthogonality and normalization to Examples 6. Starting from Heo(t) = 1.70).4. as in the normalization of asymptotic expansions of Mathieu functions). we obtain tHe„{t) (l)"n! ~ i. Reinterpreting the remaining integrals with Eq. Hei(t) Henj.e.g. Pptp 2 / 00 /2 dp. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. (6. n — l)Heni(t). 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. ( n .1 '2dt using Hen(t) = (l)n27 r . 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).4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. for the evaluation of expectation values.77) = t. Example 6.
so that /2 Q2(n+rrf in / ( p ./(n __ 2r — /i)!. (6.78) \/27rn!.jQ) n n!I ^—' ^Q fi\(n . Hence the normalized wave function of Eq.66) is \Zn\V2w Example 6.80) f(p.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. 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.q) = = e . and 0 otherwise \/2n(—l)n+mn! if n = TO.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.^ + 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 .6.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2). g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. zero otherwise (6.n)\ i s an+2rij.82) ^ . (6.5 nm .q). the coefficient of qn+2rn the coefficient of p"qn+2r m exp(. g) is n! Next.
l ) ( n . n — 1) are given by the first index.e.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. n.n + 2) + (n. n + 4)_ffe n +4(t) + [(ra. n + l ) ( n + 1. n + l ) ( n + 1. n ) ( n . The result then follows with half the normalization and orthogonality integrals (6. 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 . by n!(7r/2) 1 / 2 . Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).— — ^ . n + 2){n + 2. It follows that the coefficient of a 2 s / ( 2 s ) ! in this expression J is the quantity K given by f^ot2sge„(t)Jfen+2t. K is the coefficient of a / ( 2 s ) ! in the expression J .v)\ ^ Multiplying both sides by the value of the normalization integral.A i ) ! (6. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . i / ! ( n . the coefficient of T 2s »{sr u)\ As we observed.79). n + 2) + (n.104 CHAPTER 6. .u)\ 2s<—" v\(n . n + l ) ( n + 1.A o?(. the relation (n. . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.e. n)(n. Prom Example 6.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ .2s _oo ( a 2 / 2 ) i .(t)e«2/2dt Now.^ ) ! ( n + 2 r ." ) a2^) r coefficient of (2s)! in f^ (s .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . 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) + .' .r .^ / 2 * . n + 2)(n + 2. we obtain the result (6. n + 2) + ( n . i.v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .e.( n + Jr).ra + 3)(n + 3.n+ l)(n + 1. i. n + 3)(n + 3. /i = (n — v) _ / r o o e .. n + l)(n + 1.3 we know that upgoing coefficients are 1 and downgoing coefficients (n. 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 (6.77). n + 2)(n + 2.r .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . n + l)(n + l . n + 2)]ife„ + 2 (t) + ••• . n + l ) ( n + 1.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .i / ) ! ( 2 r + i/)! > .e '' ^ o^M ! ( n .78).83) = (n + 2r)!e a / 2 V „ „.l .v)\(s . i. n .
which is normally not discussed. in Chapter 21 (with a different calculation). This example also offers a convenient context to introduce the inverted oscillator potential. The first of these cases will reappear later. We consider first timeindependent Green's functions. we consider the sojourn time of a quantum mechanical particle at a point.Chapter 7 Green's Functions 7. in Feynman's path integral as the kernel or free particle propagator. We 105 . and is thus an important point to be noted here.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. which is a point of unstable equilibrium for a classical particle. more generally. of a differential equation of second order. With reference to this case.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms.x'. 7. and those which are timeindependent. In particular we derive the timedependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. which we write G{x. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. These are the Green's functions of the timeindependent Schrodinger equation or.x').t). which we write here K(x.1) be the timeindependent Schrodinger equation which has to be solved. Let H^ = E^> (7.
(7.g. a perturbation part like (3q4. for instance.g. . cf. e. Green's Functions H = H0 + Hi.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x .4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). e. appropriately decreasing behaviour at infinity).x') can be written t G (7.x') by recalling the completeness relation. (7. since the generalization to higher dimensions is selfevident. We can readily see that G(x.x'). although the explicit calculations can be much more involved. Eq. and Hi is some other contribution.2) where Ho consists. square integrability. of the kinetic part p2/2mo and some part of the potential.e. The boundary conditions which G(x. (7. i. x') has to satisfy correspond to those of * (to insure e. The timeindependent Green's function G(x.106 assume a Hamiltonian of the form CHAPTER 7. aq2.g. 'The case of £(°) equal to some Ei is considered later.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case.13). We can obtain an expression for G(x. where £ ( 0 ) = lim E.e.EW)xG{x.x') is defined by the equation* (Ho .x') or ^2^(x)^*(x') i = S(x .x'). i. (7. the relation X') = 5{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'\ t) (7.x'){E .7) with the initial condition K(x. (JL x'\ t) = H0K(x..x'.7.E<®)xG(x.x')(E . (7.(*'))*fr') Hi{x'))^(x') f dx'5{x .t) since ih^K(x.0) = S(xx').x. We see that the initial condition (7.E^ We can check this by considering (H0 .9) = i H0YieE^i%(x)^(x') = H0K(x.x'. i (7. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E . .x'. The timedependent Green's function K(x.£(0) .t) (K for "kernel") is defined as solution of the equation ih—K(x.E^ of the eigenvalue. x'){E .11) = = 3) = fdx'(H0 .x'.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.) = f dx'G(x.EM)X*(X) ( .#. (7.Hi{x'))^(x'). For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.t). (7. . We can see the significance of the timeindependent Green's function for the complete problem as follows.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).10) The contribution on the right hand side is the socalled perturbation contribution. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.E® (EE{0) Hi{x))m{x).
Examples in various contexts have been investigated by L. MiillerKirsten and A. All such methods of solution are based on the analogy of Eq. see E. integral equations are more difficult to solve than the corresponding differential equations. .108 CHAPTER 7. H.12) Equations of the form of Eq.11) with a system of linear equations of the form Vi = Mijyj. (7. see B. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). In cases where an inhomogeneous contribution is given. (7.6) is not defined. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. Merzbacher [194]. J. if we assume that E^ = Ej.2 (not contained in the first edition!). W. In general. Bleecker [33]. that (E .. Green's Functions Can we add on the right hand side of Eq. (7.x'. Wiedemann [183].15) ''This problem and its circumvention can be formulated as a theorem. however. 2nd ed. Instead of Eq.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. But then the Green's function (7.e. (7.t) for the complete problem as follows. (7. then at best such a contribution would be something like cipi(x). the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. D. Sec.6) we would have (7. 17. But if we assume that E^ ^ Ei for all i. This will be different. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). The theorem is also known as Fredholm alternative. Booss and D. Maharana.14) We can now see the significance of the timedependent Green's function K(x.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). i.
.20) the Green's function K(x.x'. Merzbacher [194].17) as initial condition is obviously W)t = Y. for the oscillator potential. 1 See e.Q) = 6(xx').t) describes the evolution of the wave function from its initial value ^(a^O).7.t)il>(x'. p. this relation provides the probability density \ip(x.22) According to Eq.t) is known. t)\2. x'. (7.t) = J2 fdx/^n(x)^*n(x')^(x'.2 for the computation of the sojourn time.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.7) with Eq. t > 0. *>0. (7. (7.21) dx'K(x. we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.g. i. 2nd ed.158. (7.15) which contains (7. \En)(En\ip)t=oeE^\ n (7.x'. x'.0)eEnt/in. (7. H\En) = En\En).t) = where if(x^'. in different formulations. .x'. Comparison of Eq. 109 (7.16) As usual. (6.20) is the timedependent Green's function.e. We use this in Sec. (7. e.0). We can write this expression also as§ ip(x. E.g.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. t) with the Note that when K(x.18) or. (5. (7.t) = ^ e £ " f / « V n W < ( ^ ) .' which obviously satisfies the initial condition K(x. 7.The solution of Eq.5.19) i/>(x.17) is the initial condition of \4>)t. Eq. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7.
we obtain then also the corresponding density matrix.1 The contour of integration. x')6(t) (7.i /f e «/. t > 0.x') and K(x. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.t) K(x. Green's Functions difference that (5 = 1/kT plays the role of it.«^«2 W .23) along the contour C in the plane of complex E as shown in Fig.25) in agreement with the timedependent Green's function . 7. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x.1. we obtain / W : = . As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions.t). Jc 27T „ hn .x'. x'. 24 )  ReE Fig. i. t) = J2 e £ " f M i ( x ) < ( x ' ) . 7.110 CHAPTER 7.e.x') = GE(x.E . (7. Inserting for GE+ie the expression above. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.x'. between (with E^0' = E) G(x.ie (7 . In the following we shall derive the Green's function for the case of the harmonic oscillator.
In this case the Green's function is the solution of the equation d hK(x.x'. 2mQK h h2 ih(AB) = —{4B2A). i.7.X>.31) The constant A has to be chosen such that K(x. we obtain *I£I = h2 —(2AB).e.3 The Green's Function of a Free Particle 111 7. It is clear that it is nontrivial to solve an equation like Eq.28) B(xx')2 t2 DB(xx') 2 /t (7. v ' 2m 0 v h i.x'.0) = 5(xx'). . (7.e. (7.t).29) and (7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz. which is moving in one space dimension. Thus we try the ansatz.26) This is the case of a free particle with mass mo.29) tJJ 2 2B(x .t) t h2 = .30) Inserting Eqs. B m0 n _ 2ih 2iK (7. A and B being constants.x>.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.— d2 —2K(x. and identifying coefficients of the same powers of t on both sides. / dxK(x.30) into Eq. K(x. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.27). (7.3 The Green's Function of a Free Particle The timedependent Green's function of a free particle which we now derive is an important quantity and will reappear later in Feynman's path integral method. (7.7).x'.0) = 1.
we > have to make the replacements V ^ fdk.20) and can then obtain ip(x. from K{x. Green's Functions For parameter values such that the following integral exists. 158.34) K(x. 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 . (7.x'.x'.x't) = Y. Merzbacher [194]..ihMxWn{x'). i. 2m0 (7. j3 = i(xx'). 32) It follows that x'.38) ."737) \27rii/i in agreement with Eq. (7. (7. *>0. p.112 CHAPTER 7.t) = — / dkeak2+Pk 2vr J. "See the excercise in E. we have This is 1 provided A=JB= ^ (7 .t) — for instance for a wave packet given at time t = 0 of the form ^(x.H We can insert the expression for K(x.35a) K(xy.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. (7. 1st ed.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. Mx)^M^) „ n so that J =^ .33).t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. v27r En^^. (7.oo 27T (7. (7.zEnt. t) = /jJo_ e mo(««') 2 /««.e.(7.21).0)oce"Qx2+ifcox.t) into Eq.
t) We now set h2 d2 = . §jK(x. i. .7) is the solution of d ih—K(x. / ) = ~^K(x. 1 2mo u>n ox n with the initial condition (7.t). i.e.x'. K(x. . x'. x'. .x'.. z .0) = 6(xx') at / = 0.t).x usn ot i. HQ = ^—p2 ZrriQ + \m0uj2q2.d r. (7.x'. (7.37) will later be obtained by a different method — see Eq..t) = h2 2 d2 TW . 7./ ) + fK(x.e. . in this domain *Recall that S(x) = ± f eikxdk. x'. (7.t) 1 + m0u2x2K(x.25) — in the context of Feynman's path integral method..x'. i. —ih—K(x. mQLu .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. / ) (7.40) becomes 2 . —KK(X.e. 5{ax) = ± J eikaxdk = ArS(x).e. the relation* 6(x) = o<y(os).40) Then Eq. . We consider the onedimensional harmonic oscillator with Hamilton operator HQ.x'.39) In this case the timedependent Green's function K of Eq.42) .t) + ~x K(x.33). (7.X .——K(x. (7.8).3 The Green's Function of the Harmonic Oscillator 113 The result (7.7. We rewrite this initial condition in terms of £ and use for a = const. . (21. I (7.4 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.
47) becomes (7. and we expect K(x. we obttain a' = l .45) in this sense.x'.47) We insert this ansatz for K into Eq. (7. f i.45) in accordance with (7..) a'£2 + b'i + c' = (1 . (7.50c) (7.4a&£ + 2a .b2. i. (7.114 CHAPTER 7. so that for / — 0: > a^—.42) and obtain the equation (with a1 = da/df etc. as the limiting case / — 0.46) If we interpret Eq.50a) (7.c o t h 2 / .50b) (7. (7.b2. Integrating Eq. (7. p. that is.50a) we obtain (7. Feynman [94]. W ) . it • is suggestive to attempt for K the following ansatz^ m. Green's Functions to the particle behaves almost like a free particle. 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.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.48).^ . Identifying coefficients on both sides.t) become similar to expression (7. (7.33). we must have /o = 0. .f) with a(0)»y.51) See R. a = .e. (/ = 2a. b' = Aab.48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.4a 2 )£ 2 . c(0)^. P. 50.e.4 a 2 .Z'.
In B.e. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. Comparing the Eqs./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. In order to satisfy Eq. with A. c(0) = £' / 4 / . (7.e. x'.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. (7. / —————rexp y 27rmsin(u.i) (7.55) 'For an alternative derivation and further discussion see also B. 2TT^ Inserting a(f).7. i. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . i.27). and to ensure that we obtain the prefactor of Eq. K(x. t.5. B independent of / . Finally Eq. as we shall see in the following. if we return to x.t) = . (7. 7.53) or. . we must have (besides A = —£') = /m 0 u.48). we can use K(x.x'.0).b(f).x'.45).48) 6(0) = — £'/2f.50c) yields for c(/). as one can verify. (7. we must have A = ('.33) for the Green's > function of a free particle. 115 To ensure that in accordance with Eq.51) for the density matrix PN(X. x . Felsager [91].54) For t —• 0 this expression goes over into the expression (7.40) of the timedependent Green's function with Eq.50b) Kf) = —r^77) smh 2 / A independent of / . (7.* With this result we have another important quantity at our disposal.2.t) to obtain this element (x. (5. (7. 174. (7.X'.c(f) K = 5 =exp Vsinh 2 / into Eq.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. (7.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. in particular for the derivation of the sojourn time in Sec. p.
eP E i = i + y.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. (7.. <2 (7. p.e. = jdxp(x.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.52)) the expectation value of an observable A in the canonical distribution (i.56a) the expression (7.P) 2. Thus the system is in a mixed state and the expectation value "Cf. Feynman [94].x. at temperature T): {A) = Tl{pA) = ^ > .52): Tr (PNq2) Jx2pN(x. we obtain § < Z> = ^ c o t h ^ ^ ° — . )x Inserting into Eq. For instance we have with Eq. 52. P.p)dx W ) = —^r = —p .eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. Eq.116 CHAPTER 7.uij>0 >• 0.» l. Eq.56a) {q2) = = Y.. ^—• Irpiv J pN(x. Green's Functions With this expression we can evaluate (cf. [ [ fdxdx'dx"(i\x}(x\p\x')(x'\q2\x")(x"\i) Idxdx'dx"{x"\i){i\x){x\p\x'){x'\q2\x") J2 [ [ i = fjf^»(x>)(xW)(xW) dxdx'dx"5{x" i.e. 11 For T ~+ 0: u>0 . (5. . we verify the relation: (Q2) ~ dxdx'(x\p\x'){x'\q2\x) = / / dxdx'(x\p\x')(x'\x) x2 x){x\p\x')(x'\q2\x"}. (5. (5.55).40)) 1 epEi 1 Ui = y. we skip the algebra here.x.x. R.
7. of K(x. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .54).54) is K(x. e > 0.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . 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 (7.3 The Green's Function of the Harmonic Oscillator 117 (7.50) we obtain. We assume t > 0 and t —> oo. (6.t "With the normalized ground state wave function of the harmonic oscillator given by Eq.59b) n+ l \ ftwt here 2hf exp _ i ( n + . x'. This is the first and hence dominant term of the expression (7. which means in the oscillator state \i) with eigenenergy fku(i + 1/2).56b) is that with respect to this mixed state (whose cause is the finite temperature T). setting a = mooj/h.59a) E 0 = ^fiw. 2 + x'2)le^2xx' to—>w—it t—>co 2 .21). /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. x'. In this case the Green's function K(x.J (w — ie)< e (n+l/2)Et e i(n+l/2)u. i. 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. We return to the Green's function (7.x'. and we replace to by LO — ie.e.QUJ (7. (7. t) of Eq.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. 1 mo^ exp lmow / 2 2 ft X (7.
Considered classically. the velocity of planes of equal phase. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0..1 we estimate T semiclassically. k = k.59a). as in Eq.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. i. Green's Functions For t large (i.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. In Example 7. (7.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. (7. is defined as vv = . of dpN as pN(x.e. However quantum mechanically in view of the uncertainties in position and momentum. to infinity) the contribution with n — 0 dominates.e. and with this we estimate the sojourn time T. the particle will stay there only for a finite length of time T. lie on surfaces.118 CHAPTER 7.e. In a very analogous manner we obtain the solution of the equation for the density matrix. The wavelength A is given by The phase velocity v<p. i. 7. In the following we want to calculate (more precisely estimate) with the help of the Green's function the time interval T which a pointlike particle can stay at the maximum of the potential before it rolls down as a result of the quantum mechanical uncertainties.62) . (7.x'. which are planes.5. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely.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)].
63) The relation u = w(k) is known as dispersion or dispersion law. e .)ll/(fc")l[cos(¥'(fe.Jt + a.Q ( k . ' t) dk'. i. ^ ( r .64) where / ( k ' ) differs substantially from zero only near k' = k.)¥'(*:")) +ism(<p(k') . to the function ip given by oo / fik'y^'x^dk'. (7. i. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. then ^(r.e. or dco' da The centre of mass determines the particular phase.e. / oo J—oo . A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. We now ask: How and under what conditions does the time variation of the function ^(r. e.k ') 2 . If we assume for / ( k ' ) a Gauss distribution. The wave packet describes a wave of limited extent. for which \ip\ assumes its largest value: OO . One defines as centre of mass of the wave packet that value of x for which d(p — = 0. oo (7.<p(k"))}dk'dk". a > 0 . u = w(k). xt— + — = 0.7. i ) = /*/(k')e i(k '" r '' .65) Let f(k') = \f(k')\eia and <p := k'x . (7. the fundamental postulate on matter waves in Chapter 2): E = hu. duo' da . /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.e. i. t) is the spatial Fourier transform of this Gauss distribution — as we know. essentially again a Gauss curve.
1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . as claimed for =0. we can use a simpler method. In the present case. Solution: The function to be calculated is the integral dxeax OO ezkx. (7. a > 0. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. Example 7.^ g=r a rad. = — (768) It is instructive to consider at this point the following examples.fc w> vv = ~a^ = S r a d P E «i7 Sd u . 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 = ..e.120 CHAPTER 7. .67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. i. . (7. For E = hio. The threedimensional generalization is evidently du . dE v 9 = . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const.— g(k). p = hk the group velocity is equal to the particle velocity v.69) In general one uses the theory of functions for the evaluation of this integral. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. however. we obtain the deBroglie relation p — hk.
g .0 0 7T 7 0 7T e From Eq. (7.r „• With the help of Eq.75) ' E .7. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. (7.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .. (7. W. MiillerKirsten [215]. Appendix A.72) Solution: From Eqs.Afc = 81n2. 2 f + .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 . or see H. J. (7..a) + 6{x .1 verify the uncertainty relation Aa.71) (7.69) and (7.2: Representations of the delta distribution Use Eqs. XZ (e > 0). .&)] for a + b..70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^.7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example. (7.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. \a — b\ (7.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .74) Example 7. where c = 9(0) is a constant.69) and (7... with partial fraction decomposition.70) we obtain 6(h) = lim — f°° dxeax2eikx = . Example 7. (7. (7.73) 27T 7 .
) leads to a very broad maximum of the Fourier transform \g(k)\. Physically . (7. is Afe = 2 V 2 1 n 2  (7.78) It follows that the product of the uncertainties is AxAk = 8 In 2.* o ) 2 / 2 .2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. and hence implies large values of Ak. In quantum mechanics the square of the modulus of the wave function ip(x. requires according to Eq. (7. where  / ( x )  = m a x  / ( x )  / 2 . A simple calculation yields A i = 2v/2~In2a.70) the Fourier transform of f(x) is (7.g.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.CHAPTER 7. 7. Thus a sharp maximum of the function f{x).77a) small values of a.e.77b) (7. The breadth A/c of the curve g(k) around k = kg.« 2 ( * .e.2 The Gaussian curve. The wave function corresponds to the function f(x) in the above considerations (e. when Ax is very small. Thus a slim maximum of the curve of /(a.r)1/2a Re f(x) Ax Fig. the width Ax is a measure of the uncertainty of the probability. 7. Green's Functions 1/(27i)1/2a — 1/2(2. where g(k) = max<?(fc)/2. According to Eq.t) is a measure of the probability to find the particle at time t at the position x. at time t = 0). Solution: In Fig. i.77a) g(k) = e . i.
(7.5. 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. Barton mentions at the beginning of his paper. 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.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x.2 ) and U)t. (7.h = 1. This means the particle is localized at x = 0. Barton [15]. ^In Eq. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. The result (7.e. As G. 7. nr = S(x).81) *G. In the limit a — 0 we obtain from Eqs. the less precise is the determination of its associated momentum.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «.. <7*» We assume that initially. The function g(k) is therefore described as momentum space representation of the wave function f(x).79) According to Eq.54) these parameters appear in the combinations moui/h (dimension: l e n g t h .u> = l.54) with the substitution ui — iu. Barton recalls that W. so that Ki0(x. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} . For reasons > of simplicity in the following we set in addition t mo = l.71) > and (7. (7. at time t = 0.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq.". O ) .. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.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.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . Thus G.«> = / ^ C ' W . which means all values of the momentum are equally probable. . (7. E.7. i. (7. 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.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.x'.
(Ax' + Bxf \/27risinh + (B2 + ^cothi)a. so t h a t (this defining C(t) and the imaginary part ip on the right) {B2 + icotht) sinh 2 1 + (i/26 2 ) sinh 2t ( l / 6 ) s i n h 1 .84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .79) into (7.t) = J oo P ..87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.85) = —r= o2 7 2 sinh 2i.cosh t sinh t' Wisll2b ^ (7.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7..2 (7.124 CHAPTER 7.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7. 9 (7. Green's Functions Inserting (7. + cosh 2 1.t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.82) i/f(x. (7.81) and (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .84) we obtain* ip(x. (7. A . = .86) Evaluating the Gaussian integral of Eq. ' 2\Aswht % cosh t (7. .83) ip(x.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.88) .
86) has C2(t) := b2 cosh 2 t + 72 sinh 2 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= / (7.89) = exp[x2/C2(t)} 7rV2C(t) (7.t)= where Co(t) / dx\^(x..t)\' 2 125 (7.7. In taking the modulus of ip(x. C2(t) (7.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .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).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.t)\2 Ji == 71 / V " Jo dO ? 6' (7.£ 2 (7.91) C(t)' (7. 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. We have therefore .95) . (7.90) with the expected limiting behaviour \im \ilj(x.94) / dxf(x)=g'(y)f(g(y)). •&(t)=//c(t) = / or.4 The Inverted Harmonic Oscillator The real part of Eq. _2 1 exp C 2 (i) = l + 2sinh 2 i.
99) . p.1 ) 2 \n J ^"•""ff^^V^K?..2)(p_7?2)(7.e ~' V^~i Jo Vn~i Jo ^ rx=VM _Jl 2 Using the following expansion of the integral* _ _ r e~*' dt L 2 « 1 2e~*2/2 7T X /2^7. C(t) = .98) We insert this into Eq.(4 2 .From this we obtain C(t) C'(t) Z2 ^(r?2_..=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. Dwight [81]. + 0(e'2) (7.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.126 We set I C(t)' so that dT ~d~l Now. we obtain dT _ 1 ~dJ~l 'See for instance H. B.=0 V^ C'(t) I Jt=0 V^ (7. (7.97) sinh 2 1 = .96) °° l Jr. 136.= \ / l + 2sinh 2 t.
also motivated by this paper.7. (7. 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. U) (7. 238. Fig.4.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.g.100) Reintroducing the dimensional parameters we had set equal to 1. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7.102) 'As a matter of interest we add that in the theory of a free scalar field x. It follows t h a t T~lnZ. . Explain the parameters entering the calculation. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". Zwiebach [294].4 The Inverted Harmonic Oscillator 127 for I sufficiently large. T h e following Example 7.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. B. p. The equation there describes the classical "rolling down" of this tachyon.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down. this is T ~ ]n(ly/mou>/h). this equation is the equation of motion of this scalar field with negative masssquared. See e. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. 7.
For h —> 0 the time T —• oo in agreement with our classical expectation. 2 x(0) ^ where A i . 2 10 x(0) = B u . but instead are subject to an uncertainty relation of the form AxAp > h. Green's Functions x = Aco sinh cot + Bco cosh cot. (p2):=mW(x2). the smallest macroscopic length. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. so that e.128 with x2 = LO2X2 + const. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). 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). (x 2 ) h 2mou 2mou> and so l^J^ULl). Quantum mechanically x(0). pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.e.p(0) cannot be determined with arbitrary precision.102)) p(0) = ^/(p 2 >. ^(x2)(p2). Eq. v A x A p = \ (x2)(p2) v = m0u(x2). and solution x = A cosh Lot + B sinh cot. (7. and for simplicity we take (cf.e. 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 . 2V ' Let A and Bio be the values of x and x at time t = 0. Thus quantum mechanically x(0). Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. Ap= v /(p 2 )<p)2. i. (10. x(0) = A.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. A p are defined by the mean square deviations given by (see Eq. i. .5)) Ax = <J(x2){x)2.g. / 7 ^ 2 =2J(x }. (7.
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.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E.e.(i)+/^(2) + ".2h2 cos 2 a # = 0.Chapter 8 TimeIndependent Perturbation Theory 8. 129 .E^ is already a bigg problem. i. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. In general perturbation series do not converge. Frequently even the calculation of the next to leading contributions ip^. For mathematical purists the question of convergence of the series is an even bigger challenge. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. one would set tt E = = ^(°)+/ty. in closed form) only for very few potentials. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.e.
which is known as the Mathieu equation. 0. E. Schafke [193]. In the following we explain the difference between convergent and asymptotic series. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified.+ . N. Meixner and F . h hl are in the strict mathematical sense divergent. concerning also the uniform convergence of the exponential series. Whittaker and G. .g. (8. T. It will be seen later (e.. W.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). 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.. (8. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ .= 0 < 1. 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. J..1) and its terms with the behaviour of the following series and its terms: f(x) = 1 .130 CHAPTER 8. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. the expansions in descending powers of h2.. Watson [283]. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence.2) n—»oo n It is interesting to compare the behaviour of the series (8.2 However. . e.3) *See e. 8. This convergence can be shown with D'Alembert's ratio test since''' lim .g.g. see e. at the end of Chapter 26) — and by comparing the results of WKB.* i. 581. ^For further details. E = E^ + h2E^+hAE^ 71 72 7 + .e. p.g. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •.
. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. p. and we see that the moduli of successive terms first decrease and then. /(1000) = 1 . .33333 .222. begin to increase again. for larger values of x. and in fact increase indefinitely. We can see this as follows.g. E.0. / ( I ) = 1 — 1! + 2! — 3! H • However. in spite of this the series (8.. e.1). N. 235.3) with the convergent series (8.. + •••.4) for every value of x. . + 0.000 000 06 + • • • .0.. We observe in the first place that since lim n—>oo co (8. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . for x = 3..9876.376 . after reaching 6/27 = 0. i.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.000 002 00 .. T. Whittaker and G..g.29629. However.001 + 0. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. + 0.3) diverges for every value of x. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. .22222 . the series (8..2. .8.e..5) Cf. It is evident that the larger the value of \x\...22222 .0. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .l n ( l + ^ ) n=l L ^ ' (8. 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.4938.0. +0. . 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. Watson [283]. Thus e.0. Comparing the asymptotic series (8. the better the approximation obtained. ..3) can still be used to obtain almost correct values of f(x) for x sufficiently large.
41) and (16.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. This latter observation hints already at the importance of asymptotic series in applications. meaning convergent. 152. (16. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). It is interesting to observe the comments of various authors on this point. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. In many respects Messiah aims at more rigour in his arguments. N. it is sensible to make the assumption. E.1 between Eqs. . although this has not been investigated except for a few simple problem^. Messiah [195]. 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. **A. whereas applications of the convergent series (8. This is not appreciated by mathematical purists. p. the dominant approximation is.1. Watson [283]. 11 L. that E and \ijj) (i. Whittaker and G.132 CHAPTER 8. In fact.5). 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." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. ^E. p. T. TimeIndependent Perturbation Theory where 7 = 0. ip) can be represented by rapidly converging power series in e. Sec." These "rapidly converging power series" in physical contexts are most likely very rare cases. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. the leading term of an asymptotic expansion. p.e. Schiff [243]. 371. Vol.5) are practically unknown. § However. 16. I. 236. in fact.J In2. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. Merzbacher [194]. 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.
p. (8.7) as / e'* dte* dt = 1 . Replacing t h e exponential in Eq. B.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. (8. It is therefore suggestive to write iy poo <t>(x) = 1 .. solutions of second order differential equations.." Of these three authors t h e first. We know from books on Special Functions t h a t all of these functions. Dingle [70]. ft Jx du = 2tdt. which have been studied in great detail. In fact. Vol.10) Vn Changing t h e variable of integration to u = t2x2.9) JO Jx J V 71" Jx The integral is expected to b e dominated by t h e behaviour of the exponential at t h e lower limit. II. (8.11) A .= / e~t2dt. we can rewrite Eq.7) This function has been considered in detail in t h e book of Dingle. (8. Messiah [195]. there is no need for such bias.* and we follow some of t h e considerations given there. 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. He says.^=e~x2 / e^^dt. Merzbacher.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. (8.. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion .8. *R.2. 8. 198 (German edition).2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . 2 (8. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. .
e.12).^ X e U ~ ( 1 + * \ 2 .^e~x \Ar 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. since (—4)! = y/n.' we have 2 nH 7T 7T (8.e. du.i)[(_!)«' Then.14) sin7T2. we obtain 1+ i. TimeIndependent Perturbation Theory <P(x) = 1 .12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.16) . (8. Using the reflection formula (*!)!(*)! = for z = n + i . 7T (ni)!sin{7r(n+^)} ( „ . 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .13) presupposes t h a t < 1. Then e (n —\)\( 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 (8.134 we obtain CHAPTER 8. (8.15) < 1 we can insert this expansion into Eq.
(8. convergent) only in the restricted domain u < x2. In the foregoing discussion of the error function we assumed that x was real and positive. 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. We now see how the asymptotic expansion (8.—yS!L^LLZ e x OO / _ 1\ X2 /CO dueu L \ / V n = Q 1 and we write 0(X)~1 V 7 n=0 v 2^. whereas in a convergent expansion the terms first increase in magnitude.17) 0(X) = i.l)! = / Jo Then 2 cHnxdt.19) approximates <f){x) the better the larger x is. /•oo T(n) = (n . It is fairly clear that the above considerations remain unaffected for argx < 7T.r<7r. In an asymptotic expansion the first few terms successively decrease in magnitude. much more useful than the convergent expansion. and in fact so slowly.e. the asymptotic expansion of the error function is.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. i.19) originates. The expansion (8. We can actually understand the deeper reason for this in practice.e.15) into (8.12) ignoring the fact that the latter is valid (i. (8. (8. ' arg.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. Whichever way we look at the result.8. . argx = 0. i. (8. As in the case of the gamma function.18) or that we insert the binomial expansion (8.2 Comparison of Asymptotic Series with Convergent Series 135 We can evaluate the first integral with the help of the integral representation of the gamma function.e. Frequently "=" is written instead of "~". We now want to relax this condition and allow for phases argx ^ 0. in general.
21) = y2s2.2 y V . (8. dv = 2sds = . VK Jo V^ Jo (8. (8. (8.136 CHAPTER 8.24) We can read off the binomial expansion of the factor in the integrand from Eq. we set x = iy in Eq.23) Throughout the range of integration 0 < v < y2 the integrand is real. thus with \v/y2\ < 1. however. we can try to proceed with Eq. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.26) or else return . Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. (8. i. =dv.^ = . V7T Jo Changing the variable of integration to v (8. The situation becomes critical.e. 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).20) where we set t = is.vds.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . when argx = 7r/2. (8. We therefore consider this case in detail.^ f V e ^ .15).22) we obtain 2 4>{iy) = . (8.
30) yvn f Jo V y J Cf.27). For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. and the larger y.e.28) ignores the integral contribution of Eq. However. y y 2 ( n _i)! y arg(* = iy) = i * . (8.e. (8.27) Then .17): J£° e~vvndv = n\. for large values of y.^] * dv. e ^ ( i y ) ^ ^ . i.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. and so the integral is real.^ ^ p .27) and (8..24). i (8. ey is much larger than something of order 1/y. (8.8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . (8. (8.28) n=0 We observe that this expansion differs significantly from (8. the better the approximation expressed by Eq. We have therefore <j>(iy) = ^e^TL fV e~v ( l . (8. If we proceed with Eq. which will be studied in more detail below. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . Returning to Eqs. We know that for v < y2 the integrand is real. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. (8.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . — (8.28) because then the correction term oc f°£ becomes smaller and smaller. Suppose now we consider Eq.19).29) where 5t means "real parf.2 Comparison of Asymptotic Series with Convergent Series to Eq.2 yir (8. (8. Eq. These integral contributions are of order 1/y since the integral behaves like e~y and is multiplied by a factor e+y in front.28) we observe that the expansion (8.24). i. .
7r = .33)) would have • a maximally increasing exponential e+x . For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2." — E ^ ^ forarg. The corresponding phase lines are called Stokes rays. Dingle [70].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. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. is obtained for <f>(x) with arg x = —ir/2. (8.31).. *Cf.e. (8. i.e. y. since (8.3!) which is (8.2 OO rz=0 V .25) and integrate from 0 to oo. TimeIndependent Perturbation Theory If we insert here the expansion (8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. (8.138 CHAPTER 8.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. Eq. and • higher order terms of the associated series all have the same sign.26). . (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. (8. Looking at Eqs. B.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. (8. We have therefore found that XTT n=0 ^ ' and . Chapter 1. R. .28). Since (cf. e xZ ^ (n — £)! „ ^ ) . With a similar type of reasoning one can show that the same expansion.31) and (8.
35) 8. Magnus and F. in the sum of the right hand sides of Eqs. (8.2. i.37) Cf. 9 . i. These rules. though. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. p.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.34) the contributions 1.2 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.13. —1 cancel out. '"Cf. apply predominantly to asymptotic series of the solutions of second order differential equations.32) and (8. Dingle [70]. 5. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). Oberhettinger [181]. W. (6.e. pp. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . Chapter 1.g.8. for arg x = 7r/2.2 Comparison of Asymptotic Series with Convergent Series 139 It should be observed that the phase of the Stokes ray is the phase at which an exponential is not just increasing but maximally increasing. B.e. and so may not be universally applicable.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. Thus if we set x = XR + ixj. Equation (8. 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. R. 1/1 ipv{x) v1 vl (±ix). Eq. .2).rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. See also Tables of Special Functions. 91.
0 < argz <  (8.e. Dingle [70]. (8. v — — f — 1. R. We also observe that the late (large i) terms of the series in (8. TimeIndependent Perturbation Theory i.e.37). Thus D{J\x) (i. x~l/~1ipu(x). B. R.ei<v+VxvHv{x\ argx = £ . 8.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. .36) is invariant under these replacements. = 7r/2. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. (14).x 2 oo «*)  — E^fi=0 v . ti asymptotic series (8.e. 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. > > The equation (8.7. Eq. Chapters I (in particular p.40) § Cf.39) (as in the example of the error function). The continuation of (8. with a real proportionality factor. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign.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. (8. Dingle [70]. Chapter 1. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. i. 1 1 Cf. 9) and XXI. as can most easily be seen in the case of the Mathieu potential. Chapter 17. Hence Dil){x) = xv(j>v(x). i. We consider Dv (x). D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a.140 where^ CHAPTER 8.38) are of the type (2i)! i\ i\(2x Y 2 (2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series. B.
(8. (8. another half of the discontinuity appears on leaving the Stokes ray on the other side. with \x\ = (—x) for arg a. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. sm(iri/) = —a/3/2 or a/? = 2sm(irv).41) xuct)v{x) + a{x)~u~lipu{x). i.e. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real. In (8. beyond arg x — 7r/2. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8.e. < 7T. The value of Du (x) on the other side of the Stokes ray.44) + ^iap) = 0 . where in the first line on the right hand side the second term contains the real proportionality constant a/2.42) T ^(eiwu i.8.e.e.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). for xj = 0) and therefore possesses a Stokes discontinuity there. i. = ir. we must have at arg x = 7 in the dominant factor on the right of Eq. We therefore proceed to continue (8.40). (8. as yet an unknown real constant. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. < 37r/2. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. On reaching ir the part containing ^ ( x ) .41) the Stokes multiplier a is. = ir.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. (8. the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first).41) to arg a. Since Dv (x) is real when x is real.
Hence the left hand side must have the same property.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i . (8. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . a{y)P{y) = a(u . (8. (8.e.42) that a multiplies ipv(x).1)! TT.1). i.14).45) is therefore given by a{v)a(v .1) (8. Eq.v .45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.e. 8.42) we insert from Eq. (8.46) For a given function a(y) this equation determines /3(u).1) = ±a(y). (8. V i=0 \U0 v 2w .50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.47) Equation (8. (z)\(ziy.1) = 2sin7ri/.e. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.142 CHAPTER 8. We can see that the equation is satisfied by (3{u) = ±a(u . (3{v .48) We compare this with the reflection formula (8. TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. i. Dl \X) is (cf.1) (8.42)) for argx = ir (in the second line of Eq. (8. i.e.45) that af3 = 2sin(7r^)) D^\x) = (8. i.—^ ON • — ^ > + ^ ^ L . argx = vr. .48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .51) . (8.l)0(v .
u 1 y> {2i + v)\ 1 n 0.36)) in the domain 0 < arg x < IT.e.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. (8.„ _ix2^(2ivl)l e 2 > —.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.)(.e^ . Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ ..52) (argx = 7r).35).1 ) ( . —f.2.7).50a) because this enables us to define normalizable parabolic cylinder functions. It should be noted that in Eq. (8.) (8. For details we refer again to the book of Dingle.. (2x2)1 = COSTTU(X) v . i.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8. (8.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. (8.A / ± sin^Cx)"M* 2 V ^ .8..I / . i. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.1 ) ! 4~L i\{2x2Y i=0 V2^(x).*"ei°> f ( ^ _  _ L _ = (!)««(. the error . solutions of Eq. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.. Eq.7r). Chapter I.1.36). However. We observe that for v — n = 0. which vanish at x = ±oo..e..e. (8. ' ^ (i/l)!z! ^ . In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0. i. 8. ' (X) (cf. the second contribution vanishes and we obtain £>«(*) .
36) in the form (coefficient of — x2 here chosen to be ~. 0.144 CHAPTER 8. + — + ^ + • • • .37). x°. x—>±oo (8.54) (8. We write Eq. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation.36)) 0 where + x2Q(x)y = 0. \ + U (8.. S\.. how series of the type (8.60) into (8.59) and equating to zero the coefficients of powers of x2. we obtain equations from which the coefficients So.55) The large x asymptotic expansion can also be obtained from the differential equation. ^ + 2 .38) can be obtained. Eq. previously we had ^.. (8. • • • can be determined. S2. cf.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.e. and demonstrate.e.^ + 1 ) .60) 2 x x l Inserting (8. (8. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . Solving these equations we obtain = = 0. Instead of dealing with Eq (8. (8.36). Eq. x. i.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. (8. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. i. 5„=4 Sl=o. % = {. (8.54) we consider again the important equation of the harmonic oscillator.
2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. p.63) **W.36) have correspondingly e±x / 2 ) .37).8. i. etc.62) 4 \ 2 XiFi 2'2 . Oberhettinger [181]. . Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before).3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. We. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains.z) is a confluent hypergeometric T. T h e expansions (8. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  .19). n 22e"T JziL (8. therefore.61) (the dominant terms of (8.37) with (8. are asymptotic series in a variable x. Nonetheless we shall need asymptotic expansions like (8. » 1.91. and it is clear t h a t higher order terms follow accordingly. In the following we are mostly concerned with asymptotic series in some parameter.38). a z a a ( + 1) z2 which has the unit circle as its circle of convergence.38) in a variable for the solutions of approximated differential equations. function. (8. do not enter here into their derivation and simply quote the result: W i t h \x\ .+ X M W = Q. Magnus and F. 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+ . / i N . 4~l~2 ' i i _i. The derivation of the asymptotic series from integral representations (cf. (8.35) and (8.b.e. 2 X 2Wi.
66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2.67) Cf. .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. 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. Thus there is no contradiction^ between (a) and (b).+ — + ••• of a function f(x) for a given range of a r g x . T.1 a2 an a0 + — + ^ + .4x2 (8. top of p. 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. Watson [283]. 4 (8. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . whereas e~x '4 increases exponentially.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. 8. N. 349.146 CHAPTER 8. Whittaker and G.0. E. Thus the series expansion (a) in this case is multiplied by an increasing exponential and the asymptotic equality " ~ " means an exponentially decreasing contribution has been ignored.64) (c) for —7r/4 > a r g £ > —57r/4: Dv{x) e 4x !*£=£2 2x i/i + +• (I/1)! + xe 4 x (i/+ !)(„ + 2 ) 2x 2 " ' (8.
It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis".72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e.73) Vn + q/#> + e2T/i2) + • • • .5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . 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.68) The definition(8.g.70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (871) / dx%l)*m(x)lpn(x). (8. possesses a whole chapter on asymptotic expansions. The definition is due to Poincare (1866). Chapter I. In fact. lim xneW x—>oo = 0 (8. first published in 1902.68) is (effectively) simply a statement of the observations made at the beginning. (8. which is assumed to be small.e.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. . 8. we set E^En * > ^ n = En°1+eEnV = + e2EnV + .69) and the consequent nonunique definition of the expansion. in particular with regard to possible exponentially small contributions to the expansion for which e. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. i.8. (8. A critical discussion of the definition can be found in the book by Dingle [70].Sn{x))  = oo n—>oo (8. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.
TimeIndependent Perturbation Theory We insert these expansions into Eq. 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.4 0 ) M = / ^C(4 1} . J f°° dxrn E(^ 0) .78) into Eq.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.i4 0) )V4 1} = {EM . With the help of the orthonormality condition (8.74) (8. (8.V)i. °° i J or 0 = EWJdx{rnVil>n). Ho^+V^n = EW^+E^n. i. (879) We multiply this equation from the left by ip^ and integrate over x.77) The states \tpn) of the unperturbed problem span the Hilbert space"K. (8. We obtain then the following set of equations: Hrj. ^ ) = 0. i. so that we choose / • dxrpM^O.As a consequence the state vl/) is a vector in this space. (8. . we write in the position space representation ^ ^ E ^  (878) A contribution to ipn can be combined with the unperturbed part of the wave function \P.148 CHAPTER 8. The second equation can be rewritten as {Ho . i. (8.76) The first equation is that of the unperturbed problem.n = 40)V>n.e.e.e.w » . We insert the ansatz (8.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . ( ^ .75) HoW + vW = E^+E^+E^. (8.70) and equate on both sides the coefficients of the same powers of e.n.
^ n ) + 0(e 2 ).79). We multiply the equation by ipm. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n. (il)i.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.83) makes sense only if Ef] ^ EnV for all i + n.e. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. (8. i.4 0 ) M 2 ) = ( 4 1 } . (8. (8.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. multplying Eq.i ^ 0.V)^ Setting + 42tyn. with En .Vil>n) (880) In order to obtain ipn .76) which we rewrite as (H0 . We therefore consider Eq. we return to Eq.Viprt e (8. the coefficients a\ . (8. S ' ti^+Q^)• We observe that Eq.e.^(EW ~ V ) ^ + E& / J—oo dx^miln .8. (8. (883) Thus the procedure forbids the equality of any E\ '.m ^ n and integrate: oo /•oo / oo dxrmY. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ .82) *> = ^ + E . i.
Vlpn) E^ .V^j)(^j. ^ n ) j&iift + ^n .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi. (8. FVn) + e 2 ^ ( V > n .^ ) + $& w £ (^fc. Inserting the expression (8. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.B i 0 ) ) ( £ f .Vlpn] M _ P (0K 2 ^r . ipn ) = 0 i.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .£i°») 2 Hence +^£o (£«? .V4>n){lpn. (8.81) we obtain (2) _ _ {lt>m.e. n) £f) (8.V'>pTl (E^Eny V'fc (8. V ^ i ) X (Q) (Vi.Vlpn)(lpn. (8.^(8.Vlpn)E.88) (Q) Inserting Eq.79) •/—oo oo / Using (8.89) • • • where we guessed the term of 0(e 3 ). ~Ej E E.VV .^j)(V .^w .V^n) .(1) (i&0) .4 0) ^+ e 2 E (lpk. En0) (8.90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .150 CHAPTER 8.85) together with (8.2 (V> m . j.87) (i>k.
(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution. . 8. We can therefore rewrite tpn ': The perturbation theory described in Sec. The functions ipn(x) are eigenfunctions of some secondorder differential equation.e. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w.g..5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. (8.8.91) The coefficients Cj follow from the recurrence relations. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. Merzbacher[194]). These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. One would prefer to lump it together with the other contribution. An ugly feature of the expansion (8.. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations.90) is the term with only one summation J2 m the contribution of 0(e 2 ).Vlpn) = y^Ci(lPk.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).lpn+i) = ^ Cj5k.90) can be found in just about any book on quantum mechanics. We then obtain (lpk.n(x) = J2crtn+ii (8.7 is based on this procedure.89). i.
94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. is not uncommon. in a central force problem all have the same energy unless the system is placed in a magnetic field. i. all n. —I. (8. (8. so that En = 4 ° ) + eE™ + e2EP + ••• . . are allowed to be equal since otherwise the expansion becomes undefined. that of double degeneracy En = Em . Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. (H0 + eV)^n and (HQ + eV)Vm = Emym.97a) and (8. i.4 0 ) ) ^ } = (Em] .95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. however.94a) into Eq..152 CHAPTER 8.e. (8. m = I.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . Equality of eigenvalues.93b). . For example the magnetic states enumerated by the magnetic quantum number m. + c22^m).93a)..6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. (8. .94a). TimeIndependent Perturbation Theory 8.ipm) = 0 = (^m.We now insert Eqs. Equations (8. Equations (8. (8. .96) are effectively the known equations of ipnitpm and so yield no new information.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• .93a) (8.ipn). (8.97b) are . .93b) The lowest order eigenfunctions belonging to En.97b) with ipn y£ ipm.97a) (8.94b) with (ipn.95a) (8. degeneracy. Em can now be linear combinations of ipni'ipm. Em = 4 ° ) +eE$ + e2E$ + . (8. In order to deal with the problem of degeneracy we consider the simplest case. I — 1 .e. i.V)(c2l^n + ci 2 ^ TO ).70).96) (Ho . (8. (8.e.
^ ) ) = CU[EW .H0xl>$) .ci 2 (V„. V^n). (8.Vil>m)] V^m).6 Degenerate Perturbation Theory 153 inhomogeneous equations.(Vn. .C 21 (^ m . t f ) = a k 1 } 4 0 ) " Eg>aU = 0. ^ ) = ( V .ipm ) the equations (iPn.C22(Vn.m).H0il>W) = cn[EW .(^04°))41)) = EM^CuTpn (ipl. (^n. Vl/.H0^) . But (8.cnVipn and tyi. Equations (8./>£>) = C2l[E£> . ^ o ^ ) ) = 0.98a) Setting I = n. c2iV^n + c22V7pm).^ ( ^ . and ( ^ m .V^n). cn(ipm. V^n)\ .y>m. Vlf>m)] .EJ®(lPn.99) and ( ^ ^ O ^ ) " ^ ^ .J#»a&> = 0j since ii4i = En (degeneracy).99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .98b) + Culpm) + cuV^m). Wm. Similarly Wn. (^m.^ ) ( ^ m . £ aW^°Vi) . H0lP$) = C22[E$ ~ (Vm.m.8.(lPn. Then (^. (8.ipn n) = 0 = (ipm.£ < « = ag)^) . H0lpV) . we obtain with (ipn. Vi/>n)] . We now multiply these equations from the left by ip* and integrate.
154 and V (^m. Then / > i>" 2— = V ip Hence 0 = V ' V . It follows that ip and <j> are linearly dependent.1.. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. if)' (/>' — = —. At x = ±00 the constant is zero. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. 4.C22.Vlpn) CHAPTER 8.ci2. i> 4> Hence there is no degeneracy.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.3). Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . In this particular case therefore nondegenerate perturbation theory can be used (cf. (8. the degenerate perturbation theory must be used.Similarly Eqs.5). A onedimensional system. If one wants to perform a perturbation calculation in such a case.^Vm) 0.6.e.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. .e. in particular in parabolic coordinates (see Sec. (8.100a) determine the coefficients Cn. dx i. like that providing the Stark effect (proportional to rcosO). splits the problem into two independent onedimensional systems. i. Example 11. d>" E=^.Vrl>n) . as we saw. . Let ip and < be two eigenfunctions belonging to the same eigenvalue E.E)tp = 0.e. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). ip = c<p.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate.101) This secular equation determines the first order correction En • The equations (8. 9} ip.E™ (Vn. Example 8. \nip = ln(j> + c o n s t . 11.100b) determine C2i. It is then possible that a special perturbation. (V>V</>V>) = const. i. TimeIndependent Perturbation Theory (l/>m.^ ' V = — (•>!>'<t><t>'i>). the secular determinant. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. However.
8. in particular in applications to periodic potentials. Dingle [72] and H. Muller [216]. J. See also R.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. B. The subdivisions (8. with some restrictions on the function V.104) are chosen such that the equation Dcf) + (E0 .7 DingleMuller Perturbation Method 155 8. W. Eq. (8. (8. . V = V0 + v.e. the method has to be applied in each domain separately.* In various versions this method is used throughout this book.e)<t>. J. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. Thus the method is a systematic method of matched asymptotic expansions. of course. W. W.104) "This method was developed in R.V)(p = 0.105) (8. J. We consider an equation of the form Dcj) + (E .V0)<f> = 0 (8.V0)<t> = (v. (8. H. i. cf.106) (8. It is immaterial here whether D does or does not contain a first order derivative — in fact. Dingle and H. and the solutions then have to be matched in their regions of overlap. and such that one can even obtain a recurrence relation for the coefficients of the expansion. anharmonic oscillator potentials and screened Coulomb potentials. Applications of the method are. T For instance spheroidal functions can be studied with this method. We assume that V can be expressed as the sum of two terms. the considerations given below would be valid even if D contained only the derivative of the first order. Muller [73]. (8.103).102) can be written D<f> + (EQ .102) where D is a second order differential operator and E a parameter. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. a potential.t Since in general a solution is valid only in a limited region of the variable. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. Muller [210].103) such that if E is written correspondingly E = E0 + e. also possible in areas not of immediate relevance here. B.
Vo)(j)m+s = (Em .108) with coefficients (m.110) If we write the next contribution <f>W to <p(°> in the form m+st (8. the function 4>(0) = <t>m (8.V0)<f>^ .m + s)4>m+s s (8.m + s)" in order to relate them to "steps" from m to m + s.e ) ^ = 2 ( m .113) .E0\ (the latter in a restricted domain).111) s then (see below) 0(1 = E (m.m + .+s ' (8.e\ < \V0 . TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.102) with E ~ Em. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . m + s) which are determined by the recursion formulae. v = {D+EmVo)<t>m. and so v (8. (8.109) D(j)m+s + (Em .VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . 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. (8.107) represents a first approximation to the solution (f> of Eq. where m is an additional integral or nearintegral parameter (depending on boundary conditions). m + s)0m+s.Em+s)(/)m+s. the equation Dcf>m+S + (Em+S . Then if ]e < \E\ and \v .) This result is obtained as follows. We write the coefficients "(m. s an integer: (ve)(pm = 'Y^{fn. since D<f>m + (Em .V0)4>m = 0.156 CHAPTER 8.V0)<f>m+a = 0. Considering (frm+s we have.
TP \(T? . . (8. . (8.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.m) E J^ ( m ~ Em+S) ' • fa.116) m ^ v (m.m + s)(m + s.m + s)(m + s.8.m + r)(rn + r. .m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .n\ ^119) + .m) + > ^ .Em+S) = 0 (second approximation). £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.118).102) to order (1) of the perturbation (v — e). The factors (Em — Em+S) in the denominators result from the right hand side of Eq. V ^ V ^ {rn.T? S The first expansion determines the solution 4> (apart from an overall normalization constant).113) must vanish. . .m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m. (8.Q (m.m + s)(m + s.7 DingleMiiller Perturbation Method and so (m.m) 1^1^ IW .119) follow a definite pattern and can therefore be constructed in a systematic way. (8. (8.117) „ .m) — —. Eq. the coefficient of the term (m.119) is effectively the secular equation).m) — 0 (first approximation). and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.„N This procedure can be repeated indefinitely.110) and are therefore related to the inverse of a .114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq.^ (Em .m + s)(m + s. /01_.. Thus (m.e. (8. (8. We observe that the coeffcients of expansions (8. We then find altogether ^  l^m  Em+S) + y^y^ SJ.m)4>m in (8.
Other applications can be found in the literature. An explicit application of the method (not including matching) seems suitable at this point.1)( 2 + 64J3] — .2). . This means the solution is not normalized.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. 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.e. S. An additional aspect of the method is the full exploitation of the symmetries of the differential equation.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. (8. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . W. 1 . . as will be seen in later chapters. The eigenvalue expansion obtained is an asymptotic expansion. as already mentioned after Eq.120) n4 .[q(llq2 + 1) + 2(33g 2 . n = 0 . Miiller [211]. § H. for instance. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. this is — so to speak — the price paid for obtaining a clear systematics which is essential. . and is typical of a large number of cases. The result is the following expansion. This can in fact be done and will be dealt with later.[ 4 ( 8 5 g 4 + 2q2 . .423) + Z(2720 9 3 .I2q + 64) + 256Z 3 (41 g .158 CHAPTER 8. Kaushal [146]. The following Example 8.* Example 8. 2 .6q + 1)1 + 24(5? . as will be seen later. i. i. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. TimeIndependent Perturbation Theory Green's function. since it will be used extensively in later chapters.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. does not occur in any of the later contributions. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. for instance. ^The potential A r 2 / ( 1 + gr2).38). for the investigation of the behaviour of the late terms of the perturbation series. has been treated by R.9) + 4096J4] + 0 ( l / g 3 ) . in which q = in + 3. (f>m in the above.e. (8. J.
Therefore in the general case we may set k2 + g2 = ga(2l + q) .z2 is a normalizable function if a = —n for n = 0 . (fc2 + <?2) 4ga and b = I \—. Sees.l)ip(a .123) The first approximation (note the last expression as convenient notation) ^ . „2 J„2JL. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.121).8. (8. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . (8._ „!2 ^ ( " « 2 r r 2 > .122) where A is of order 1/g. The solution i.0(z) = zl+1ez2^fa.b. \2 .1). multiply the equation by onehalf and set h = —a/g. a . .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. 11.2a2 A. a)ip(a) + (a. 2 The solution XQ(S) = $ ( a .a + l)if>(a + 1) + (a. (8.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2. 1 1(1 + 1) 1 .6 and 16. 3\ 2 V ' 2. (. 2 . S) is a confluent hypergeometric function. .. Setting q = in + 3 implies k2 + g2 ga(2l + q).123) the contribution fl<°> Ah + .b.V ! 4 I z S *! V2 ipq(z). 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. . .2 '(l + l) . 1 . We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. . . i/> with Dq = d2 . We now insert this equation into Eq.2) where 1/ . b.
2.a + 2]i P .a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". In fact.3). fa.^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously. of course. (8.a—lli. .a]i = . a + 1) = a = In general CHAPTER 8. i.Z. one could now use the RayleighSchrodinger method and rederive the result (8. a ] i + L ' ^ J 1 [ a + l .a]1=0. a]i + h fa. _ J [a .b = (g + 3) . ( 5 m ( a . This equation shows that a term fiip(a + j) on the right hand side of Eq. a + r) = 0 for r > m. 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.e.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except.160 where (a. TimeIndependent Perturbation Theory (g .2a = I + q.120). a + j)i/>(a + m). „ . (a. a) = 1. when j = 0. If we now evaluate the various coefficients in the last expansion we obtain the result (8. a ] i + 0(/i3) [a. the latter determining to that order the eigenvalue.l . a + r + l)(a + r + l .120). 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. (a. a ] x + L _ 2 J 1 [ « .a — 211 . a + 111 r [a. As a check on the usefulness of the perturbation method employed here. So (a. . a . a + r — l)(a + r — l . 0 = h[a.S m (a. a + i) = 0 for i ^ 0. a) = b .j^0 ' + jU+1 J ^(a + j) with h[a. a + r) + S m _ i ( a . The coefficients S m ( a . one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. Now we make the following important observation that Dqip(a + j) = jip{a + j). where [a.1) = a . a + r) with the boundary conditions Sn(a. Hence the next order contribution to tp^ becomes oo i+2 r [a a . a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . a + i)(a + r.a]2+L ^ J1[a + 2 . a + r) satisfy the following recurrence relation: Smi(a. o + r) + S m _ i ( a .i ^(i)=£V+i i=0 J2 j = (i+2).a + j}i+1^(a + j). fa.
161 . proceeds along similar lines.Chapter 9 The Density Matrix and Polarization Phenomena 9. Regrettably this has to be left out here in view of lack of space. We therefore begin with the appropriate classical considerations with a view to generalization. light. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. i. the generalization to many particles. 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. and we shall then see that these lead to analogous results as in Chapter 2. and this means electrodynamics. as we shall see in Chapter 27. 9.e. although not exactly straightforward. In continuation of our earlier search for an approach to quantum mechanics as close as possible to classical mechanics by looking for the generalizability of classical mechanics — as already attempted in Chapter 2 — it is reasonable to consider also classical systems with a wavelike nature.2 Reconsideration of Electrodynamics In this chapter we consider some aspects of classical electrodynamics with a view to their generalization in the direction of quantum mechanics.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.
5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. This is the equation of an ellipse as indicated in Fig.i). 9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized. (9. i.1. t) given by A(r. (9.4) (9. (9.1 The vector potential ellipse.1) Fig. A 0 = A0R + iA0I.e.2) (9. In general the constant Ao is complex. It follows therefore that A(r. 9. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.t) = 3ftA. The Density Matrix and Polarization Phenomena potential A(r.162 CHAPTER 9.6) . t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.3) (9. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized. t) = A 0J R cos(cji) + A 0 / sin(u.
A 2 . AW = £ fc=l.7) We consider now measurements with polarized light. From the vectors A(°).8). (AW.y)) components A i . (9.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. determined with the help of a polarizer. the vector potential A. the connection must be linear. the direction of polarization. (9. i.A2 is the sum of the moduli. I=<l2x2>:=E40)*l*40)i.11) 'This means we consider the polarization. (9. The intensity 7 of a light wave with amplitude coefficients A\' .t in which k and ui remain unchanged. In classical electrodynamics.e.A« we can construct only the following scalar products with this invariance: (A(°\A(°)). which leads to an outgoing wave which we give the label "1". the fields E and B are observables. Here we are interested in observable properties of light waves — like.A«).A(°)).2 *i*4 0) (98) with coefficients F^. for instance. (9. See for instance A. Observable quantities can only be those which are invariant under translations and rotations. A2. 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. p. not.g. (AW. Then their connection is given by Eq. 17. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2.2) the vector potential A is completely determined by its two (say (x.e. ie. Rae [234].AW).2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.k (910) We give the initial wave the label "0" and subject this to an experiment.e. however. .9. the intensity of a wave.A i = 0 = k . as e. (A(°). i. we recall. a calcite crystal or a film of nitrocellulose. We set therefore A 0 = Ai + A 2 . represented by a filter "F". i. with k .
13). the connection between the initial polarization and the final polarization.12) i. (A(°). (9. pki = AkA*.13): j2^FikAk = Y^AiF*kAi = Y.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). (9.13) the quantity F must be an hermitian matrix and thus representative of an observable. it must be real and hence (F) = <F>*.k i (9. if (F) is an observable quantity. (9. (9. Then according to Eq.164 CHAPTER 9.e.k l f c 4 ° \ (9.e. We use definition (9. (9. so that (F) can be looked at as the expectation value of an observable F.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments. i.k (9.15) in order to rewrite Eq. We write the observable quantity as (F):=Y. Thus (F) = J2FikPki = ^(Fp)a. (F)=Tr(Fp).A( 1 )) = ^ ^ F i. Thus.A*kF£A* = E and hence F = Fl A p i tkAk. (9. i.16) i. i.e.A*FikAk. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. To this end we observe that as a consequence of Eq.17) .
like that of a bulb. i. that of a pure state.15) with A\ = a. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions.e. . i.* For a lightwave travelling in the direction of z. A2 = b as </*) . 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. "incoherently.21b) This state is still a pure state. is given by a single polarized wave.18) The vector A describes a particular ray of light. The socalled pure case. In the present case of a polarized beam of light we have two possible. mutually orthogonal polarization directions x and y described by the matrices (see also below) * . (9.(AAt) = ( £ £ ) . For a 45° polarized wave we have 1 .e.( J o ) . This admixture is described by introducing matrices p.e.9. i. (9. we have (9 19) ' In this case there is no preferential direction. 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. 165 (9. b = 1 we obtain the expressions (9. But light. The density matrix for the pure state (9.22) For a — 1.21a) follows from Eq. <»^> (9.19).d "»=(o ! ) " In the case of an admixture with equal portions of 50%.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). is in general unpolarized. b = 0 and o = 0.
23) 1 . We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M .15) Pik — A i A k . Equation (9. (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". 1 = . (9. l l\ " i °\ ) ' ^925^ 9. Pi35° = \ \ 2 ? )• (9. (9. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics. in which the intensity I = Tr(p) is not conserved (e.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.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. as a consequence of absorption).166 and CHAPTER 9.™ V 0 n kmAm / ^^jPim^mk' .8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F.g.Z^ 3. Then according to Eq.8) A(1)=E^40)> k R = R(<*). b = —=. We set in analogy to Eq. (9.
q.e. (9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq. that the initial intensity Jo is equal to the final state intensity I\. in the state functional) in analogy to the time dependence of W(p. (9.31). In the case of Eq. the observable F remaining unchanged. the dependence on a is contained in F' = R^(a)FR{a).29) where (i. in other words without creation or annihilation of systems. i.33) i. In the case of Eq. (9. (9. the observable is transformed.9. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . 9.30) Ti{FRp{®I$) = Tr(F.32) (9.31) (9.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.pW). here this contains the dependence on a. The corresponding condition here is that no absorption of the wave takes place. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)). The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations.28) (9.4 The Liouville Equation 167 pW = RpMRl (9.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. (9. and p(0' remains unchanged.e. i.34) .e.e. on the other hand.t) in Chapter 2 in what we described as the Schrodinger picture there. (9.
p].36) (9. Eq.40) Equations (9. We set therefore M^:=iH.B}:=i[A.B]. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.p(°\ao)]da.46) suggests to define Poisson brackets of matrices by the correspondence {A. (9.41) We see therefore that the 2 x 2 matrices entering the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics in Chapter 2. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. .33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H. CHAPTER 9.35) Thus the matrix R has to be unitary. j£ = i[H.e. In view of Eq. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. = Tr(R^Rp^). (9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.e.M . In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9.168 i. (9. (9.39) and (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.39) If the dependence on a is not contained in p^1' but in F' (cf. Comparison of these equations with Eqs.F}. (9.40) can be considered as quasiequations of motion.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H. (9.2) has to be taken into account. (2. the matrix M is antiHermitian: Mf = . (9.37) (9. i.21) and (2.
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. we establish their Heisenberg equations of motion. Thus in particular we establish the uncertainty relation for observables in general. To insure that the 169 . 10. Heisenberg and interaction pictures. the Schrodinger.Chapter 10 Quantum Theory: The General Formalism 10. (10.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. and consider timedependent perturbation theory. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. 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).
we must have (u\u) = 1. we assume this now. Naturally one can also write (10.y. such as for instance of the angular momentum operators Li. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. 10. Here.* a conspicuous exception being for instance a spin system.g. Pi • Pi.2. Such a correspondence with classical mechanics is not always possible. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. 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. E.1 U n c e r t a i n t y relation for observables A. classically qiPi = PiQi.q). which clearly does not hold in the case of operators.Pj] = 0. The phase space coordinates qi. where i = 1. and assume that A = A(p.3 represent the three Cartesian coordinate directions of x. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. y.Pi are called fundamental observables. \pi.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.2) (F(A)) = {u\u) <"lf. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i.170 CHAPTER 10. q). [qi.4) *Even in those cases where this is possible the operator correspondence may not be unique. however.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A.Pj] = ihSlj. .Qj] = 0.2. (10. (10.B] = ih.
. B}= ih. AB defined by the mean square deviation AC=((C2)(C)2)1/2. = \\A\u)\\2\\B\u)\\2 (10.12) the Schwarz inequality (cf. Thus AA = AA = ( i 2 ) V 2 . 111 (10. Ai = = = ((A2) . (10.9) ((A2) + (A)2 ..1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.9a).2(A)2 .B] = ABBA > ^h.7b) = = [A. Since A = A— (A). Then [A.7a) (B) = 0.8) and.0)1/2 = AA.12) where \u) 6 "K is a ketvector representing the state of the system.2 States and Observables then their uncertainties AA.B\u) G IK as vectors) we obtain (for an application see Example 10. (4. (10. Applying to the expression (10. so that (A) = (A) .13) AB = AB = (B2)1/2.(A)2)1'2 (1 °= a) ((A2 + (A)2 . using Eq. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). C = A. (10. (3. (10. (10.{A)2)1'2 {{A2) .2A(A) + (A)2.B.5)) (for A\u).(A) = 0. (A{A))(B{B)){B{B))(A(A)) (10. (10. (10.10.11) (A2) = (A2) . we have A2 = A2 .2A(A)}  {A)2)1'2 (10.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A)..10) .6) B:=B(B).7b). Eqs..(A)2 = (AA)2.
A\u)=cB\u). Zi (10. (10. (10. R= (AB + BA).172 CHAPTER 10. for a complex constant c.15) as (u\AB\u) = R + iI.14) Hence (u\AB\u) = /^(AB where. (AB + BA) This means (AB + BA) is real. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. Zi + BA)\ + ^ih.4). and hence for condition (b) 0 = = tSee Sec. Zi Zi (10.17) as had to be shown.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. (10. (10. From this we obtain (uiJBu) 2 = i22 + / 2 .3. (10. since A and B are hermitian.16) I = h. (10.18) >R2 + I2. We can therefore rewrite Eq. and (b) when (AB + BA) = 0. .e. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u).13) (AA)2{AB)2 i. AAAB>I=h. The equaltosign applies in the case when (a) it applies in the Schwarz inequality. and with Eq. 4. (10. The condition (a) is satisfied when.
i relation holds: (ALi) 2 (AL.e.y. (Ly)=0. For states \u). i. such that Lz is "sharp". (A{A))\u)=i(Zc)(B(B))\u). ALy = (u\L±\u) ^ 0. 9ftc = 0. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li.e.10. in view of Eq. implying {Lx)=0. (ALy)2{ALz)2 > ~h2(Lx)2.) 2 > = x. Example 10. Thus in this case Lx.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. = 0. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. (10. {u\LiLj\u) = R + iQ. i. (Lz)2^0. i. q) = ih.Lj] = (LiLj + LjLi) + ih£ijkLk. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. (Ly)2=0. i.e. m\lm)). 9 = h{u\eiikLk\u).1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. (10. for which one component of L is "sharp"? Solution: Applying Eq.7a). (Lx±iLy) ^ 0. h2(Lz)2.20) or. + LjLi\u). (ALZ)2(ALX)2 > \h2 = (Ly)2. Ly are not "sharp".3 OneDimensional Systems 173 Thus c = — c*.13) to operators Li — Li. The observables like angular momentum are functions of p and q. ALX 10. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. the following uncertainty h2{Lk)2. .e.z. Separating the product into symmetric and antisymmetric parts. For these we have in the onedimensional case the relation [p. = 5R2 + G 2 > 9 2 = hi2(Lk)2. What does the relation imply for states u) = \lm).
21b) now follow with the assumption that A(q. i.C\. and (c) the spectrum is not degenerate.21a) Equations (10. This is why the operator p requires a representation in the space of eigenvectors of q.34d)) [A.qn] = ih^(qn).B]C + B[A. A — A(p.p] = ^ihp ~ Proceeding in this way we obtain [q. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).p) can be expanded as a power series in q and p.p)]=ih^^and \p.A(q.p] = ih we obtain therefore [q. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). i. Then (see below) we can derive for A the following commutator equations [q. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable.e. (b) its spectrum is necessarily continuous.q). (10. (3.23) = ihg(q2) (10.21b) (10. Using [q.A(q.pn]=ih^(pn). q]q + # > v] = ~2ihq and \p.21a) and (10. Next we have a closer look at the operator q. .p}p + p[q. Q2] = [p.i h ^ ^ . We have (cf. as remarked earlier.22) d ih—(p2).BC] = [A.p)] = .p2] = [q. Similarly we obtain \p. We obtain these equations as follows.174 CHAPTER 10. Eq.e. they do not have a common system of eigenvectors. (10.
For the cases (b) and (c) we consider the operator U(a — pipa/h (10.1 The delta function for K — oo.f^ ^ J—oo dke ikx —ikx' (10. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. = eipa/h = Uia)'1 = U{a). > We begin with (a).26) it Here p is an observable.21a) A by U.28) i. U] = ih^= aU. (10. 10. and a a cnumber.10.24) lim SK(x — x') — S(x — x'). we obtain [q.25) It is shown in Fig. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1. (10. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization).e. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.e.27) (10. i. U is unitary. 10.3 OneDimensional Systems 6K(x) 175 Fig. . Replacing in Eq. an operator.1 how the delta function arises in the limit K —> oo.
^Observe that {iji\A{p. 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 (—00.176 i.e. q any other observable F(p. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). Evidently every eigenvalue has only one eigenvector. (10. 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. Eq. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). Since this holds for any arbitrary value of a in (—00. (10.q)U^U\ilj).30) (10. (10.e.3. Thus all these values belong to the spectrum of the operator q. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.q)\i>) = (i>\U^UA(p. that with U(a) = e~ipa/h = U. i. we can see that also p possesses only a continuous spectrum (from —00 to 00).00). b a cnumber). WU = 1. CHAPTER 10.29) This means: U\x) is eigenvector of q with eigenvalue x + a. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). q) can be constructed representing a dynamical variable. or qU(a)\x') = (x' + a)U(a)\x'). the spectrum is not degenerate.31) .00). which means that the spectrum of this operator is a continuum. cf.1 T h e translation operator U(a) We saw above. or qU\x) = U{q + a)\x) = (x + a)U\x).29). 10.
x" . (x'\p\x").x") .e.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x").36) i.10. \c{a. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' . i. In the expression of Eq. offdiagonal elelemts of the {x}representation of the momentum operator. Comparing Eqs.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . for instance.35) (10.x" .31) and (10. the operator U(a) acts to shift the value x' by the amount a. (10.34) {x"\U\a)U{a)\x') c*(a. We choose the phase such that U(a)\0) = \a). The operator is therefore called translation operator.3 OneDimensional Systems 177 Moreover. \x') = U(x')\0).e) ~ 5{x' .a).x') = = i.x") U{x'\p\x" % pz. (10.x')\ = l. i.e.37) With the help of the latter expression we can calculate. we have q\x' + a) = (x' + a)\x' + a). (10.e. since q\x') — x'\x'). (10.x" .39) .38) h 6(x' .e) (x \p\x ) = . U(a)\x') = c\x' + a).33) (10.e.x")c(a.5(x> .x')5(x" = (x" + a\c*c\x' + a) x').32) we see that \x' + a)ocU(a)\x')./ (10. (10.32) Since U is unitary.e. it follows that 5{x" . (10. i. (10. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).
44) (10. (10. i. .35). 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)). H\uE(to))=E\uE(to)). a conservative system. .f dpipeipx.t0) = HU(t.27) and (5.45) (10.x") = (x'\p\x")*.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. (io.tQ) = eiH(tt<Mh.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . it follows that 5'(x) is an odd function of x.° ° d(p) (ip) e"« = S'(x). Differentiating U(t.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. to) with respect to t we obtain ihjtU(t. Quantum Theory: The General Formalism We mention in passing that.t0). In this case the classical Hamilton function does not depend explicitly on t.L fdpeipx. The operator U(t.4. i. 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). 6'{x) = !.e. so that U{t.to) exists which is such that^ m)) = u(t.4i) We assume first of all.178 CHAPTER 10. (5. 5. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.e. (10. since 5(x) = .43) (10. See Eqs.to)m0)).to) is called the time development operator. 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.e. This is in conformity with our earlier postulate in Sec.40) 10.
Ufa.t) and therefore [U(t.t) = l^eH{t).10.41) we obtain ^(t 2 )> = = J7(t2.t).tiM*i. = U(t0. (10.4 Equations of Motion 179 We now demonstrate that Eq. From Eq.to) (differentiating we obtain immediately Eq.47) e. t0) (10. But obviously (cf.e.t0)\^(t0)). Hence U(t. The operator H is d efined by this relation.z / dt'H(t')U(t'.t0)]1 Let us set for small values of e: U(t + e. (10.48)).*l)^(*l.48): [/(Mo) = 1 .to) can be expressed as an integral as the solution of Eq.t0) or ih—U{t.t) = 1.to) = 1.44)) we have U(t.t0) [U(t + e. t0) = E/(t 2 .t)l]U(t.to).t 0 )^(*0)> U(t2.49) . (10.t) = l. (10. (10. (10.45) has very general validity. Then with U(t + e. Eq.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.*l)^(*l)> = I7(*2. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.48) with U(to. t0) = H(t)U{t. Evidently U(t. (10. i.t)U(t.t0)U(t0.
In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.e.t0)\ii>s{to)) (10.30).53) . the instant at which a measurement is performed on the system. >(*)> = i. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. 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.e. Now let \tp(t)) be the state of the sytem at time t. The operator U(t. Eq.41).52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). (io.51) Since </. the Schrodinger equation ju(t. Another such description is the Heisenberg picture.(£ + dt)) = U(t + dt. the predictions of different descriptions are identical. (5. (10. 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. requires the identification H(t) = H= Hamilton operator. it follows that U(t + dt. The eigenvectors of these observables are also constant in time. 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. Since the only measurable quantities are moduli of scalar products. i. Then \Mt)) = u(t.5o) ihjtm)) = Hm)). Comparison with the equation postulated earlier. and these remain unchanged under unitary transformations.t) = l % ^Hdt) \il>(t)) Hdt (10. (10. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.180 CHAPTER 10.t0) m0)). t)\t/>(t)) =U~ and since H is hermitian.
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. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.55) and using Eq.59) This equation is called Heisenberg equation of motion. t0)AsU(t.(tf HU)(rfASU) (10. i.HH]. we have to take into account the contribution dAs/dt. For As(q. Thus. 181 (10.58) HHAH (U*ASU)(U^HU) [AH.HH]. . Differentiating Eq. we consider only cases without explicit time dependence of As.t0)\ips(t)) = \ips(to)) = constant in time. if ^ = 0. As(q. = so that ihjtAH or ihjtAH = = [AH.p. i. {AH. even if As is not timedependent.i). as a rule.48).55) Thus AH is explicitly timedependent.p) in the Schrodinger picture dAs/dt = 0. subject to a continuous change. It does not always have to be the case that As does not contain an explicit time dependence. for instance in cases of nonequilibrium or if part of a system disappears through absorption. (10.4 Equations of Motion and \^H) = U\t. (10. in order to obtain the Heisenberg picture (description).H]U + iW]^U. Here H is the Hamilton operator in the Schrodinger picture.e. (10.H]U = = = U]ASHU AHHH — U^HASU .HH]+ihU^U. In the Heisenberg picture it is HH = U]HU.e.57) and rf[As. (10. a scalar product) remains unchanged: AH(t) = U\t.With explicit time dependence.e.54) Observables transform correspondingly with a similarity transformation. (10. t0). so that a matrix element (i. but here. However the associated observables are timedependent.10.56) U^[As.
HH}=0. We see that formally one obtains Hamilton's equations of classical mechanics. in the case of the fundamental dynamical variables qi. Orfanopoulos [224]. We also observe that the momentum pi is conserved when \pi. For an extensive discussion see B. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. The "pictures" just explained — with regard to the time development of a quantum system — are not to be confused with the representations of the theory.Pi): and where the expressions on the right follow from Eqs. (10. . Solution: We have for an observable A (A) = (i>H\AH\TpH). Finally we add a comment concerning conserved quantities.13) to derive the uncertainty relation for energy and time.Pi we have in the Heisenberg picture (replacing in the above AH by qi.21b). i. In the treatment of the pictures we used unitary transformations of operators (and vectors). This holds also in the case of the Schrodinger picture. In our treatment of representations. [CH.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. An observable CH is called a conserved quantity. if ffCH = 0.61) Thus conserved quantities commute with the Hamilton operator. provided the Poisson brackets in the latter are replaced by commutators.e.21a) and (10. we used in essence unitary transformations of matrices. In the Heisenberg picture essential properties of a system can frequently be recognized more easily in relation to their counterparts in classical mechanics — in both cases the time development of a system is given by the time dependence of the dynamical variables.H] = 0. Example 10. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. A. In particular.182 CHAPTER 10. We end with a word of caution. (10. In many cases the Schrodinger picture is more amenable for explicit calculations. since in the transition from one to the other the commutator remains unchanged.
in a different form: The eigenvalues of an observable are called "good" if [A. since \IPH) is timeindependent.68) To put it The eigenvalues a of A (i. (10.p) we then have 0=^M) = U[A.e.2 > h. > (10. (10. If r is the smallest such characteristic interval of time. (10.e. at ^±AH>h.15): (AAf(AH)2 Using Eq. a position or momentum observation is independent of t. is displaced by A A in the interval A T = TA.4 Equations of Motion Then. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t.66) This means. dt in. We construct the following vectors using Eq. if we set AH = AE we obtain TAAE (10. 2 d{A)/dt " and.)\2 ( = \{i. i. i.62) and dAs/dt > \WAH\i. the "centre of mass" of the statistical distribution of A.64) = 0. (A). this is practically the same as at time t for times t with \t — t'\ < T.7a).e. .H]) = 0. and TA O C (10. (10. we can replace ([AH.e. H] = 0. (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 ) . Let \ip) = \"4>H) represent the state of the system.62) The Hamiltonian H does not depend on time.10. (10. 183 dV ' \ dt >"•"»» + ( £ ) • (10.Such a characteristic time interval can be defined for any dynamical variable. AW = {A(A»V>.63) so that with Eqs.13) and (10.\HAM\2) \(nA. H\tl>) = {H(H))\f). This means that for a physical variable A(q. i. of A\<t>) = a<£)) are then called "good quantum numbers".H}\f)\2.65) and rA AA d{A)/dt' .HH]) Hence we obtain by ih(A). then if a measurement is made at time t'.67) 1 d(A)/dt is oo and AE = 0.
69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.Eo(x)eiEot/he'^2h9(t).70) we obtain: oo / oo TPEO (x.tyEt/hdt = = = / J—oo dt J dEJrl>E.H ) ( £ .69) and (10. First we obtain from Eq. Recall for instance radioactive decay. the wave function ipEo(x.t) — ipE(x. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.(x)fEo(E')e«EE'Wh 2irh f dE'^E. This means that in the case of decaying particles. i ) = J dE'^El^)eiEltlhfEo{E'). i. This means.70) we can determine the function fEo{E) integration with respect to t. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10.72) = V'EoW 7/2h + 0i(EE )/h  oo . for instance. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime"..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 . The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.71) From Eq. (10. (x)/ E o (E')5(E . Quantum Theory: The General Formalism 10. states with sharp energy (AE = 0).)exp(iEt/h)). (10.69): oo poo p / oo *PEo(x.184 CHAPTER 10. as is observed. (10.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x .E1) 2irhfEo{E)4>E(x). /oo (10. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. (10.70) by From Eqs.e. have around EQ the form ipEofct) = i. the particle density diminishes in the course of time (t > 0). the wave function ^ ( x . But the number of radioactive (thus excited) nuclei decreases exponentially. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. (10.
and is called "interaction picture" or "Dime picture". VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. We saw previously that if the state of a system is known at time to. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. We obtained the result (10.to) = HU(t.75) (10.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). (10. i. i.76) **For specific applications see e. (10. This description contains effectively parts of the other two pictures and is particularly useful when Eq. . i. (10.t0). i.e. (10. Considered as a function of E.105) and (20. 7 > 0.e. every unitary transformation of states and operators defines a possible "picture").6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). (14. They are called "resonance states". of course. Let U^°\t.e. (C/(0) unitary).toM{to)). 185 i.10.11).41).g. (10. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ.74) Thus the solution of this equation is the main problem. W)) = U(t.e.e. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. Eqs. £/ = £/W(Mo)^'(Mo). We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. that** EQ. ^(*o))> then \xp(t)) for t > t0 follows from Eq.45). where U is the solution of Eq. to) be an approximate but unitary solution of this equation. The state with lifetime r — h/'y is not a discrete state.45) can be solved only approximately (in actual fact. 10.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. ihU(t. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state.
dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).e. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. i.186 Then CHAPTER 10. = HU<®U'.h^U> = at ( io.t0) 0 = l.78) (10.e. [/' has only a weak ^dependence (i. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. Quantum Theory: The General Formalism ih^(uWu') i.e. with Eq.79) ^'~°i. (10. varies only slowly with t).77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. with Hamiltonian H = H^ + H'.e. £/(°)£/(°)t = 1. In addition we have (10. i.83) ih^U^ ishermitian.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. (10. Now let H^(t).e. H(°Ht) = ih^uW.82) (10.e. (10.e. .81) But [/(°) is unitary. i.
(10. (10. (10.86) at = U^HU^U'. (10.78) at (10 85) (10. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).e.89) where ^s(i)) and As are state and observable in the Schrodinger picture. (10.85) = where we used Eq. (10.10.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. H = H{®{t) + H'. lfe(*)}.78).85) multiplied by U') on the right of Eq.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10. (10. (10. (10. i.84) U^ where H' is also hermitian.80). and obtain from Eq.92) . 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^. U<®\Mt)) = \Mt)). (10.U^HU^U' H'fi'.90) = U^H'\^s(t)). We also set H^U^H'UW. U^H'U^U' (10.
94) _Hf)Al 1 + lhdAl+AlHf).e.89).89)^0. (10.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. comments after Eq. so that with Eq. we differentiate with respect to time the operator defined by Eq. H'j = UWH'UW. (with ihjtAI(t) = [AI. In order to obtain the equation of motion of an interaction picture observable Ai.e.188 CHAPTER 10. cf. dAs/dt=0) .e.60b) with iff0) instead of H.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). (10.90) and replacing the left hand side of this equation by the right hand side of Eq. (10. a perturbation so that H ~ H(°)). (10.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. (10. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). Since H'j is assumed to be small (i. i. (10.60a) or (10. ot i. ^ (0) l^(*)>. near zero and \tpi(t)) is almost constant in time. the vector varies with time (different from the Heisenberg picture).93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. we have i. (10.55)) at + imm<Ms_u(o) at dt _umH(. (10. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. i.H\0)}.94) Then we have (normally with dAs/dt dt (io=8o) — 0.e.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ . A^t) = U^AsUi0).e. i.e.
where according to (10. H0\ipn) = En\(fn). i. (10. i. (10.e. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. (10.100) (10. ^ \<Pn){Vn\ = 1n (10. i.i0) = e<Ho(tto)/ft.e.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. ih±\Mt))=Hi{t)\Mt)). (10.93).94) Hi = uW(t.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture.e. H' = H'(t).t0)H'uW(t.97) The equation to solve is Eq. {ipn\<Pm) = Snm. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).98). (10. (10. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.80) as E/(°)(t. We have the Hamiltonian H = H0 + H'.102) We assume that before the perturbation H' is switched on at time to.7 TimeDependent Perturbation Theory 189 10. We assume again that the spectrum and the eigenvectors of Ho are known. We obtain the operator U^ from Eq. Since .99) (1098) Instead of the original Schrodinger equation we now solve Eq.87) and (10.t0).10.
the time dependence of \ips } can be separated as for stationary states. (10.107) Jto With this we obtain from Eq. we can write l 4 0 ) ) m = elHot/hWm) =e . We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the timedependent perturbation depend on time.e.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). (10. (10.104 ) Here we insert Eq. Hence we set ^> = £°«(*)l4 0 ) >n. Quantum Theory: The General Formalism HQ is independent of time.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). (10. (10. (10.^ V ™ ) . i. 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'.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s).105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10.102) and truncate the series after the first perturbation contribution.190 CHAPTER 10.103) for m — n and use Eq. 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. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.108) .106) We insert this expression into Eq.
for instance. but that otherwise the perturbation is timeindependent. or if by emission of a photon with continuously variable momentum an atom passes from one state into another. A further example is provided by adecay. (10. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions. < KO?t . Thus we set H'(t) = H'0(t).110) With this we obtain from Eq. We assume that the perturbation is switched on at some time to.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^. if in the scattering of a particle off some target its momentum p changes to p'.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En .109) 10.112) for a^0.Em)t' (<Pn\H'(t')\pm) = Wnm(t). This is the case.111) Different from Eq. » — < (10. (10.10. t o = 0.Em)t} \{tfn\H'\ipv h 2h(En .8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .En (10. where the momenta of the aparticles form a continuum. (10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0.109) the transition probability Wmn(t) = jp Jo . (10.
if the energies belonged to the discrete spectrum.192 CHAPTER 10.E7m)t} = (10.2.2 The delta function for £ — oo. (10. and no other state would be nearby. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. Hence lim St(a) = 5(a). t—>oo Hence also (shifting t to the other side) sin2{^(ffn. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En.— = o (En 2h En 2h5{En . so t h a t Wnm would be zero. we would have En — Em ^ 0. Otherwise. Fig. though sufficiently small.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. 10.Em). 10.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . .Em)t} r i 7 TTT. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . > We see t h a t this quantity has the same behaviour (in particular for a — 0) > as the function 5K(x) we considered previously.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.io.
266. then Eq.117) were derived by Pauli in 1928.115) and (10. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .8 Transitions into the Continuum 193 as equal.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. If we are concerned with pure scattering. f See E. In the case of the functional (10.117) in many applications. (10. Let p(En)dEn be the number of states in the interval dEn.r Z x 1 ^ ) = 4>{Q).113) this is* hm . _ 2irh AE > > Se.* In view of the usefulness of Eq.W ./ t>oc TZ J_00 dx . Schwabl [246].113). 142.3: Application of test functions to formula (10. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. 75. We evaluate the integral 0(0) = 1. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. Fermi [92]. for instance from a potential. Fermi gave this formula the name "golden rule" . p. . i. Example 10. so that initial and final states belong to the continuum. F Jo . Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). sin 2 (tx) ax i—( * According to F. Then.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . Pauli [227].e. (10. if p(En) varies only weakly with En. in the article of W.113) Use the method of test functions to verify in the sense of distribution theory the result (10.4). (10.116) The transition rate V is the transition probability per unit time.10. pp.117) provides the outgoing particle current (see Example 10. (10.
<p). = ( — ) dk= 2ir (—  2irJ k2dkdfl. p.( ^ V V • < ih 2 _ i k n .e " 1 * 1 = 2 lim — . E = h2k2/2fi. Gradshteyn and I. Jin § m = See I. 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).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 .. ^o n .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . 2 .194 CHAPTER 10. Then ^ . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . formula 3. dEk = h2kdk//j.( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<).947. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. x . The ingoing particle flux is therefore v / L 3 . . Quantum Theory: The General Formalism with the help of Tables of Integrals. .tan p 2 + a 2 . . . M. Then p(Ek)dEk Since Ek — h2k2/2[i. p = 1 we obtain I = .„I. . 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. and is obtained as ih.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10.k. E x a m p l e 10..a 2 ' i that for a = b = t . . S. with v the velocity of the particles. and hence . Ryzhik. . 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. with n . .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . 1 . V +(ab)2 a _x 2p6 7 m —^ '. = 0 ./ cfc a n p 2 +fc 2 . § We have r Jo \p\x dx sin ax sin bx  V. . 2 T / ^ M. * ^ = . iknx . 491.? 2TT / L \3k2dkdUn 1 /V(x).. L \3k2dkdUn It follows that T. Hence the result is 0(0) as expected. . 77^ + .
69). t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. Schiff [243]. specifically in the case of the Coulomb potential. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>). we present another derivation using the usual method of timedependent perturbation theory. 148.x)^(E'. (21.^°>(E.119) only the time dependence of HQ can be separated from the state \ip) in the form of the exponential factor of a stationary state.^ We start from the Hamiltonian H = HQ + H'.x) = S(EE'). Thus we set ^(x./continuum = <San(£)</4°)(x.t). 1 rfx^°)*(x)^)(x) = 8mn.120) In the case of Eq. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known.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. 10.10. . pp.122b) S e e also L.t)+ / n a{E. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.t) = ^On(t)^(x.121) . (10.122a) (10.t)^°\E^t)dE (10.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn). see the discussion after Eq.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. (10. (10. (10. For a possible application of this result. 193. by which we mean without leaving the Schrodinger picture. We expand the wave function V( x .
(10.t)rJ><®(E'.127) we obtain from Eq.x).iJ5 »*/Vn(x). H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). (10. We multiply this equation from the left by </?£(x) and integrate over all space.123) We now insert these expressions into Eq.120). Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.. Next we use the orthogonality relations .x. (10.x. Observe that these conditions do not contradict Eqs.128) .E').196 CHAPTER 10.. into dip (10.e. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x..„. = e^EE'^h5(E Thus with Eq.126) + H')<pn^)eiE^h}. t) = San(t)<pn(x)eiE^h. (10.i) = e... dnptMrnW = { ^ _ En) (10. (10. .122b). . The second term on the left of this equation and the first term on the right cancel out.122a) and (10.t)dx.129) . .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. (10.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. . V>(x.E') = 5{E . / *. (10.127) i n c a g e Qf c o n t i n u o u g £.„„. i.x. f 6kn in case of discrete tp's.t) = eiEtlhy(E. since f rJ>(°>(E.125) where an(t) — dan(t)/dt. ^0)(E..
131) and insert this into Eq. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . i. (10.130) H'kn by \H'kn and at the end we allow A to tend to 1.10.130) we develop a perturbation theory. (10.134) . (10.120). Comparing coefficients of the same powers of A on both sides. 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. let's say am / 0.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. We restrict ourselves here to the two lowest order equations. (10.e. (10. Thus first we have ak0) = const.e. SH'kn(ty^E^l^\t) (10.132) These equations can be integrated successively. In the first place we replace in Eq. (10. they are fixed by the initial conditions.130) This result should be compared with the Schrodinger equation (10. In order to solve Eq. to perturbation theory in the lowest order. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. i.130) along with the parameter A in front of H'kn. determine the state of the system at time t — 0.133) The numbers ak = const. since it is equivalent to the latter.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. and we set ak°}=5km or ak0) = 8(EkEm). Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. SH'kn(t)e^~E^ha^(t). 0. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10.
In this case we obtain from Eq. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). (10. What is the physical significance of Eq. (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. (10. (10.137) .k . ) ' X V(x')dx'. Equation (10. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . H'^y^"^^/hd£.ml2sin2{^frn (En . (10. ifm{kl) = f J —oo (10.135) 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. The amplitude / B o r n is a Fourier transform in spatial coordinates.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. ip) in the context of timeindependent perturbation theory.Emy in agreement with Eq.e.135) / Jo L Hfrfc ~ Em)/h Hence the probability to find the system at time t in the state n ^ mis (with Al) 22 i rjv I1 r i r IW a(l)m2 K ( j l = i^kl\t) = H'km e^E^dt> = H'km 6 \nnm\ ei(EnEm)t/h i(EnEm)t/h _ j [EnEmy 4g. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum.132).111). (10.136) we see that both "amplitudes" are Fourier transforms and look similar. i. i. ipm) into the state k.e.135) We put the additional constant of integration equal to zero. Integrating the second of the equations (10. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'.136) Comparing Eqs.e.135) and (10. so that ihakl\t = oo) = 0.198 CHAPTER 10.
Both of these important potentials are singled out from a large number of cases by the fact that in their case the Schrodinger equation can be solved explicitly and completely in closed form. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.i be respectively the momentum.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.y. ( 1L1 ) We let pi. r=  r . such as the electromagnetic field (the quantization in these contexts is often called "second quantization". Let r : = (x.Chapter 11 The Coulomb Interaction 11.2. meaning the inclusion of the creation and annihilation of field quanta). Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.m.2 Separation of Variables.Ti. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. The Hamilton operator is then given by w= jt + ji_^M. 199 (1L2) . Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. Z<i = Z) y(r) = _ M ^ . the position coordinate and the mass of particle i with i — 1.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. 11.
y. P = pi + p 2 . where m 0 = mim 2 . (11.4) P = p i + p 2 . (11. Z) and r = (x.3) for the individual position coordinates we obtain ri=R+ •^ —r. so that as expected we obtain the expression for p in Eq.2/ mi+m2 The mass mo is usually described as the reduced mass. so that as expected according to Eq. . p are such that as operators in the position space representation or {R._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. Y.U7&. r} representation they have the differential operator forms Setting R = (X. r = rir2 mi + m 2 (11. mi + m 2 iTi'2 r2 = R ^ —r. Solving Eqs. mi + m 2 va\ (11. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . 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 ' .5) The momenta P . (H4) \mi m.200 CHAPTER 11. 2Pip2 m\m2 pip2 mi + m 2 .4).3) and centre of mass momentum P and the relative velocity p/mo by setting r» . /'Pi p = m0 P2 A v. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. (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.
r) is a canonical transformation.pj. we mean n = rx s x. (11. (i. (11.z).pj]=iMij. for which we postulate the commutator relations* [ri. with M := mi + vn^R = ^ P.y.13) * and = M ' 1V1 P = 0. p on the other are completely separated. 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).Ri. etc. pj. r = —.10) we set H = Hcm + H. Pj (which in the following we write again ri.e. [Ri.Pj).16) Although we write ri.12) (11. P on the one hand and those in r.Ri. p = W.Pj} = ih6ij. (1114) m0 We observe that the equations of motion in R. 2(mi + m<i) 2mo 2m\ 2m.10) One can convince oneself that the transformation from (ri.2 It follows that p2 2 P2 n2 n2 n2 (H.2 Separation of Variables.= ^. The canonical quantization must always be performed in Cartesian coordinates.15) In accordance with Eq. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.8) and so W =777 v + 7 T . (11.+ P. (11.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion.+ ^(r) 2(mi + m 2 ) 2m 0 (H. Angular so that Momentum 201 ^ ^ .r2) to (R. Pj over to operators ?i. Ri.pj. Hence Hamilton's equations apply and we obtain.11.+ ^2.il) ""Op' i.j = x. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. (11. .
V(r) = V(r).20) r2MAR_ $(R) = ER*(R). where Hcm$(R) = and Hip(r) = with ^total — ER (11. r ) = J E t o t a i*(R.202 CHAPTER 11.17) (11.23) y — r sin 9 sin tp. it is reasonable to go to spherical coordinates r.r). = rcos9.21b) + E.e.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). Thus the twoparticle problem is practically reduced to an effective oneparticle problem. (11.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R .r) = *(R)V(r). ip: x z = r sin 9 cos ip.Q 2M' + V(r). 0 < 9 < ir. (11. —n < tp < ir. i. The Coulomb Interaction nc H = 2m. (11. as in the case of the Coulomb potential.24) . 9. 0 < r < oo.22) If. 2mo (11.21a) L*+v{r)] ip(r) = Ei/j(r) (11. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .
11. k = 1.2 Separation of Variables.27) We can now write (see verification below) p V = h2Ar^ where 2. that the above expression follows from l:=rxp=ift(rxV).pr]=ih. (11. = (p2 + ^ V (r + 0). otherwise. = 0.26) However._0di})\ x We now demonstrate that 1.28) .k a clockwise permutation of 1. n ( .29) .1 it is shown that pr is hermitian provided lim n/>(r) = 0. . the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r.5r = .A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. i.33) i. _ h2 \ . (11.3. i—>0 (1L25) (11.2.2. where (r x p)j = ^eijkrjpk.3. = —1. (l + . is the operator of the relative angular momentum. k an anticlockwise permutation of 1.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.3. i( i. 2 1 9 / 1 9 . (11.j.j.e. with tijk — + 1 .)In Example 11. We have I2 = (r x p) • (r x p).i ' .32) (11.2. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. (11. if i.31) . (11.12 — I2. j .
(r • p ) 2 + 3ih(r • p) .44) . (11.k Using Eq. This means that l2 = so that 9 9 1 (11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.41) o = r pr.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.42) (11. ih6jkrjpk (11.ih) — rpr(rpr With Eq.ih)pr and x 2 2 9 9 2 2 Q O (11.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p . (11. Eq.40) (11.39) + ih).43) r2(p2p2).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. (11.35) j.37) .ihr • p r2p2(rp)2 < + ih(rp). (11. (11. r— = r • — or or (11.k ~ rjpkrkPj) (11.ihr • p ) .38) + y* + z .ih5jk)pk (11.27) we obtain r{prr)pr = r(rpr . (11. * ' S Since r =xz we have (cf.j.
e'iP^} .e.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11.6. <p)* .45) oe Thus in Eq. 0.<p G [0. 9 € [0. i.<P) = Etl>(r.e. ip 6 X>Pr C 7i. (11.1 Separation of variables We deduce from the expressions of H and l 2 that [H. <p)*Pri>(r. /'(r. <p)*rip(r} 0. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.e lim rip(r) = 0.ip) = (<£.2. <p)r2dfls = .! /'). prip) = = I <j>{r.2TT]. (11.47) Example 11.d. H 1 2mo pi + K) +V(r). ¥ p)r 2 drdn s = (p r 0.[ i J With partial integration with respect to r this becomes (<l>. 6.30) I2 is to be identified with the square of the angular momentum operator. 11.Pril>) J / r= dQs rcj>(r.46) (11.e.e.— (riP(r. Pri/Oi V0.11.<p)*?(ri.<p)drdna " r r=0 J \ll§r~(r4'(r. Angular Momentum 205 Comparison of Eqs.Q ' 2mor2 with the condition (11. i.[ 4>(r. r := r : lim rip{r) = o i .6l. <p))r2drdns % J r or r<t>(r.e. 6.<p))drdns. Then {<j>.d_ r dr the operator pr is hermitian..<p) (11.28) and (11. r—>0 il>(r. S o l u t i o n : Set d£2s = sin.26).7r]. (11. reR3. h f l d —(r<l>*(r. or J . 9.<p))ril>(r.l2} = 0.(r. 8. <p) d °° .2 Separation of Variables.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).6d0dip.. (pr4>. pr = .48) .
206 CHAPTER 11.l2] = 0.49) By cyclic permutation of x. zpx] + [zpy. we have [h.lz]=ihlx.xpz] = [ypz.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. [ly. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0.e. The Coulomb Interaction i.lz (or l .lx or l . xpz] = y\pz.q2 = V. g i = x. [lyA = 0. i.e. 3 .53) Either of these operators is the hermitian conjugate of the other. /z> *xJ = l <i\}ylx + tx'j/Jj [h. H and l 2 possess a common system of basis eigenvectors. However. z]px + py[z. z we have altogether [lx. they are incompatible variables. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. 11.50) Thus the components of 1 do not commute pairwise.l2} and correspondingly [lx. (11. i. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.51) (11. Since 1 = r x p .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. 2 2 2 = 0. (11.lx] = ihly. for any two of the components there is no common complete system of eigenfunctions. 2 . so that [lz. (11. zpx .qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .e. Example 11.ypx) = ihlz.ly]=ihlz.2: Verify that [h. (11. ly] = [yPz ~ zpy. We therefore first search for a complete system of eigenfunctions of l 2 . These operators here play a role analogous to that of the quasiparticle operators . [lz. y. ?3 = z). In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.pz]x = ih(xpy .52) This means that l .
60) Similarly for l+ with the help of Eq. (11.58) {l0 : max.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. A^ in the case of the linear harmonic oscillator.i] = m. assuming a finite dimensional representation space.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. Hence we must have Z+I0=0.h~lz(1156) (11. (11.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.3 Representation of Rotation Group 207 A.54): lzl\l) = {llz . (n. and with Eq. [i+M = 2%iz.l)hl\l).). With the help of Eq. y . in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.59) = 9) h(l0 r)\lr).i+] = m+. Similarly we have lg{l)2\l) and more generally lz(l.e. We write or define: / . (11.r) (11 = {l02)h{l)2\l) (11. (11.49) we obtain the following relations: [iz. In the present context the operators are called shift operators for reasons which will be seen below.61) But lo is by definition the largest eigenvalue of lz.54) On the other hand we obtain from Eqs. l\l) is eigenvector of lz with eigenvalue (IQ — l)h.r) := (Z_)rZ) so that lz\l .Y\l) = (lQr)h(lY\l). (11.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). (11. .57) i. (11. since lz is hermitian. (11.52): 0 = [ l 2 .54) l 2 = \{l+l~ + 11+) + ll = l+l. (11. Therefore (lo + l)h cannot be an eigenvalue of lz.lh)\l) = {l0 . [ig.51).11.+ ll. (11.Z_] = [ l 2 . i + ] = [l2. (11.
68) .60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i. The Coulomb Interaction (l+l.Z_] = 0. i..64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11. i.56) CHAPTER 11.y\i) l)h2\lr).l2\l) = l0(l0 + l)h2l\l). i.54) (n = 57) (11..e. (11. (11. (11.\ll). (11. IQ(IQ 1 0 (11. l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.66) It then follows from the second of relations (11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_. (11. (11.n)}h2\ln).n).67) i (11 60) = But according to Eqs.e.h)\ln) {(lQnf(lQ.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .e.+ 11+) + l2z\l) = i+i. Eq. (cf. (11..e.60) and (11. (11.\lr) terminates at some number (let us say) r = n (n a positive integer).+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). \l).n) (1 = 60) {l)n+1\l) = 0..63) + l ) ^ 2 . l\l .64) i.208 Thus for l 2 we obtain: V\l) ' = 11. [12.55)) we have 12Z_Z> = l.From Eq.
.l.. (IQ — l)h.m). \l — n) are eigenvectors of lz with eigenvalues loh. We write these* for integral values of IQ: \l.m / + l)h . .1). The operators l+.m±l\l±\l. (11..63) and (11.. Satchler [38]. The basis vectors \l).m) (l.. M.71) can be established in analogy to the method used in the case of the harmonic oscillator. The nonvanishing matrix elements of 1 = (l+.70) they satisfy the eigenvalue equations (with m = l0Jo l.m) = IQ(IO + l)h2\l.m)..69) Prom this it follows that IQ must be either half integral or integral. m + l)h. A..m) = = = = y/lo(lo + 1) — m(m +l)\l.m y/lo(lo + 1) — (m — l)m\l.m) with — lo<m<lo2 (11.m) = mh\l..  U o ..m\lz\l.l)h \/(lo + m + l)(l0 .n .3 Representation of Rotation Group so that with Eq.n)} = (l0 ..lz transform the vectors \l). The dimension of the representation space is therefore n + 1 = 21Q + 1. *D. i.l_. —loh.. p.2. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.m + 1)(Z0 + m)\l. lo = l0n.{lo .l0): l2Z.11.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . R.. (11. We therefore write the 2/o + 1 orthonormal vectors: \l.l)h. The normalization factor in Eq. where lz\l. (11. y/(l0±m + l)(l0^m)h. y (lo .m) — mh\l..m) : Mo).m).n){l0 . § See e.m) = = mh. lQ = . According to Eqs. lz\l.1).. (11. we skip this here.g. 24..m)\l..Z 0 ).n) 2 . In the following we consider primarily the case of integral values of IQ. II. \l — n) into one other.65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2.m) l\l. As in the case of the harmonic oscillator we introduce normalized basis vectors. Messiah [195]. Brink and G. 17. Vol. p. (11.m .e. The phases are chosen such that§ l+\l..70) According to Eq. lz) are then: (l.
74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).210 Then. (11. $ oc eim*.<p). .e.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. i.yp. and m = —IQ. as this is sometimes described. where IQ = 0. . for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). for the simple "rotor"... ip. . 11.4 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. . ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ ...<p) = i0{io + lzY£(0.77) $((p) = m2<b(y). (11. d2 sin2 9 dp2 _' (11. mhY£(9. l_\l. ip) = e(9)$((p).76) d2 (11. The Coulomb Interaction Z+Mo> = 0. We . /o. CHAPTER 11.2.1.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation.<p) = i)n2Y^{e^). In these coordinates we obtain: h l± so that h2 = xpv . as remarked earlier. 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.l0)=0.73) 1 d d sm0 06 \Sm9dd) + (11. We now perform the analogous procedure for the angular momentum or. —lo + 1 .With the ansatz Y(9.75) the variables contained in the expression (11. with i2Y£{e. (11. as above. 9.
and that their phases are such that Yt0(0.0) is real and positive. (11.81) 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. orthonormalized system of eigenfunctions. (11. These functions are defined in such a way that they are normalized to unity on the unit sphere.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs. is given by (2Z + l ) ( Z . (11. m > 0.82) .78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}. This is the case if one demands that l 2 and lz possess a common complete. / 3 / 5 J — casO.(p) are called spherical harmonics. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.^ = *(n .i».11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.n>).ypx = tfrK(11.* ? ^ . Uniqueness implies immediately that m has to be integral. (11.t»'\Y7n((l'.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability. We mention in passing that in spherical coordinates d lz = xpy .76) and (11. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.<p) Jo = 6mm'5w.79) . so that im(ip±2ir) The functions Ylm(8. (11.(5cos 03cos0).80) y? Furthermore Y°l = V^fcosO).r o ) ! YT{e.V>)Yir'(8.\ = W .
212 CHAPTER 11.. It is instructive to pursue the appropriate arguments.e.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function. in the replacement ip —> if + 2K. (11.1. Instead of the values I = 0. 22 (11. f.l)xK+i~2 i .p.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .2xy' + 1(1 + \)y = 0.x2)y" .x2)Pf . (1 . i.88) y = ^aiXi+K..e. (11. i=0 (re lowest power > 0). (11. oo (11. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .88) we obtain for every value of x ^2 ai(K + 0( K + * . (11. of Ytm. we would now have to search for those values of I for which the solutions are square integrable.m(m + l ) ] i f = 0.h..86) y = 0. i.89) Inserting this expansion into Eq.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .e.g.87) we obtain the equation (1 . the designation s. (11. i.x2)y" . If we did not know yet that I = IQ = an integer > 0.1(1 + l)]xK+i = 0..e. Setting y = (lx2)m>2Pr(x) m .. For the solution we use the ansatz of a power series.m = 0. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.2(m + l ) x i f + [1(1 + 1) .x2)P[' .2xP[ + 1(1 + l)Pi = 0.d.90) . Consider the equation for Pi(x).2. respectively.85) (11. is also in use. exp(imip). i..
E Xl(r) =0 (11. Murphy [190]. 93. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . > 1.11.1(1 + 1)].1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.92) This is a recurrence relation.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. (11.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.*" 11.94) ^For further details see H.95) Vl = rxiThe hermiticity condition for pr then requires that (cf.94) from the left by r) h2 d2 h2 2 + H! + \)IL. (11. l 2 and lz are the solutions of the Schrodinger equation of the form (11.Qr2 with (cf. Eq. We see that this happens precisely when I is a positive or negative integer or zero.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. but also those for a different series for a. Margenau and G. Schmidt). (11. We have therefore (multiplying Eq. However. p. these are of little interest to us here.78) to be square integrable. M.1. ai(re + !)« = 0. .26)) yi(r = 0) = 0.46)) p.96) + V(r) . (11. Eq.93) where Xl(r) is the solution of the radial equation (cf. (11. when the series terminates after a finite number of terms. For the solution in accordance with (11. For \x\ < 1 we can obtain from this the coefficients of a convergent series. they have to be polynomials (recall the orthogonalization procedure of E. l(l + l)h2 _2mo + 2m. first line of Eq. This is precisely the case.91) from which we deduce that K = 0. (11. .
(11.1)). Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0.1. The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11. In the first place we investigate the behaviour of yi in the neighbourhood of the origin. (11.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. Z\ = Z2 = 1 in Eq. and others.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0.96).e.100) with (11. (11. L i + + . The equation also has an irregular solution.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. an analogous consideration applies to hydrogenlike atoms like. r m0 = meMp me + Mv (11. we obtain the socalled "indicial equation!'' s(s . We now consider the hydrogen atom with V(r) — —e2/r (i.1)+1(1+ 1) = 0. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). since it satisfies neither the condition yi(0) = 0. s = 1 + 1. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. The Coulomb Interaction (11. Equating the coefficient of 1/r2 to zero. Obviously the irregular solution has to be rejected. for instance. i. He + .e. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r.102) . This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. and substitute this into Eq. which behaves like 1/r1 in approaching r = 0.214 with CHAPTER 11. oo) of a onedimensional space. nor the condition of normalizability for l ^ 0.99) e ip(r) = Eip(r).
It remains to determine the function vi(x) for a complete determination of the solution. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 .104) W i t h these substitutions we can rewrite Eq. Inserting the ansatz into Eq.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.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. Magnus and F. (11. W. p.e.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. We can see from Eq.+if>x/2 vi(x).6. *The factor 47reo appears in SI units. Oberhettinger [181]. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. .529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.x ) $ ' .105) 11.108) This equation has the form of a confluent x$" + (6 . behaves there like xl+1) extended to an exponentially decreasing branch at infinity. (11. (11.88. T This equation is known as Whittaker's equation.a * = 0.e. see e.105) we obtain the equation for V[(x).103) °y moe2 ( = 0.11. hypergeometric (11. (11.6 11. We set therefore Vi X J. i. i.e.107) equation. (11.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). (11.g. (11.1 Discrete Spectrum of the Coulomb Potential T h e eigenvalues Here we are interested in the solution which is regular at the origin (i.v) vi(x) = 0. for E < 0 they > decrease exponentially.
i. it is convenient to use only the factorial notation.z) = ^f^. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . with many of these around in some calculation.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). the factorial for an arbitrary argument. .e. called dupZication formula.1/2)!. (z)!(z . (4) r ( i ) = V5F. (5)r(n + l ) = n ! . + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. ^ ( 2 ^ ) ! = 2 2z z!(z . are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases.(0_i)!(6 + n l)!n!' (11. (2) r ( z ) r ( l . and hence the *Some writers. (11. (11.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. In general the function $ is an infinite series and behaves at x — oo like » x{l+l+u)ex_ Obviously such a series is useless in our case. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .! ) ! = imlFI' c a l led inversion or reflection formula.216 CHAPTER 11.106). particularly pure mathematicians. (11. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. 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. However.z(zl)\ = z\. The function T(z) is defined by the integral /co T(z) := / eHzldt. more precisely.112) The gamma function is a nonvanishing analytic. meromorphic function.J. and for both cases.
I) r(vl p)sm{n(i/ I . = n = Z + l + n'. in our case: a = 0. d = 1. i. For every value En. (11.V))T(v .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. This is achieved. . One defines as the principal quantum number the number n.. Eqs. i.1.110) as follows: T(l + 1 + p . This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).1) + n = n(n . (11. Thus the infinite series must break off somewhere. for z. . and so to the number of different values of m.101) to (11.. (11.113) This is a quantization condition. The quantum number n' is called radial quantum number.1) + n = n2.. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. 1=0 (11.» • » • » .2.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. . which are also described as its nodes.104)) 1 n =v = K.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. 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 . 1 .e. n' = 0 .. . i.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. 2 .11. With (cf. the orbital or azimuthal quantum number I can assume the values 0.e. .e.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . if the series terminates for certain values of u.v) T(l + lv) _ ~ K sin{7rQ/ . which is thus seen to be a consequence of demanding the square integrability of the wave function. every n. must be a polynomial.
1 = 1 (called diffuse series. nf . and hence the fine structure of the energy levels is excluded. np. (b) The spin of the electron was not taken into account. I = 0.'.l = 2 to n = 2. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. One describes as Balmer series the final state n = 2..1 = 1 to n = 2. I = 0 to n = 2.1 Hydrogen atom energy levels.218 CHAPTER 11. from n..1 = 0 (called principal series. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.'. The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig. .p. respectively.d. 2 . . stand for I = where n denotes the principal quantum number and s. hence s). . / = 1..4. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy.1.1 = 0 (called sharp series. . 11... and from n. ne.. nd. from an initial state n > 2. (c) The nucleus was treated as pointlike and not as something with structure.§ One also writes n2l+1l. The following spectroscopic description of states is customary: nl —> ns. ' O n e describes as Lyman series the final state with n = 1. hence p).\ I = 0 or 1. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. . 11..1. from an initial state n > 2. hence d).. 0.
We distinguish between different definitions in use. . The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. . S. Vol. l .p)\(2l + 1 + p)\ The spectrum (11. Gradshteyn and I. For clarity and later use we denote the function used there by its normal form L^. Apart from factors. (11.6. (u. • fc. in fact. and I. Each can be recognized by reference to the generating function.x).117) The function *Lk(z) is a polynomial of degree r.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.r f .118a) where Lk.414(3). Unfortunately the definition mostly used in mathematical literature differs. Oberhettinger [181]. Magnus and F. 11.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. With the help of Eq. 3 .1. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. . see. several different definitions are in use. the polynomial contained in y\ of Eq. . • (11. however. 84. M. I.2l { + 2. (11.116) Laguerre p o l y n o m i a l s : Various definitions in use! The polynomial obtained by terminating the hypergeometric series (11.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). p.z) = (r + k)\ Lkr(z). Unfortunately. 2 .2.115) is effectively the associated Laguerre polynomial normally written Lk(z). We mention here three different definitions used in the literature. formula 7. Appendix B. Messiah [195]. (11. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). . W.11. for instance.114b) yields the binding energy of the states.= t .r = 0 .2 (11. a + 2:*) = t v . p. Ryzhik [122].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. k + l.115) ' =0 (ri . • ( . 844. "1'f„+11" P] „. so that one always has to check the definition.
Appendix B. N. It is said that this is the definition preferred in applied mathematics. I. (11.z .220 CHAPTER 11.„M„ z"*L.121) In Example 11. Sneddon [255]. and Ll(z) = l. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. L\(z) = 2z. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 .122) "See A.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. r r Z r=0 v tr k k (11. 162 .119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. N. **I. ^ F o r this definition and the following equations we refer to I. There at some points n + 1 is misprinted n + 1. is defined by^ dk (—'\\k(r^\2 (11.3): e2t/(lt) (1 .3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. p. .2.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk. which we write ^Lk.d z—^z + {k + l~z)—+r Lkr{z) = 0. Sneddon [255]. L2(z) = l2z + z2.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11.164. 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.1. Vol. . 163.. pp.. Example 11.„\*Kk. Messiah [195].** The associated Laguerre polynomial here. Ll(z) = 33z + ±z2. L1(z) = l . . The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'.
(11. and partial integration. ^L\(z) = . (11. and obtain with this the normalization condition (11.124c) Example 11. N d rk 0.119) derive the generating function of associated Laguerre polynomials (11. Example 11. (11..11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z)..3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.4z + . (11.118a) and (11. ^L2(z) = 2 . (11. '"''Comparison with Eq. Solution: We consider the following equation of the form of Eq.120)) it)k& zt/{lt) ~*P p=k fc (11. i L1(z) = l .% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.120).123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !.124b) Comparing Eqs. and *L\(z) = ..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 .4 + 2z.119): ty"(t) + (a + 1 .8): (11.z . *L\(z) = ..1 . Then. . Jfpd>2s{p))eptdp.124a) (i*) pi The orthonormality condition is (cf.121).1 8 + 18z 3z2. (11. The limits will be determined later.107) implies k = 11 + 1 and r = n + I. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.
t > 0.222 CHAPTER 11. which removes [. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt.( a + l ) p s ( p ) + 6s(p) dp 0. this is the rejected irregular solution. We demand (to be considered later for integration contours) the following condition.) 2 J J p 6 + i ( l .124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials.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. agreeing with Eq.p ) « + i q c + i ( i ..a = 21 + 2. 2TT_ Jq=0qb+l(lq)a+l 1 . p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. j p = (P 0 "V?.p ) o + i 7 t = 0 . Thus in the case of Eq. (11. and (4) around p = 0 if b > 0 and integral. .^ ./ ( l — <?) —— = G(t). Hence the expression in square brackets vanishes. p and one has (as polynomial of degree 6.q) and P~ t ( . 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. 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. since for n > b : (b — n)\ —• oo. case (3) similarly.] from the above equation: C(p) := p(p + l)s(p)ept = 0. (11. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). Then dp = dq/(lq)2. s bK> (lq)°+i (11.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. It follows that 1 d i I _i. (3) around p = — 1 if a + b is a negative integer. Case (2) implies the exponentially increasing solution ~ e + t . with requirement of a finite solution. the only possibility is the last.. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . (2) from —1 to oo.119) with a and b as there. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* .p + 1 = 1/(1 . Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a.
and 0 if cj=b.111). Eq. so that I = (c + a)! 1 f . Clearly only the second form is sensible.—i.112). The wave function of the state (n.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.6. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = .(x). I.127) .125a) and the factor a~ 3 ' 2 with a = Bohr radius. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11. 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. Jo (11. .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.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.?). (11. 11. is introduced for dimensional reasons.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . /•OO / Lg(t)£?(t)taefctt: (a + c ) ! . (11.11. where Fnl(x) = yi(x) (11.
11. For the ground state (n = 1.n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .224 CHAPTER 11. thus on the average the electron is the farther away from the proton or nucleus the larger n is.(2Z + 1) + 1} = ^[Zn21(1 + 1)].128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).(l) n+l.3fc + 2).6(n + Z)[(2Z + ! ) .. (11. = 0) we have ) r /u =.130) We see that the mean value of r is the larger the larger n is. 2 (11. + .v (6n2  6nk + A:2 + 6n .3(2Z + 1) + 2} 2 {2(n + 0 . (11.8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .1 + 1) (11. from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix ()r2dr 2 ^i + /„.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 .! ] + (21 + l ) 2 . a (r)is = ~a.2.7. and for the explicit derivation of the second case below Example 11. The Coulomb Interaction One finds in particular (see also Example 11.131) . This behaviour is indicated in Fig.2Z .
j .133) for large r..l ) n ] =  [ 2 n 2 + n] an[ n + (11. / ! J — . \f{r ) ~ n a.6 The Discrete Spectrum of the Coulomb Potential r.11.134) *These are taken from I. S.( n . pp. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . i. with n +1 replaced by n and 21 + 1 by k.J n 1s 2s 3s *r Fig.2 Hydrogen atom wave functions. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 . In the case in which I assumes its maximal value n — 1.(r) = x W 2 n + 1 =. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.2 I 3/2 Kl r 225 \2 1 R„.132) (11. 173/174. we have and tp becomes particularly simple.e. = l a nl n N'nll rFnl.2 1 A r = V (^2) .^ We obtain therefore for the quadratic deviation (r) . or (r ) ~ n a 2 2 ^ [ 3 n 2 . 11. v ^ V2n + 1 (11. Sneddon [255].20. Problem 5.
determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory".e. we have pg = const.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). and separating this in cylindrical coordinates r. 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). „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . r.e. However. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. = ngh.— > where ampler is the length of the semimajor axis of the elliptic orbit. pJ = —— and j> Pidqi = nth. Example 11.226 CHAPTER 11. Thus. . for these values of I it is not distributed with uniform probability over the spherical surface. Since 6 is a cyclic coordinate. (c) The nucleus was treated as pointlike and not as something with structure. p2 = p2 + — pg • eV. The Coulomb Interaction or TT = ^ = (11.3.r. This means. but rather like classically in a plane (i. \/v vT 7 v^ See also Example 14. in which periods are obtained from the BohrSommerfeldWilson quantization condition. we consider in Example 11.135) This expression becomes very small for large values of n. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). i. In order to obtain an impression of relativistic corrections in a simple context.4 the relativistic equation in cylindrical coordinates. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.6. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. the states Ytl are predominantly concentrated in the xyplane). i = 0. 6.
which are related to Cartesian coordinates in the following way.11.T).136a) (11.* See L.6. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. 11.2. which correspond to a new choice of polar axis. y=y/%qsanp. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. 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^(9.4 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 £..3.V . ( v c + i ) . . as also indicated in Fig. where ip = 0. 11. (11.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. where the constants are identified by comparison. (p). z=^rj). Schiff [243].(p . The deeper reason for this are symmetry properties of the problem. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. <p = <p.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». where r = ^/x1 + y2 + z2. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . it is possible to separate the Schrodinger equation in other coordinates. r\ = r ( l — cos#) = r — z. p. Whenever there is degeneracy. 2 2 Therefore the energy is no longer a function of n = nr + ng. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'.9. ip are the spherical coordinates.136b) £ = r ( l + cos#) = r + z. 86. y — 0: V ^ c o s tp.
(11.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. so t h a t the variables can be separated. 1 . Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . We rewrite the first of these equations with the help of the relation M (w ^  m + e) .228 CHAPTER 11.22) becomes — as we show in Example 11.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. = 0.3 Parabolas £ = const. 2 . for uniqueness of the wave function. . . Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. and r\ = const. there is really only one basic equation to solve here. The Coulomb Interaction Fig. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. 11. a separation constant. with zaxis as axis of rotation.136b) the Schrodinger equation of relative motion (11. In terms of the coordinates (11. .
(11. as well as the Laplacian A = V and the Schrodinger equation. we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". T h e n by comparison with the spherical polar separation the answer is as in Eqs. .11.6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. In the spherical polar case we had (cf.± ( m + 1) = 0 . . Adding now / . Eq. . . . . 1 . n 2 = n" . 1 .125b) (dividing by p 1 / 2 ) £ oc e .i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . We leave the calculation of the degree of degeneracy to Example 11. To obtain a closer analogy with the separation in the spherical polar case. . m0Z1Z2e2 ^ 2 = n (say). 2 . ds2 = dx2 + dy2 + dz2. p = /?£ and cr = /^r?. . " .6. from Cartesian to parabolic coordinates. .5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric.114b). (11. E x a m p l e 11.e.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). i.106) and (11.105)) Hence we write I = (m — l ) / 2 . .o. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. . 2 . T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) .
.e. i. N2 where g^dS. all cross terms cancel in the sum of 7 the squares. Thus dz2 = . 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .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. / — cos ipdl.136a). + l 4 {^^)drl f i + v \ . > . The Coulomb Interaction Solution: We begin with z of Eq. Similarly we proceed with dx and dy. for m / . and so dz = (d£ — drj)/2.230 CHAPTER 11. We use the method of selector variables.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. for instance. </.2d? dr)}. n 2 (1UJ)2 The coefficient of a .114c).tx2 . We then sum all possibilities and select the coefficient of ojn~1 in the result. / J„^2 . Beware! For m = 0 there is only one mstate. and attach a factor to to each term introduced in the wave function. We consider first the case m = 0. z = (£ — rf)/2. 0 there are two. (n _ !) n = n2 §See e.2) and summing over m. Solution: For the wave functions ip = £(£)J\f(r])&(ip). Thus we want to obtain the coefficient of wn~1 in £2 .e.g<p 9 \ . (11. we have to find the number of values of the quantum numbers n\. . = e±im^e("+^'2{par'2L^{p)L^ (a). see Eq. 2 + ^dV . Oberhettinger [181].l)a + (n . Here we have (see preceding text) 711+77.136a). Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). Observing that the highest possible value of m is n — 1 (i. (11. Thus.e. are elements of length.g.2 .2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . Eq.E E n l „ .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr.2 = n — 1. W. as t h e two signs of t h e exponential indicate.. (11. the total number of wave functions we have is the number n with m = 0 plus twice the number with m ^ 0 starting from m = 1. Magnus and F. In the case m ^ O w e have n\ + 77. 2 2V? V^ One now performs the square of this and obtains similarly dy2. dx = d{ y £77 cos ip) = .1 in this expansion is seen to be n. 712 and m which give n.2)d} = (n. .e.1) ~ . i.l ) ( n . {g^Y. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. 2 _ = (9id02 + (gvdvy + /„ .. n the double sum ]T]2 is TI — m. etc. " . \— < / — cos ipdr] — \/£r] sin ipdtp.[df + dr]2 . when m = 0 = 77. last chapter. i.2)d/2.
1 (cf.6 The Discrete Spectrum of the Coulomb Potential Example 11. i.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.128). which will then give the change in energy of that state to the first order (i. 8. = j (2ni+m+l).) = ^eF(p . Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e.•* n (l) = / Jo It follows that . eF{z). where F is the electric field. fc — m. I. (11.127) and setting n = r + k.™(p)] 2 By contour integration (see Examples 11. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). {p — a) — = dxe~*xkvLkn (z)] 2 . { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. with RayleighSchrodinger perturbation theory). we can relate the expressions Iq to those of Eq.e.—r~n = 3(mn 2 ). With the help of the wave functions we derived above. Z2 —> e.g.. with z = r cos 9. . Thus e. (11.a). i.e.8 and 11.6). Ikr+k. However. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£.128) if one makes there the substitutions n —> n\ (or 712) + m. L ni\ ny.11. . . Sec.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. (11. . and determine the average value of v for a given state.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. .7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz.r. with the connection » (11. In fact. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. with Z\ —> Ze. s 0/ . these expressions agree with those of Eq.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. if we expressed the perturbation in spherical polar coordinates.124c).e.drid<fi oc (p + o)dpda. we would have to use degenerate perturbation theory and obtain the same result. we now have (pa) J dpdaip* (p — a){p + a)ip J dpdaip* (p + a)ip Jo°° Jo°° e(e+<r)(p*r[L™ rm+2 i rm 1 n1 n2 _ Jrm+2 1rm "2 n1 (p)L™ (<r)]2(p2  a*)dpda /o°° Jo°° e(o+>(p<T)[L. with Eq.
( 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)!.p ) ( l .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.pK — — —L .0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .3: /"OO I := Jo et[L^{t)]2tm+1dt. the integral arising here becomes with Eq. c positive integers.g )(l. s = at.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. _ „ .9 )« Solution: Using the generating function for both associated Laguerre polynomials.n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .128).1 ) ! (n . ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .3). The Coulomb Interaction E x a m p l e 11.1)! n\ This verifies the expression I™t in Example 11. the integral / is the coefficient of pnqn in the expression _ / .p ) ( l . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 .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 . 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 .a. and is a special case of Example 11. E x a m p l e 11. (n + r . In Example 11. tt„:_ (1 .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 .3 by one power of t.8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. (11. (11.^~T^——^ ) = coefficient of (pn .232 CHAPTER 11.p ) r + 2 [(lp)(lg)]"l+1(lpg).9 we consider integrals with an arbitrary power of t.l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .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.124d) of the alternative integral representation obtained in Example 11.. dt = ds/a.9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. .7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.x ( n + m)! ( n .
e. we have to have y + (c + b — 2r) = 1.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. c + b even) and/or j / = 0.11. .p ) ] . we have 2r = 6 + c = even.(i) = i ! ( .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 «. = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.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. — (cr)!r!(6r)! Kv. c + b odd).L.w(r) = and / a ) = r\(P ~ r .l ) * x coefficient of a* in the expansion of 7 ( a ) .111) of the gamma or factorial function.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.?(t)t a e.e.r ) ! a ' . c + b — 2r = l (i. c + 6 — 2r = 0 (i.a t [Lg(i).^ ( l . so that = (q+fc±£) . where (a+**£)! c j n Jo = . (a) j / = 1. i.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .e. ° ^ + a ( l . we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l .f ].124d) of an associated Laguerre polynomial. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . Using the contour integral representation (11. 7£.™( r > (1 + a)P r £^0y^w Jc+6_2r = .q)(l .e.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 .1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. (lp)(lq) + a(l q)(l . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. i. .p) .a _ 1 = [1 + a ( l .
cannot really — and this means in a rigorous sense — be applied in this case.^)(. t We have the Schrodinger equation of relative motion of particles with charges Z\e. also with regard to earlier literature. . however parallel to the Coulomb case. ^A very readable source to consult. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. Taylor [267]. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). Since such a rigorous treatment is beyond the scope of this text. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. In view of the slow falloff of the potential at infinity. . These results are seen to agree with the expressions 1^. 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.7.~* 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 . the standard principles of nonrelativistic scattering theory.1%^ 11. + (!±<)«. R.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16.)!' f^)(. is the work of J. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.7 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.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. b\ V cited in Example 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.
z).87 under Kummer functions.b. z = rcos<9.. i. Magnus and F.z) + W2(a. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0. (11. where ^.z) = Wi(a.. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. with the asymptotic expansions (11.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). (11.+ ^  + .11. b. Our question is now: How does ipr behave for r — oo. (11.b.b.109) we can therefore write the solution ipr = Aeikz$(i>y. that we encountered previously.z)). for r — oo) • ipr = Aeikzf(r . 86 .. W.140) and using (11.108) and (11.140) The simple Coulomb potential permits particularly simple solutions.143) and v = ikrj.136b).e. pp.146) See.7 The Continuous Spectrum of the Coulomb Potential Then Eq. i. (11.e. . . . According to Eqs. what is the > behaviour of 3>(a.^ W ) = 0.144) *(a. for instance.z).. „ az 1. A = const. (11.141) Inserting this expression into Eq. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr).M) = l + . a(a +1) z2 ] (11. (11.142) This equation is of the form of a hypergeometric equation. (11. Oberhettinger [181]. v = ik(r .138) becomes 235 ^A + k2 . (11.
^l*7 l + O (11.z)T(l . The Coulomb Interaction W2(a. (11.150) ^(rz)]1"^ 1 + 0 T(«7) i.236 n\Zzab CHAPTER ^T(n '£ fo 11. with 2. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.149) (11.151) with Eq.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre. i. > A = Aeikz[W1(i1.z) + la)T(n + b. = r cos 0.e.b.152) determines the quantity f(9) as /(*) = 7 r(i + i7) exp k(lcos6)r(lij) 2sin2(f) ?7 In sin 2 7 2 exp 2 i a r g r ( l + ij) — ij In sin ( ^ 2&sin2(f) .i 7 ) (11.152) Since z = r c o s # .l. (11.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].kz+l^ktrz)) C .(11. a comparison of Eq.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.153) The asymptotic solution represents t h e stationary scattering state (E = const. J\") ci(kr~/\n2kr) (11. if tbr were „ikr ibr ex el"z + f{6) r (11.l.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo. (11.7 ln*(rz)) k(r .154) .148) . Without t h e logarithmic phase factors.a) (z) na)"~ r(la) T(ba) n\ ' (11.e.ik(rz))} =^ + ^d.
(11. The problem of the logarithmic phase. For the incoming wave ipi = exp(ikz). we see that this expression is the exact analogue of a current density in classical considerations. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. In other words. the current density is * h . 2im.156) .Q hk mo v0. z — — oo.11. that we encounter here is a characteristic of the Coulomb potential. We define as current density the quantity h J := 2imo [^(v^)v(v^n.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain.7 The Continuous Spectrum of the Coulomb Potential 237 the wave could be looked at as the sum of an incoming plane wave exp(ikz) and an outgoing scattered wave. (11.n ..155) Since the gradient is proportional to the configuration space representation of the momentum. 11. 11. The quantity /(#) 2 would then be a measure for the scattering in the direction 9. also called phase anomaly.4 Variation of the observed differential cross section with 0. do dQ K 12 K Fig.7. As a consequence of the masslessness of the photon. the effective range of the Coulomb potential is infinite.
152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.238 CHAPTER 11. This implies that scattering takes place for the attractive as well as the repulsive potential.154) we chose the prefactor of the incoming wave in (11. is called scattering amplitude.152) as 1.e. That this formula retains its validity here in quantum mechanics is something of a coincidence. which yields the scattering cross section. in the current this would in any case contribute only something of order 1/r. 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. for 0 —> 0.157) 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.153): 4fc2 s i r  The expression /(#). In reality this divergence does not occur. In Eq. 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. We make a few more important observations. We recognize the result (11. (11. Correspondingly the other part of (11. the square 7 2 ). thereby screening itself off from the surroundings. We see that a depends only on the modulus of the potential (i. i.159) as the Rutherford formula which can also be derived purely classically.e. (11. 'screened ZlZ<2e2 cr/r0 ° . The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). since in actual fact the Coulomb potential occurs only in a screened form. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V .
t) = e . t 0 )^k(x.165) ^ ( x ) = £ fe Vk(x). i. 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. (11. of the equation r fi.2 fr2 .i i f (''°>/Vk(x. the wave packet in the absence of a scattering potential. v0 = .8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.yJ ^ j A k o W 727)3/2 e 0. of continuum wave functions.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. 11. and which provides the uncertainties arising in quantum mechanics. be given by ^ f dk AL. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . (k).e.t). i. i.161) where Ak0(k) ~ A(k — ko) and ^ k (x. / dk 7^340(k)4(xxo. t 0 ). 2mo V + F(x) k (11.t). t) = U(t.164) Ek~Eko + {kk0)v0h.166) Let the incoming free wave packet at time t > t$. m0 m0 and V. 11.4. t 0 ) . The wave packet is a superposition of waves ip(x. The particle velocity v is v = — . 2 (11162) where Vk(x) = ^ ( x . and if H is timeindependent one has the stationary wave function ^(x.e.11. t0) = e.8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig.x° ' r j . (11163) (11. which in the case of free motion (zero potential) are simply plane waves. (11.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * .^ ^ % ( x ) .e. whose maximum moves with the particle velocity.
Then with k = ^(ko + k . The Coulomb Interaction In Example 11.10 we show that the solution V'k(x) of Eq.167) we have ^k0(xx0.170) A ko (k) e .167) by replacing the plane wave exp(zkx) by this scattered wave.x °Vk(x). i.ko) k0 (11.t o ) / V k ( x . (11. the stationary scattering wave.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.240 CHAPTER 11.154)) that for x — oo.171) The phase of fk(0.iB *(**°)/ R °'tj + 1 pikr e * . (11.k0) ~ k0 + and the expansion of E^ in Eq. (11. ^k(xxo) ^k(x) ~ = e.169) and (11.xo. (11. Eq.e.173) . k = kez.<p) + Ol Akr l (11.161) and (11. (2TT)3/2 eik* + —fv(P.ik . i. (11. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf. this solution > has the following asymptotic behaviour in which f(6.k 0 ) 2 ~ Jk20 + 2k0 • (k .169) The wave packet in the presence of the potential is obtained from the free form (11.x 0 ).i) dk (2TT (11. From Eqs. With Eqs.t) = J ^ A k o ( k ) e .165) ko • (k .172) h h k0 {rv0(tt0)} (11.166).163) we obtain </>ko(x .e.^ ( * . ip) could be carried along but we ignore it.ip) is the scattering amplitude with respect to the incoming plane wave.
(«)/ ipko(x . with the energy of the maximum momentum of t h e wave packet. 11. r ' ) by the equation (A + fc2)G(r. t) . t) f.8 Scattering of a Wave Packet so that with Eqs. i.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. however. U(T) = ^ V ( r ) . (11.z Fig.167).5 Scattering of a wave packet with scattering centre O. Here ^ (x' — xo.to). (11.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . the solution can be written iMr) : (2TT)3/2 ^dr'G(r.r')V(rW). . and determine the Green's function G(r.5. Solution: We define the Green's function G ( r . Adding a suitably normalized solution of the free equation.r'). (11.171).x 0 . r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.to) has t h e form of t h e wave packet in radial direction as indicated in Fig. A particular solution of the Schrodinger equation is — J G(r. '.^y /k0^.e. 11. T h e scattering amplitude is t h e same as t h a t given by Eq.154).^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.y)eiko'Xoe'[fcor£.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. E x a m p l e 11.*0ko (x .11.r')l/(r>fc(r').174) r /k0(^.fco(tto)/ft]^0o)(x/xo. r ' ) = 47r<5(r .r ' ) .x 0 . *. (A + fc2)t/.
one obtains another Green's function given by G_(r)= r eikr. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . These are the socalled outgoing and ingoing Green's functions. .6 The contour C+ in the upper half of the k'p\ane. 11.=k+ie/2k 1 d f. The Coulomb Interaction We find a set of Green's functions G(r. e > 0 small. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r .fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane.fc ) oo dfc' 1 d nik r 7rr dr / . 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . 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.Rek' kfc/2K Fig./2 dk'd(—cos 6)dip iki.6. displaced slightly away from the real axis as indicated in Fig. Hence we write G(r)G+(r). 11. Imk . dk' ik'*r 2TT2 fc'2 .O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. Thus we consider the contour integral taken around the contour C+. which is such that the contribution along the infinite semicircle vanishes.242 CHAPTER 11.
21 + 2.100) — 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.e. Eq. (11. z) + W2(a.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. (11.0 + W2(l + 1 + iy. b.£). the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. e. b. Then yW (11. b.9 Scattering Phase and Partial Waves 243 11. Before we introduce the general form of the socalled partial wave expansion. 2ikr)] [Wi{l + 1+11.g. z).z2 27fc 1(1 + 1)' yi = 0.e. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. Then £^<fc + (2l + 2 £ = ~2ikr. (11.176) " 0—A ~ (I + 1 + iy)<t>i = 0. i. i.11. If we proceed from the radial Schrodinger equation.178) (11.2l + 2. z) = W^a.21 + 2.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. .145)) F(a. Thus we proceed as above. under the stated conditions 2/07 = Z\Zie2 = —Mo. Eq. whose continuation to infinity is again given by (cf. 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. where Mo is a parameter in Chapter 16.9 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.140) (previously we solved this only for the bound state problem) § . (11.2Z + 2. (11.
244 CHAPTER 11.183) r(Z + 1 + 17) := Reie. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.ry) = e"^r(Z + 1 + ry). The Coulomb Interaction Using Eqs.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .146) and (11. .ii)' (11.180) ^ ' (11.n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .j syyll iy il( _ .i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.«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) (11.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 . Then e2i5< = Reie Re~ J2i9 9 = Si.' .l e*"" + T(l + 1 + H ) T(l + 1 . (2kryr{i" V 7 .181) so that eldlT{l + 1 . T(2l + 2) (2fcr)*T« (Alliy ikr .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 . (11.
185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). as we saw.114a) reveals).139) n' = 0.e. (11. i.e. In the present quantum mechanical considerations I is. considering I as a function of E extended into the complex plane._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.188) in the form y/E. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. That the energy E appears in Eq.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. .. kr) the regular solution with the following asymptotic behaviour: r 0 (11. One defines as regular Coulomb wave Fi(j.. with Eq.1.e. For instance the poles of S are given by Z + l + z7 = . However. 245 (11.n ' .11. i.2. where the continuum of the spectrum starts. (11..9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.e. g = e«* = rSi + 1+ *V*ri (11. i. (11.187) contains almost the entire physical information with regard to the Coulomb potential. a positive integer. i. . considering I —• an(E).184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase.
(11. (11. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt.7. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. 11. Singh [252]. we obtain trajectories which are known as Regge trajectories.246 CHAPTER 11. we obtain the partial wave expansion of the scattering amplitude as follows. and thus in general. The Coulomb Interaction and plotting these for different values of n. 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.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. In these cases.191) .7 A Regge trajectory of the Coulomb potential.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. For the Coulomb potential a typical such trajectory is indicated in Fig. 11.
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. (11.g. Oberhettinger [181]. The mathematical expression can be found in Tables of Special Functions (there cf. The potential disturbs this phase relationship (by delay or absorption or redirection). W. r e. . pp.~ J > Z + l)Lil*e~ikr .196) "See e.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.eikr\P^cos9).i f c r .k)=me2iS^k\ in which r\i^\ indicates absorption. Magnus and F.11. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .22.192) Thus the incoming and outgoing spherical waves have 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„ ' . (11.r ^ V ^ Wostf).193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. 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 ' .i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. (11.^2l + l^S^k) lik (11.195) with the scattering matrix element S(l.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9. 21 . 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.
This page is intentionally left blank .
was developed only in the last two to three decades. It is very instructive. one of the main and now wellknown methods to perform such calculations. which in one form or another. which a particle in classical mechanics would never be able to intrude. i. however. Razavy [236]. Similarly difficult is.* *See e. as — for instance — for trigonometric or double well potentials. For various general aspects and applications of tunneling we refer to other literature. 249 .Chapter 12 Q u a n t u m Mechanical Tunneling 12. It is not surprising. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. These newer methods will be presented in detail in later chapters. in texts on quantum mechanics is rare. states of a finite lifetime.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. i. therefore. that quantum mechanically a particle may have a small probability of being in a spatial domain.e. that the explicit derivation of such quantities in nontrivial contexts.e. in general. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. Such a vital quantum mechanical phenomenon is socalled tunneling. M.g. In fact. namely the instanton (or more generally pseudoparticle) method. to explore first much simpler models in order to acquire an impression of the type of results to be expected. the computation of finite lifetimes of a quantum mechanical state. can be found in any traditional text on quantum mechanics.
t) that the one particle concerned is somewhere in space at time t. so that p = T/>2 is the probability density for a spatial measurement at time t. Quantum Mechanical Tunneling 12. t) was developed by Born in 1926.4) (12.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. r (12. i.t) = HiP(x.250 CHAPTER 12. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. M^ Then dt + Vj 1=0.t) (12.1) as the probability described by ijj(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.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. According to Eq.e.t)\2 =1 (12.t).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.. ihr(x. (12.
If there is no absorption. higher dimensional delta potentials do not have properties like that in one dimension. 'Maybe the reader dislikes being confronted again with the onedimensional case. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. which is the case we consider here. the problem is analogous to that of the scattering process from a potential in a onedimensional case.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. we can do this also here. ~hr> k = n^ (12 7)  .12. This case plays a special role. However.33. the equation can be integrated at every point. The Schrodinger equation for this case is* ip" + [A. Gosdzinsky and R. 29 . Nonetheless.2a6(x)]il> = 0. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. pp. Schiff [243]. as we mentioned above.2 . 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. t For the singular delta function potential a modification of these demands is necessary. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. 2m E 2 V0 > 0. Tarrach [119].3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. 12. In much the same way that we define there reflection and transmission coefficients.6) . Delta potentials in more than one dimension do not permit bound states and scattering. This effect is known as "tunnelincp. 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. Such a region is for instance the domain where V is larger than the total energy E. even for discontinuous potentials. If ^>(x) and V^>(x) are known at every point x. where „ mpVo V(x) =+V0S(x). since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). although in general small. Thus it is possible that the probability for the particle to be in such a domain is nonzero. 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. (12. as shown by P. From this follows that both p and j have to be finite and continuous everywhere. 1 0 ~x o a= f See L.
Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). i.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. The continuity conditions at x = 0 are therefore for V : and according to Eq.8) (12.[^']o.e. except in domains of the potential V. 2 (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.ik(C .ikB) = 2aC or 2a(l + B). (12. i. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .e 1 + B = C.9) for ij.(ik .e.1) . (12. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.: (i2iQ) Ce for x > 0. (12.11a) ikC .2 a J 5{x)4>(x)dx = 0. a — ik £Qr < { eikx a eikx •t" . [<//]0+ .ikC = i. .e. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.252 CHAPTER 12.13) ~A R= for x > °(12.12a) (12.11b) 2aC. i. a 2 — a 2 a — ik a2 + k2 ' . e Q (12.' : It follows that ik .= 2a^(0). (12.
ip'(0) = .e «x = J \ CeKX for for x > 0. (12.12. = . as coefficient of the outgoing wave. R.2 = . .K C . the quantity T. . and C follows from the normalization: OO A. K = a. Proissart [223] or other books on scattering theory like R.(K)C = .16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. 2m 0 (12.2 a C . poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. f Ce~KX c.e. oo (12. We write the solution ^ so that ^(0+) = ^ ( 0 .g.2 o ^ ( 0 ) . Thus ^ = Ce. z<0. possesses simple poles at a ± ik = 0. i. See e. 'Note that the physical pole has negative imaginary part. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. at k2 = —a2. when normalized to 1. Newton [219].) and V>'(0+) .i 8 ) . ( 1 2 .19) Thus the delta function potential supports exactly one bound state.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. Omnes and M.K 2 < 0.17) da#(s) 2 = 1.a l s l. i. or h2 E = —a2.e.
12. (12. . V(x) V=E a v=y Fig. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.20) for for \x\ > a. Schwabl [246]. The latter contains also specific applications in nuclear physics. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. 58.a]. p. p. or in the lectures of E.22) + De Feikx + Ge~ikx *We follow here largely F.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. oo]. Fermi [92]. 12. Quantum Mechanical Tunneling 12. x G [a.1. \x\ < a. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.1 The square well potential. (12.4 Scattering from a Potential Well We can immediately transfer the above considerations in a similar way to scattering from a potential wen* as indicated in Fig. We want to consider the case of stationary states for E > 0. —a].254 CHAPTER 12. 55. x€[a.
27) v ' Similarly Eq.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.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12. Fikeika . (12. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. I (12. (12. M(a) .28) Prom these matrix equations we obtain ( B ) = MWM(fl)_1 ( G ) ' where M(a)1 = v ^12'29) i / [i + n i K\eiKa+ika *)e (i t\ K\„iKa—ika fcje \ ] J fl2 30) • J ' 2 \ ("I — K'Jgita+Jfca Q _i_ K\p~iKaika \ . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.26) K^gifoa—i/ca fc. ( V C n vD ). (12.4 Scattering from a Potential Well 255 The connection relations obtained by equating ip(x) and its derivative from left and right at x = — a and a yield the equations Aeika + Beika = CeiKa + Df.12.24) yields F ^ G. fc Af(a) = .ikBeika = inCeiKa + DiKe~iKa.ikGe~ika.e.iKa ? ikAe~ika and . ).. (12. .f )e" t\ K\0—tKa—ika .
e.^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. x € [a.e. xe[a.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave. i.256 CHAPTER 12. (12.31) we have A AT(a) = )MFo B ^r)sin2na . (12. \R\2 = T(E)^sm2Ka (12.35) with with with x G [oo.oo]. i. (12.e sin 2na IJlika e B = r)F sm2na.a . Quantum Mechanical Tunneling Hence with the complex quantities IK (12.32) [cos2«.4T(£)e ifci. = < Ce~lKX + DeiKX .22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. . we obtain A = F( cos 2na — . (12. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.a ] .i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.37) .a].(1 + e 2 ) sin2 2na 1 .—2ika [cos 2/ta + \e sin 2«a]e~ (12.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .34) into Eq.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.
(12.40) we obtain cot(na) — tan(/«j) — i K IK . 83. (12. p.12. cot(fca) = i— or tan(fta) = K K .39) 2 i. in the domain of A imaginary and K real. (12. B.42) These equations have no solution for K. for VQ < E < 0. i.e.41) i.44) . However. an equation of the type f~1f = g19. (12.35) and using Eq. 2\k K. (12. Dwight [81].31): 1\T(E)\2 257 1 . We see in particular from Eq. (12. i.4 Scattering from a Potential Well On the other hand according to Eq. and k real. 2m0\E\ h? ' (12. they have solutions. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)].e. formula 406. Li ( — H— I sin 2na = 0.e. which is satisfied if either / = g or / = — g'1.12. (12. .43) Thus we set k — —i ' H .e. nor when both parameters are purely imaginary. (12.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12.38) as expected.
The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. (12.42) which then leads to bound states with associated even and odd wave functions. .46) This condition implies the eigenvalues E^ER = n2^^ V0>0. 2Ka = nn.47) These states in the continuum are called resonances.20). The amplitude of the transmitted wave assumes its maximum value where (cf.46). k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. (12.258 and obtain the equations CHAPTER 12. from Eq. Quantum Mechanical Tunneling .35)) for E > 0 sin2Ka = 0. (12. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1.g. Eq. i. The remaining wave part becomes exponentially > decreasing. pp.' We observe that T(E) has poles at those values of E which supply binding energies.45) These are the equations that one obtains from the connection relations for the case of Eq. Schwabl [246]. (12. n = 0 .34). We now return to the scattering states.49) Taylor expansion of the denominator in powers of (E — ER) yields . T(E)e2ika = — r .48) We expand T{E) around these energy values.50) See e. . (12. ± l . ± 2 . (12. i. 2m0\E\ and Kcot(Ka) = y—^—L. „ . t2 2 (12. Also. At their positions cos(2/«z)£.e. 64 . around the values given by Eq.e. 2m0\E\ Ktan(«a) = — y——^— .l ) n . A = 0. . This can be understood.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . . The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . F.fl = ( . (12.c o .e. and is itself infinite there. Then according to Eq.65. (12.l . i. = 2/rn (EER). (12. t&n(2Ka)ER = 0.
2a<x<2a. T(E) has poles at E= ERi: (12. These are precisely the resonance states. . 12.e. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. A potential of this type is known as a PenneyKronig potential.' ''' 5a 3a \ a ~~~.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. 12. a 3 a. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. s \ V0 /• ~\ 'X . .52) Here ER is the energy of the resonance and 2/F the lifetime of the state. 12.2 The PenneyKronig potential. We restrict ourselves to the period — la < x < 2a.' ' 5a \V / Fig.12.^ l / . This result is physically very plausible.51) i. V(x) = 9(x + a)VQ6(ax).5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12.2. V(x) x=2a x= =2a . = 0.
2 < a .e. We set . = EVip(x).55) The Hamilton operator of the problem.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. the differential operator is even in x. and we choose (12. with V as parity operator. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. and hence for = Eip(x). VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). E^En = ^ ^ t n = 0. (12.a ) = 0 = ^(a). i. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo. The particle of mass TUQ whose quantum mechanics is described by the Schrodinger equation is free to move in the domains under consideration to the left and to the right of the central barrier. . i.1.Vo) n.2.. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.e There is the additional boundary condition (for Vo = oo) ^ ( ...53) *w«w^( « j)'*^{i i ^ . HV^(x) = H.2 = W h 2 _„2 = 2mo(E .e.. In view of the periodicity we have ^(x) = ^(x + 4a).260 CHAPTER 12. . (12.. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r.
— a]. This wave function is no longer restricted to the interior of a box. which.59) Asink(x + 3a) with x & [~2a. ±Asink(3a — x) with x€[a.il>]?0.2a]. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . — a].a]. VV(°) = °. so that (W meaning Wronskian) W[tl>+. The central region of the solution is the classically forbidden domain.63) = KBe~ F = ..12.2a. (12. (12. We > obtain even and odd functions ip± by setting ^ = ±1. = Ka ±Asin(k2a).2a].2). (12.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . V+(0) = const.58) D sin k(3a — x) with xE[a. we shall not exploit here. ^kAcos{k2a). We therefore have different boundary conditions. however. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. This means that i>±(x) = <  = ±1. 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. We expect.a]. of course. At x = ± a we now have to connect ip and ip' of neighbouring domains.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].. (12. xe[a.Be Ka (12.60) The boundary conditions that ip± have to satisfy. 12.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a. V'(O) = const. that these are such that for VQ — oo they approach those of the previous case. are ^_(0) = 0. (12. BeKX±Be~KX with xe[a.
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. say.e. + 0 ( e . i. the odd solution with. we set tanhx = l .68) . say A. For large values of Vb. is the one belonging to the even solution. (12. i. (12.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . (12.nir + 0[{2ak . the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.even.— (1 ± 2e~2A + O (—).e.262 CHAPTER 12..mr)2}.+ • • • 1! cos^(n7r) 2ak . (with reinsertion of nir/2a for k) 2ak ~ nir . 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.e. where Akoce'2™. i.2 a . Quantum Mechanical Tunneling Fig.2 e . i. k = fc0dd.65) and we expand tan(2afc) about nir. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo.nir)2}} = 1 ± 2e~2Ka.3 Level degeneracy removed by tunneling.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. However. for VQ — oo. In general one now rewrites the equations in matrix form.67) The value of k belonging to the upper sign. Taking the ratio of the equations in both cases.e.nir) + OU2ak .64) we obtain ~[(2ak k . (12. 12.66) Inserting these expansions into Eq. = AA. F ^odd (12. It is plausible therefore to use appropriate expansions.
V ^ 0 are given by the equations ^o + " J . The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. w(x0) •• = = mogy(x).4.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E.3.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. Classically the car is unable to overcome the bump. p. . i. Compute this probability.2 dxy/V(x) (12.^ 0 = °' Hence in dominant order •00 — exp . Fermi [92].V{x)\i. b > a. . Example 12.4 A car meeting a bump in the road.v = 0. y(x) = sin and y(xo) = 1. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. exp .e. (12. 2m0E „ 2mnE .70) xo = (a + b) 2' .. as indicated in Fig. exp r rb rb / •J a 2mo (V(x) .. 12.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. 12. . 57. 2mn. but quantum mechanically there is a finite probability for this to succeed. and is a reformulated version of a problem set in lectures of Fermi. 0. .12. 1pV 2ro 0 4>o '. Solution: The wave functions i/>o>Vv in regions V = 0. 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. ^v + ^[E.** " W ~^7~ Fig. '.
384. p. 950.621. 369 and 8.055 X 1 0 _ 3 4 J s .72) Assuming a mass of 1 ton of the car (i. 1000 kg) and using the following values of the natural constants: g = 9. Emde [143]. p.17)). and assuming a length of b — a = 100 m of the base of the bump.2 . one evaluates an approximate probability of exp[6.9191 ~ 1. which give /•TT/2 dx sin'M .1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. 14.l .1 / 2 ) ! = v^r) TT/2 /. (12. we obtain ( .807 m s . Jahnke and F.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.71) The integral can be looked up in Tables of Integrals.2. : 2 ^ B ( ^ . M. ^ ) = 2 ^ .1. **For instance E. We find™ (with ( . h = 1. Ryzhik [122]. p. formulas 3.**. S. dx sin ' x •• Looking up the value of the factorial in Tables. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. (15.y) is the beta function (see Eq. .4 x 10 3 9 ]. We can now evaluate the probability.1. Gradshteyn and I.e. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. n I .264 we have ro / dxyfV(x) Ja CHAPTER 12.
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. This is the quantum mechanical version of the problem of Galilei in one space dimension. 265 . in the book of G. the last case allows only a discrete spectrum. Example 14. Then considering the probability distribution determined by the squared modulus of the wave function.1 Introductory Remarks We distinguish here between three different types of linear potentials. for instance. * A highly abridged treatment of this problem is given. 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. Siissmann [266]. 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. p.3.* In Chapter 14. 13. Thus. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves. we consider the corresponding case of a particle above the flat surface of the Earth. Finally. whereas the first case permits only a continuous spectrum. This is the problem of Galilei in quantum mechanics. we consider the linear potential in three space dimensions that forbids the particle to escape. in Chapter 15. each of which propagates with the classical momentum and energy of a Galileian particle.Chapter 13 Linear Potentials 13. 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. 144.
The present case is an exception and a good example of its applicability.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.= l V 27T J_oo 1 dkeikxijj{k. The equation of motion follows for instance from the derivative of the constant energy.1). (13. V(x) = —mogx. We see " that the case g < 0 can be related to the reflection x — —x. (13.1) where g is the acceleration due to gravity and we can put Vo = 0.t) = .t) = . (13. 13. i. Therefore we use the Fourier transforms il){x.t). A different > "linear potential' is the socalled "confinement potentiaF V(x. (13.2) which is the case with a discrete spectrum.2. and yields rriQX = mog. (13.t). g > 0. x > 0.266 CHAPTER 13.= l V 27T Joo dxelkxip(x. from d_ m x2 0 dx + V(x) 0. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle.t) = mogx + V0.t) = m0g with potential V(x.4b) . We consider here the case of Eq. Recall the classical treatment of the freely falling particle.3) In general the momentum representation is of little importance. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.4a) f°° ${k.e. c> 0.t).t) = c\x\ + V0. (13.
2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.e. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i. d \ ~f1 dkr[ k + m0gt ^r't (13. (13.t) the expression (13.9) .7) applied to every quantity in Eq. (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.< (13. To this end we consider the following substitution in Eq.7) In the wave function on the left hand side of Eq. Equation (13.13. i.6) and wish to solve it. we obtain the differential operator on the left of Eq.t).e.3) yields us now by inserting for ip(x.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13. Applying the total derivative d/dt to this function.t)im0g^k.8) With this result and the substitution (13.d m +imo9 .6). We rewrite this equation in the following form th (13.t). (13.t) m0gx /2TT m0gxip(x. • fd ~(' m0gt %h . multiplication by a test function and subsequent integration).t) dk nikx k=oo '2K (im0g)il)(k.6) the argument k is then replaced by this. (13.t) h2k2 d ~J(k. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt. (13.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.t) dketkxiP(k.5) m+tm9Wcr{ktt)=2^^ktt)t (13.
6).i) .1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.12) . Our next step is to reexpress the solution (13.6) to the solution (13. the following result h 2moi tp[k\ ^—.= k rmt.6). Thus addition of both indeed leaves the expression on the right hand side of Eq. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k. (13.10) Note the amplitude function ipo(k) in front. (13.t ) ' U ( f c . we have to reverse the substitutions and replace in (13.10) k by m0gt back : k —> k h The solution of Eq.268 CHAPTER 13. We have thus obtained the solution of the transformed equation.t).6).t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. (13. i. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.11) solves the partial differential equation (13. To obtain the solution of the original equation (13. (13.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc.11) in a neater form. E x a m p l e 13.t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.6). On the other hand the second operator expression implies dtj>o imog^—ip(k.e.t In Example 13. We define the wave number kt as kt. (13.gh [ d t ' j f c + ^ ( t ' . Thus we have first (the second term arising from the upper integration limit. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. t ) .11) *[k?f. Solution: We apply the two operator expressions on the left of Eq.11).1 we verify t h a t this solution indeed solves Eq.6) is then $(k.
/„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.14) (13. _ k t  h2 3 Inserting here the explicit expression for fct. 27T ^ .o o 2 —  exp iktx h mZgH' .— [It(k) . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.11) can be written j'A ^. .17) = J_/ C X ) h2 [It(k) .} fA ' _ _tkt 2 k+ t) 2 k . (13. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .I0(k)]. we have ip(x.19a) .t) into this integral and recalling the property (13.16) Next we evaluate the Fourier transform of ip(k.+^f (13.= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.t) defined by Eq. m0g (13.o ut = 2mo —kf 2mo A.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.4a). .15) we obtain / = .2„2+2 mfaH 3^3+3 . (13.t) = .t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . Inserting ip(k.gfci + 2 g2m0 2/i 2 ' (13.11). we obtain rp(x..18) h =k mogt p h S' h .13) rn0_t+2 + mlg2t3 .13.
7 u>t'dt' (13. With the help of the de Broglie relation this wave vector corresponds to the particle momentum Pt = P — rnogt or to its velocity vt = v — gt. To this end we convert Eq. fm0gt\ \ 2h ) (13.v = kh/mo.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.t) =y= dkip0(kt)) exp i< fctx V27T J . Such a decomposition is possible only for homogeneous force fields. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.t) as a decomposition into de Broglie waves.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .20) expresses the wave function tp(x. 13.2.19b) h3 3 we can rewrite ip(x.18) into a • different form in order to be able to perform the integration with respect to k. (13. (13.The associated kinetic energy of the particle is mv+ = huJf. Linear Potentials f dt'.o o Jo This result can be interpreted as follows.ut> = Jo t2 g2m0 i 3 k t .t)\2 for t — oo (distant future or past).19a) exactly as for a falling particle. (13.20) (13. the relation (13. These wave vectors vary in accordance with Eq.21) In other words.t) as 1 I00 ip(x.270 and CHAPTER 13.23) . 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.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 .
e./m 0 a. /m0gt\ \ 2ft J ° ) I m09H3 ft J 6ih m0g2t3 _ m0x2" 2Aih 2hti m gt IT / dk^oik t) exp oo ht k2mo« ~fti 2ft~] (13.18) thus becomes ip(x. integrate and then take the limit fi —> 0). rnogt m0g2t3 tX^r~. m0xcl ht m0gt 2h gt O v: °° 2hti (13. fi > 0.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.t) = 1 1 oo 271 27T /.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ .13. (13. (13.t) 1 exp r . It follows that ip(x.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.QX 2 ^ 0 7 (m. may be put in front of the integral.26) with the formula 1/2 (13.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. (13. We evaluate the Fresnel integral in Eq. being now independent of £. Then ip(x. TOO#A ft and.2 The Freely Falling Particle: Quantization The integral (13.26) d£e ie £ kt^ m0' 0. 2h h 24ift ' mogt^ r m. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f./oo dkip0(kt) exp l x ixI exp oo .28) (for a verification replace a by a + ifi.29) .24) We now set fit / 2m 0 V" 0 V m0a.
30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.dA 4>o(h (13.1 A turning point at x = —XQ. (13. V(x)=m 0 gx V=E Fig. For stationary states we set xl){x.e. 1/2  ±^(2mlg/h^x^ / 2i (13.1) has only a continuous spectrum.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 .33) oo for E <C \mogx\. xc\). For x exponentially increasing and decreasing solutions are possible.272 CHAPTER 13.t) ~ W —exp i V irlt m ktxd J cot.31) and obtain from Eq.3 Stationary States In this section we show that the Schrodinger equation with the potential (13.t) = eiEtlhct>{x) (13. kt and ut 13. 13. Linear Potentials Replacing here x by the large time value x c i. this expression can be brought into the following form: if *\ 0 ip{x. (13.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. . These are incoming and outgoing waves.
cf. t).34a) Then d &k 1 f°° ih—il>(x. and hence is continuous and extends from —00 to +00. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. is reflected. i. i.6) that also for 4>{k.36) m0g \ 6m0 with c = const.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.00]. exponentially decreasing. t) = Eil)[x.4b).4a) and (13. Fig. H2k3 4>{k) — cexp Ek (13. is therefore called a turning point Quantum mechanically this is a point at which periodic solutions of the Schrodinger equation go over into exponential solutions or vice versa. 13. The Fourier transform is given by the relations (13. so that ih dtp{k.35) It follows that /W tm0g \ 2m0 i.t).0 0 dkelkxE^(k. t) = — = / V27T J . i. i. Akx (13. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. but also d 00 dt Eijj(k. at which the classical particle bounces back. i. we .13. J \ „ (13. E)dk.e. its momentum p would be imaginary there.e.t).t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m.1. probability. 1 f01 dketlcx^{k. However quantum mechanically this is possible with a small. The point — XQ. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. at which E — V = 0.e. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. which is not possible for real particles.34b) It therefore follows from Eq.e.e.t) = dt (13. (13.e.
. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).39) we obtain dkcc*exp m0g It follows t h a t cc = &(# . apart from a constant phase which we choose to be zero. (13.. / <>~ikx VZ7T J (13. (13.L dxeikx(f>E(x).38) we have for <j)E{k): OO /"OO / oo (13. </>E(x) = .a : ' ) . Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). / dE(j)E(x)(t>E(x') = ^ s .e.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 .36) into (13. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').37) (eikxdk.£') = <*(#£')• 1 27rm0g' = /o .(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.38). (13.e. V 2lT J Mk) = .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. 13.36) (with c replaced by (13.M *K Joo Setting dt cos ( ts n Jo st).40)) into (13. i.43) y=2^x+2^E.39) dk4>*E{k)4>E.L [ dkeikxMk).41) J—oo Next we insert (13. (1340) i. h? h2 . In a similar way we can verify the validity of the completeness relation.
.45) One can now derive normalization and completeness integrals for the Airy functions. / s M .43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x) s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) r 2^rJ 2W dxe^Qe^^eU*3*3) _^Pi(Sy~tz)ei(s3t3) S(y = j£j**w I ^LJt{yz) 2TT _ We will encounter the Airy function Ai(s) again in Chapter 14 in the matching of exponential W K B solutions to periodic W K B solutions.46) This follows since with Eq./ 2m 0 E (13. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables. 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= ^ . and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z).J * . (13.^x + ^ E (13.3 The TimeIndependent we can write <J>E{%)'4>E{X) = Schrodinger Equation 275 2ir^/m0g J dkexp h2 yk — k '2m20gV~ 3 W i t h the substitution k = [it we change the variable to t.^ . (13. whereas the other is of exponential type. the function 4>E{X) i and setting s y = — = 1 . and we shall see t h a t one branch has periodic behaviour.13. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s.44) is then expressed in terms of the Airy function: ..
43) contains the phase (with x — z): > *(*) := sz .49) (13.51b) . .e.r 3 cos 30 J + i ( sr sin 9 . . i.TT < 6 < 0 for r > oo. (13. 13.f dzJ*W 2TT JC with z = reie.r 3 sin 30 ). (13.e. i.TT < 30 < 0. or n <6 <TT o for r > oo. Thus we must have sin30<O. (13. (13.276 CHAPTER 13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig. o (13.\zz.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 .. We want the integral to exist and so desire that lim e**^ = 0.4 The Saddle Point or Stationary Phase Method The integrand of Eq. Linear Potentials 13.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . (13.2 The contour C for s < 0 with ends in the angular domains (a) and (b)..
Then Ze TeC : V^se±i7r/2 = ±i\fzrs.48).56) 2n(s)^Jc But I daefr"?*"^.55) s < 0. so that exp(±2i9s) — 1 and — s — rl. 13. provided its ends at infinity satisfy these conditions.51b).S )V2 (13. we choose zs = iy/^s. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. i. (13. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. (13. Then the ends of the contour lie in the domains specified by Eqs. (13. oo] as indicated in Fig. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.55) into (13. i.53) Since we shall have the contour C in the lower half plane.51a) and (13. Then 6S = ±7r/2. (13. at which the phase becomes stationary. (13. where the derivative of the phase function vanishes.58) .z3s and V^s + .52) = exp(±i7r) Let s be real and s < 0.13. and we set with a real: z := zs + a (S)V4' (13.45) exists for the contour C.54) i ¥~s)3/2' (13. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.54) where a is the new variable with —oo < a < oo. and the main contribution to the integral comes from the a region around the saddle point.2.57) $"(z s ) = 2% s.e. Such a piece can be found in the neighbourhood of the socalled saddle point zs. 0 = &(z8) = s~zt s = za „2„2i6>.e.4 The Method of Stationary Phase 277 The integral (13.( \ / ^ ) 3 2z. Inserting (13. —i ${zs) = szs .
44) and (13. 8. (13.44) we can convert Eq. Ryzhik [122].45) 4>E(X) ~ Ai(z). This is an important aspect in the WKB method to be considered in Chapter 14. M. (13. as one can see from the following equation^ ri' + Ffz •2^2. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here._ 1 ..32) with the solution (13. 2 s 3/2 .278 CHAPTER 13.2/32u = 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.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). For (—s) — oo the integral converges towards T/TC.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). S. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. (13. Sec.491. cos . According to Eqs. With the help of the substitutions (13.64) tSee I.56) decreases exponentially around a — 0 for real values of a. V^s / 4 V3 (13.42) into an equation with a more appealing form. (13. .s ) V 4 exp \(s) 3/2 (13.2 we show that for s — oo: > Ai(s) 1 . V 2h2 (13.59) In Example 13.60) The reason for describing the point zs as a saddle point is that the integral in Eq.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. > so that Ai(s) 2 v ^ ( .32) therefore becomes dz2 + zd) = 0. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. (13. Gradshteyn and I.
43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).51a) and (13. (13. Thus there are two symmetrically placed real saddle points on the real axis.59) and (13. Eq.13. we obtain f° Joo dp=—[a+ i\a\]da.s 3 / 2  .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. It follows that for s —> oo: Ai(s) ~ . cos  .. * " ( z ± ) = =F2Vi. Then. We have therefore r /*0 poo zs = ±y/s = z± / Jc dze^M = / Joo dze^^ + JO dze^z\ In these integrals we set respectively P with *(z) = #(z±) + i ( z . changing to the variable p.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. Example 13._ . 27T Ic Jc 3 This time s is real and positive.z ± ) 2 * " ( z ± ) . the conditions (13. with *"(z) = 2z. Choosing the contour C to lie along the real axis.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.^ e ^ / V2 4 [°° Jo dae'2^. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .51b) are satisfied. . P Introducing a new variable a by setting p = ai\a\. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2.
Then. 13.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). Jo Jo \ . ^/2KJW1. z o = n . i. Thus here / ( z ) = — z + n l n z .111). one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en.Be 2 Fig.e.3. n!= / e~zzndz= / e~z+nlnzdz. since f'(zo) = 0 and (z — ZQ)2 = —x2. (11.3 The saddle point at z = n + ix. ~ 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.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. and hence integrating as described above through the saddle point indicated in Fig. / " ( z o ) = —n/z2 = —1/n. Solution: Briefly. .*. 13.280 CHAPTER 13. Linear Potentials E x a m p l e 13. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix.
The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. i. Jeffreys [144]. and the role played by h in our considerations here. 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. 292 . 316 . the linkage of solutions above a turning point to those below. where *G. Kramers [153] and L.e. of misunderstandings and other aspects may be found in the monograph of Dingle.317. ^ The central problem of the method is the topic of "connection formulas".g. Dingle [70]. B. for misunderstandings pp. For the history see pp. § See e. *R.295.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs.* The method had. Froman [99].e. O. named after its first promulgators in quantum mechanics: Wentzel. A. H.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. 281 . Brillouin [37]. A critical overview of the history of the method. Wentzel [282]. however. It goes without saying that such comparisons are very instructive.Chapter 14 Classical Limit and W K B Method 14. Kramers and Brillouin.§ 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. ' T h e most prominent earlier expounder is H. will there be that of a coupling parameter. N. Froman and P. been developed previously by others in different contexts. The WKB result is therefore frequently described as the semiclassical approximation.
g. For stationary states ^/.1) (14.282 CHAPTER 14.V(r)]^(r) = 0.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. (14. 14.t)\2 = \iP(r)\2. In this equation we set. for which the observation of position and momentum is independent of the time at which the observation takes place. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).7.3) .t) = eiEt/h7P(r) and \V(r. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . Classical Limit and WKB Method E — V changes sign). It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition. the function S being called phase function. the derivation in Example 18.* We illustrate the use of this condition by application to several examples. V>(r) = A(r)eiSWn. the energy has a definite value E with V(r. • {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. A(r) and S(r) being real functions of r.
7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.1) and separating real and imaginary parts.t)}2.10) m 0 ^ .Q A 2i 8A 1 (14. V(r.12) VA A (14. h2 ih— = H$ A + V(r) * . (14.4) (14.t)eiS{r't)/h. we obtain the following two equations: E .+ VAVS + 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. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.* VAVS+AAS + V(r)A. (14.14.V(Y) + and 2VA • VS + AAS = 0.13) . (14. These equations are equivalent to the Schrodinger equation.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. We can identify Eq.6) 2mo dt now with A also a function of t. h h (14. We set in this equation.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *. (14.
15) and (14. (14.20) With this we can rewrite Eq. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real.9) for the phase function S. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.. the implications of this equation are also those of the equation of continuity.17) With Eqs. (14. TTIQ (14.+ AVA • VS + A2AS = 0. + h2 AA V = — ^ .9) as 9S 1 mo^ 2 + T. (14. we have mo—(A2) + V(A2)VS + A2AS = 0. ^ + V • J = 0.10) we obtain = — [V(A2) • VS + A2AS}. Next we investigate the significance of Eq. m0 But now (14.10). Since J = — A2VS = PVS.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. (14. Since we obtained this equation from Eq. (14. (14.16) BA 1 moA—. (14.21) . (14.14) A2VS.J = 0.18) ot ot This is the equation of continuity that we encountered already earlier.16) this becomes m 0 1 rr + m o V . ie. (14.01.284 CHAPTER 14. . mo <™> (14.n. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS.
d — = (14.20). dt 2 Constructing the gradient of this equation. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. r o t v = ( l / m o ) r o t g r a d S = 0.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ.e. This means. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.V ) v + V 7 = 0. which for very small values of h yield the dominant contributions.23) (14.2) for the probability amplitude. . '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.e. This result is very general. In the first • place we require the ansatz (14. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. Such series are typical for expansions in the neighbourhood of an essential singularity. This expression with h in the denominator shows. and hence''' ^(m 0 v) + m 0 (v. the wave function \I/ describes effectively a fluid of particles of mass mo. V • J = 0. a singularity different from that of a pole). where. we obtain ^ V 5 +  v ( v v ) + V V = 0.22) (14. We deduced this equation solely from the phase S in the limit h — 0. However. so that V ( v • v) = 2(v • V ) v + 2v x rotv. they are asymptotic expansions. We see therefore that in the limit h — 0 the particle behaves > like one moving according to Newton's law in the force field of the potential V.14. The particles have no interaction with each other.24) d 1 dr d •— = d Lv _ .1/h. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. (14. that the quantum theory implies an essential singularity at H = 0 (i. The equation of continuity then describes the stationary flow of a fluid. (14.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). i. with the help of Eq.
is an asymptotic expansion. (14.3. such as "LiouvilleGreen method' and "phase integral method'.4) and (14.30) ax *For details see R. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).e. i. z (14. There are also other more specific descriptions.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.26) We observe already here that an expansion in rising powers of k. Sec. the method is widely familar under the abbreviations of the names of Wentzel.In A(x). .27) (14. We set. or h2/2mo(E — V). As remarked at the beginning. §We follow here partially A.5)) and 2 ^ f dx ax + All=0.28) The calculations we performed at the beginning imply the following equations (compare with Eqs.286 CHAPTER 14. (14. 318. 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.[EV(x)}y = 0. Classical Limit and WKB Method 14. We consider the onedimensional timeindependent Schrodinger equation y" + ^.3 14. i = eiW(x^h. p. similar to our earlier procedure with A and S real functions.2. Kramers and Brillouin.$ 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. B. Dingle [70]. 6.§ y = A(x)eiS^'h where W(x) = S(x) + . Messiah [195]. (14. The expansion in powers of h2.
.2rn0(E V) = h2 lS'A 2S'0 (14. Li This expression can be substituted into Eq.29) becomes (note that this is still exact!) (S')2 .28).A = const. (14.35) can be integrated. (14.32) is therefore (S'0)2 + 2H2S'QS[ .\ In S'.10)). But first we note that: A dA dx2 2 = c{S')ll\ = A 15^ 2 S' + ^ = A ax 3 (S") 4 (S')2 2 2 S' With this Eq. (14.V) and (14. (14.30) corresponds to the equation of continuity (i.4 (S>)2 1 S'" 2 S> (14.31) \a.32) s" = sz + h2s'( + •••. Since the expression (14.3 The WKB Method 287 Equation (14.2mQ(E . Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .14. 3 / 5 .35) . Eq.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E . (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq. + (14.33) 3 (S")2 .e.36) Equation (14.
41) We now ask: What are the conditions of validity of Eqs. (14. (14. .I In S' + In c h 2 2 2 exp ^{S0 + h Sx) .288 CHAPTER 14. Chapter XIII.and So = ±h / — — A J k(x (14. it is reasonable to set (cf.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.39) where a and /3 are constants.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . Dingle [70].' For the expansion (14. (14. (14.37) For Eq. we expect that in any case So » ft2Si. Classical Limit and WKB Method has the form of a momentum. p = h/X) k{x) :Then S'n +. y/2mo(V(x) . (14. The solution is seen to be periodic provided k(x) is real.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2).38) h y/2mo(EV)' (14. i. Eq. 1 S e e R.39) and (14.27) we obtain now with Eq. (1.33) to make sense as an asymptotic series. and instead is asymptotic which we shall not establish in detail here. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.19).e.41)? We note first that the expansion in powers of h2 does not converge. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13).40) rx dx v "7 W)\. B. E > V(x).
E > V(x).14.42) [(^). Eq. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .45a) . Since k(xy 1/52\S'0 (14.36). (14.V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (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 (14.3/2 s&T + 3 {So Il« we have (cf.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.3 The WKB Method We have 289 S0 = ±h and Si follows from (14.1 / 2 ]" [§0SS).
V(x) = V(a) .46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. (14. In terms of X the original Schrodinger equation is y" + = y = 0.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l .48) The (dominant) WKB approximations are exact solutions of this equation. How do the solutions of Eq. *2 A(z) = V / 2m 0 (£ . the WKB approximation is in general invalid (see below). (14. In a similar way we have for E < V(x): l'(x) <C 1. Roughly speaking one therefore requires E to be large in both cases. i. V(x)) (14. (14. h = 1) .44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. In the neighbourhood of E = V(x).V{x) ~ Or . Classical Limit and WKB Method we can rewrite the inequality (14.a)V'{a). and hence (with UIQ = 1/2. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.48) behave in the neighbourhood of such a point called "turning point'.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). around the classical turning point.e. 14.290 CHAPTER 14. at which E — V{x) changes sign? We note first that in this case near x = a: E .3.
On one side. 14.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. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. We shall not derive these conditions here in detail but rather make them plausible in Sec. since this requires different approximations there. As indicated in Fig.1. The transition is provided by matching relations.1 there are WKB solutions on either side of the turning point and some distance away from it. 14.e.1 The regions around the turning point at x = a.48). in the domain E < V the WKB solutions are exponentially increasing or decreasing. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . (14. i.1. 14. 14. (14.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x.3 after stating them first here. in the domain E > V the WKB solutions are oscillatory. It is therefore necessary to find other solutions in the neighbourhood of x = a and then to match these to the WKB solutions in adjoining domains as indicated in Fig. which connect one domain with the other.3. 14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a.14. V=E Fig. and on the other side.
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.292 CHAPTER 14.52) For a potential rising from left to right. in order to make the formulas independent of whether the potential is rising or falling. has the same asymptotic behaviour in the domain i « a a s Ay\. as in Fig. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi.H— J =^ v/exp + m (14. Thus. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. 14. (14.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . This means the asymptotic part suffices only in t h e exponentially decreasing case. Note also the extra factor 2 in yi. a interchanged. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. one could simply replace everywhere /•a / pa • • • b y / Jx J x .50) x S> a. 14. (14.1) 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 / —. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . however with the limits of integration x.
(14. 1 ~ (x .48). One should remember that the matching relations above are those for the dominant terms in the WKB expansion.3 The WKB Method 293 In that case Eq. 14.53) Since at x ~ a.14. the matching relations have to be altered accordingly. (14. the equation becomes approximately y" = (xa)xiy. 2 1 X (14. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.54) .3.2 The overlap regions around the turning point at x = o. i.Airy overlap* Fig.e. the equation V approximately linear around turning point V=E Airy solution WKB .3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. (14. is singular at a turning point x = a. as we saw. whose exact solutions are the dominant WKB approximations.a).46). y" + Toy = o. If higher order contributions are to be taken into account.51) — always with the decreasing exponential on the left — is valid in both directions. Therefore we consider now this original equation in the domain around x = a. 14.
A second solution is written Bi.5960.4. (13. In the following we require the asymptotic behaviour of the solutions Ai(—z). (14. i I 772 \3 + > for ia>0.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0.57b) UV2 3 11 by J dx . (14. With the substitution z = (x . This is given by the following expressions. .56) We have discussed some aspects of solutions of this equation in Chapter 13.57b) To the same degree of approximation as the equation y" = (x — a)xiy. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. (14.291.1 / 4 by \^. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~.a)3/2 ^ /2 =^ 3/2 o Q. B. (14.57a) and (14. or M. (13. Abramowitz and I. (14. — a ^> 0.59). p.55) the equation becomes the Airy differential equation d2y dz2 zy.Bi(—z) for \z\ — oo. Cf.a)X\/3. formulas 10. p. J (X (i4 58)  Replacing in Eqs.294 CHAPTER 14. Dingle [70]. 448. R. > for a. Stegun [1]. In particular we encountered a solution written Ai.^ 4 I 7T x — a <C 0. as > may be seen by referring back to Eqs.60). A. we have r x * f x ' /2(x a)i/2d*=^(x .
At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. 14.1: 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 / — = . n = 0. e.3 The anharmonic potential well.14.(zf2 by J we obtain the matching relationsX (14.59a) .52). As an illustration of the use of the WKB solutions we apply these in Example 14. We summarize what we have achieved. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation).2.1 to the quartic oscillator. by Ai(z). and hence must be proportional (in view of the uniqueness of the solution there). As we approach a limit of this interval in going away from the turning point.3 The WKB Method and 295 •. Example 14. we enter the domain of validity of a WKB solution. are only valid in a small interval around the turning point at x = a as indicated in Fig. In this way different asymptotic branches of one and the same solution are matched across a turning point. (14. In the direction to the left as indicated in Fig. The latter domain is a small region where the solutions from either direction overlap.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.g.l. These solutions.l v/2m0(EV(x))dx 7r = (2n + l).2. however. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig.. In this sense we have now verified these relations.51) and (14. 14. 14.
extends only over a length of very few wavelengths and is therefore valid for n small. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. regions 1 and 3 in Fig.296 CHAPTER 14. ^ A = (2n + l ) .61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique).59b) Solution: We consider an anharmonic potential well as depicted in Fig. (14. (14. Sometimes this distinction becomes imprecise. i.60) Using the matching relation (14.. 2 n = 0. Classical Limit and WKB Method where x = a. 14.l. . 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. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule.(x) = c ' v Acos I / — I for a < x < b. 1 . (14.3.e. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. . The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained.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 . Here we are interested in the discrete states (the only ones here). . 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 ( . We argued earlier that the WKB method is suitable in the case of large values of E. . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. Hence we search for solutions which are exponentially decreasing in the far regions. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J.3. n = 0 .2. (14.y o dx TT T provided 6 _ 4 /.51). The condition for this is ya{x)=yb{x) for xe(a. 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. 2 . whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. According to Eq.44) this implies large values of n.^ . as in the case of the harmonic oscillator itself. /n = s u ^ TJcos^ya TJ+cos^a T J . 14.. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. j/i. (14. . In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a. b are the two turning points. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .b). Then around these points the potential would be well approximated by a linear potential (as we saw above). Let E = V(x) at x = a and x = b.
1. Wilson by supplementing classical mechanics by Planck's discretization. This relation is sometimes wrong. 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.4 Three cases with different pairs of zeros. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. 2. Bohr.4 BohrSommerfeldWilson Condition 297 14. equivalently.e. 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. and knows of no published form or script. n = 0. In this socalled "old quantum theortf (i. . with n — 0.63b) *The author learned this in lectures (1956) of R. . Sommerfeld and W. Dingle.x number of turning points TTH. ri2 I Jqi :pdq n + 1 — . (14. B. A. .1.64) (14.63a) or.x number of turning points h.14. The corrected form* is. (14. . 2 . . pdq = n + 1 — . (c) V Fig.
where n —. 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. if /(g) is negative.e. There are three cases. Since sine and cosine have zeros spaced at intervals of TT. 0(g) ex . as illustrated in Fig.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. i. (14. 14. Before we continue we recall from Eq.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0.4(b). i.66) J<1\ Thus in this case the socalled old quantum theory is correct. this is an exponential zero.e. We have the case of trigonometric solutions. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. 14. In quantum mechanics we have dq2 + Ki/j(q) = 0. / • n an integer > 0. the case of tp(q) — 0 at q = gi.e. this means dqyf—f = (n + l)7r. i.65) We use the knowledge of these solutions to find the eigenvalues of the equation. i. (14. where /(g) is negative. where /(g) is positive.67) Case (b): Next we consider. Jqi ^ fpdq J =2 pdq = (n + l)h. together with q — — oo. this is a trigonometrical zero. Case (a): The case of two trigonometrical zeros. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. In this > case the wave function for g > go is . (14.4(a).298 CHAPTER 14.
^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. ri\ / J do y/f(q)dq (n + 1) 1" (14.2 and 14. Thus in this case the socalled old quantum theory is wrong. (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. 14. or Pdq (n + l ) " 2 h.68) (14.4(c). Solution: As in the case of the quartic potential.14.e. (14.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa .3. this is the case of two turning points.e. (14. i. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.63b) to obtain the eigenenergies of the quantized harmonic oscillator.69) pdq = (n + l ) " 4 h.63b). i. E x a m p l e 14.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). As illustrations we consider Examples 14. Therefore their arguments differ only by (n + l)ir.70) We can therefore summarize the results in the form of Eq. 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. For the wave function to be exponentially decreasing to the left of qo and using Eq.63a) or (14. Hence we have pdq= i n +  .2: The harmonic oscillator Use the relation (14. and • • for the wave function to be exponentially decreasing to the right of q'0. the sines must be proportional. 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 .
2 . The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . whereas a turning point does.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 . .300 With potential V(q) = 2ix2mov2q2. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it.28 x 1 0 . 4 / x CHAPTER 14. and h = 6. Calculate its quantized energy. in principle.807 m s . Example 14.63 x 10~ 3 4 J s . g = 9. where g is the acceleration due to gravity. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo.V(q))dq The result is E/u. 14.71) For m 0 = l x 1 0 _ 3 k g .5 A particle above a flat Earth. so that E = hu{n + 1/2). permit this although with rapidly diminishing probability. :: V=m0gq Fig. 14. Thus there is one turning point at E = mogq. as illustrated in Fig. one obtains E ~ 2.5. 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.
5 Further Examples 301 14. d dE V)={n+)n.( 1 — = — = it — = h\ n H — 2TT OJ 2K J p dE V 2 In the case of the harmonic oscillator we have hi n E (14.73) Jx! rp2a dx I^<E 2 v) = ^ n — = . E x a m p l e 14.14. 9 .5 Further Examples E x a m p l e 14.72) \\2dxp=Qdx^{Eh Hence. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). In the case of screened Coulomb potentials (cf.7.4: 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 £ = . h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .= — <*— = h(n+\. Chapter 16) one obtains the quantization relation (for Mo = const.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 . with p = modx/dt. ra 2/ = 0. and u> is the oscillator frequency in either well.l.2.75) which verifies the formula immediately. Solution: The proof is contained in Example 18.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points. 1 .f V ft h JXl h 2TT (14.5: Period of oscillation between two turning points Use the BohrSommerfeldWilson quantization rule to obtain for the period T of oscillation of a particle of mass mo between two turning points the relation T 1 m0 f dx d ( — = . (14. (14.
Eq.<?2i<?3 for E > 0 given by E — V = 0.a . q = y+ a . (14. i.6.6: 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. which determine the imaginary part of the energy. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.77) Determine (a) the turning points. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.y2)(y . 14. The equation E — V = 0 then becomes after a few steps of algebra y3 ~ Py + Q = (y P = ^ a 4 and Q = 2 ^ yi)(y .e.y3) = 0. 14.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. The q term in this equation can be removed by transforming the equation to a cubic in y.~ \ (14. (14. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2. (11. Fig. where . Example 14.78) (a) We determine first the three turning points at qi = <?i.79) .133)).302 and T CHAPTER 14. where 1 2 • 1 2 y = Q.6 The cubic potential.
6. 212. H e r e cos(<9 ± 120°) = .k 2 2tan# \/3 + tan! . B . V cose' + sin 8 (14. 0<6O°.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0.g i < 1.9 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 .< 1.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .' : ± 6 0 ° w h e n E = 0.c o s ( 0 + 12O°) + c o s ( 0 . q3 ~  . B y r d a n d M . 92 = a2[l . 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.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. 78: sin 120° = v ^ A c o s 120° = .{ V 3 e } > 0. p . 14. tan60°=%/3. 80 a n d 361.a 2 [ l .78)) c o s 30 •• 4P3 ( ' \ aeb2} .2/3 i n t o q = y + a 2 / 3 . 91 < 93 < 9 < 92(14. (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. q3 = a2[\ . (14.gi = 1. § T o h e l p we cite H. D . (14.3/2.\tan0^ v/3. p . e > 0. _ \ / 3 — tan(9 2 V92 .93). 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 . ' P . a n d t h e a n g l e 0 = 0° t o E = £ m a x . (14.04.= s i n t 3 v 3 (14. V3 + t a n 0 2 .83) T h e i n t e g r a l m a y n o w b e e v a l u a t e d w i t h t h e h e l p of T a b l e s of I n t e g r a l s . F . M .2 c o s ( 0 . p .84) 92 .81) m a 91 . i.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.T h o m s o n [196]. p .9 2 [ .79) a n d t h e n s e t t i n g m o = 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 = 1.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . .1 / 2 .120°)] = .q)(q .9 3 — = = 2 " 3 " [ . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . W i t h t h e o t h e r i n t e g r a l from T a b l e s we o b t a i n (K(k) b e i n g t h e c o m p l e t e *L. we h a v e qx = .14. ui=sn 1 \l— — = l. we set 0 = 60° — e.c o s O + cos(6» .0^6O°. gi ~ y {V3e} < 0.g i ) ( g 2 . g 2 ~y{3}>0. (14. M i l n e . 6 = 0 i m p l i e s E m a x of E q .80) I n s e r t i n g t h e r o o t s yi.93 2a2 [ .120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . 37. (l_fc' 2 a + fe'4)l/4.2 c o s 0 ] . D w i g h t [81]. f o r m u l a s 236.9 l ) f f / du s n 2 u en2 d n 2 u . 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 .e. 92.cos 0 + cos(6> + 120°)] = a2 92 .2 c o s ( 0 + 1 2 0 ° ) ] . F r i e d m a n [40].ui=K(k).08. 93.
92) ''P. as in (14. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . D.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.91 F(TT/2.91 .86) 15 tan. Byrd and M. 72. sinV = q3 qi ~ = 1. (14.k).fc'2 + fe'4)"1/4. (14. 2 2 .2(fc2 . Eq.qz) (14.91) a2[cos0^] y/3tanfl ? L .00. V?2 . E(u = 0) = 0. (14. p. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . Whereas the integrand of the above barrier integral is effectively a momentum.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . Next we evaluate the required integral across the well from 91 to 53 at energy E. . (24. Friedman [40].?)(?3 . formula 233. G(fc) =86aS (ITS£ li[fe. F.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .85).S .84) we obtain Carrier = 2a 2 sin6> 2 b( .) . The WKB Method elliptic integral of the first kind.~ y ^ ) a< [ COS 5 + . 91 < q < q3 < 92 bJc. g= 2 .^ V3 / V V3 y a / c o s e J. 1+72 = and J o s 0 + ^ = (1 . 0 (14.= —^ = k a2[cos0+^] v ^ + tanfl V=".304 CHAPTER 14. E(u = K(k)) = E(k). with K(k = 1) = oo. the integrand of the integral across the well is effectively the inverse of this momentum./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. We can evaluate this integral again with the use of Tables of Integrals.87) Inserting G(fc) together with the expressions for the prefactors into (14. V 93 .32)).2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it.^ + ^ ) (l + fc'2)2 .sm t %/3 £\. Thus Iwen=lgF(ip. E(u) the incomplete elliptic integral of the second kind. A:') = # ( * ' ) • (14. . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .
99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. (14. (14.88) in rising powers of fc' .5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14.63b).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. formulas 8. .. we obtain E a6b2k'4(l + k'2). (14.96) Expanding the coefficients of the elliptic integrals in Eq.3 and 8. so that even in the case of the ground state the zero point energy will contribute to the prefactor.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. 2a 5 b K(k) + 2 E(k). in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. 32 (14.4 2n + 1 fc' = 4 — ^ — .** However. 906.85) (14. (14. Gradshteyn and I. . Ryzhik [122]. M. 905. pp.93) where w is the harmonic oscillator frequency in the well. a°b n = 0.2 IE 4 fc' ln 8 (" «V iU^fc' 4 .113. (14. and hence at E = 0.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .114.14.3. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1.80). 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.l. we obtain > . To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). we obtain 15 .2. Thus one obtains from (14. to represent a physical decay rate. With q = 1/z the integral (14.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. S.
i. this has not yet been done.96) for the imaginary part of the energy as in the Breit. 2n + 1 — / 26a5b \ .„.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. /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .7 . so that we obtain 4 . We observe that with \j2je as 1. (18. V (14.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.178)) call the Furry factor set equal to one. (14..e. .— (14. It follows that (for one complete orbit back to the original turning point) . also Eq.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) .73).100) 'barrier = ~0°b In [ ^ — j .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. _ exp[2/barrier] >.j . .101) (14..> tf = = 2/wfiii = ±tv8a°b——P.ba / 25a5b\n+2 ±l_l eT5 a f >. (0) i hw En = E\> . The WKB Method For k' —• 0 the first and the third terms dominate... the result agrees with the ground state path integral result (24. 8 Bfc .J + 0(VoJb).Wigner formula (10.e. 15 a . oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s. _ T T • (14. +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . 7 ~ —.2 7 b a r r i e r ] Z Z7T where (cf. (20.fc'2 + fc'4)1/4 * ± t ^ . i.102) .e x p [ .306 CHAPTER 14./^~^l^ab ^aub . n ~ V27T .
Yan [82]. after the realization that quarks as their constituents might not exist as free particles and. K. The study of this potential. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. at least indirect. C. in fact. B. Gottfried. the classification of the various states of nucleons and mesons and other particles. Eichten. C. describing free fall under gravity.g. [231]. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. [233] and H. Thacker. There are numerous. now known as quantum chromodynamics. along with inclusion of angular momentum and spin effects and their interactions. led to a spectrum which contains only scattering states. which permits only bound states. Quigg and J.1 Introductory Remarks The particular linear potential we considered in Chapter 13. See particularly E. This potential became widely popular in the spectroscopy of elementary particles. L. Rosner [230]. i. the Coulomb potential. that this is indeed the case.Chapter 15 Power Potentials 15. [232].* 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. D. K. Kinoshita. T. Rosner [269]. 307 . L. 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. Lane and T. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. Quigg and J. indications from the nonabelian generalization of quantized Maxwell theory.e.M.
r acting between two particles is constant and directed towards the origin (of the relative coordinate).6) = (I + l)u(r)/r = R ~ rl.4) Instead of considering the linear potential. h2 2n V 2 * ( r ) + [V(r) . (15. This force maintains this value irrespective of how far apart the particles are separated. m\m<i M= : .1) i. </>)R(r). We leave consideration of the logarithmic potential to some remarks and Example 15.308 CHAPTER 15.e. R(r) = ^ .2 The force The Power Potential F = . f dr{u(r)}'2 1. 15. and the normalization [ dr\y(r)\2 = 1.m2. (15.3) with the boundary conditions''' u(0) = 0. ~ (I + l)rl (15. and so u'(r) . v > 1.£ ] * ( r ) = 0. In order to have a clear starting point. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r).1. These particles therefore cannot be separated by any finite amount of energy. mi + m2 and r is the relative coordinate. Power Potentials general power potentials. where \x is the reduced mass of the two particles of masses mi.4>)ru{r) = Ylm(9. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. i.V 7 = const.u ~ r i + 1 (/ + l)R(r). A > 0.e. and u'(0) > [u(r)/r]0. For a central potential V(r) we write (15.2) tt(r) = Ylm{9. we return to the Schrodinger equation in three dimensions. (15.5) u'(0) u[r) r=0 = R(0).
V(r)) = (2nl) +1 N.8a) we can write the quantization condition (14.2. In order to to obtain the . (15.63b): 2 and V(0) = 0 (15. (15. 3 .2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.1. odd integer ^=(nll)vh.9) into the threedimensional case in the following form: / Jo C dry/2n(E . x > 0. (15. and the turning points are at a — — b := — xc. This means that we transcribe Eq. is to be interpreted as the principal quantum number in three dimensions. which has been reduced to an effective onedimensional problem in the polar coordinate r. Then k(x) = k(x). 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.8a) ^ JO = (2Ar + l ) £ .4) in the threedimensional case."MA. (14. (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.8b) This condition implies that we have only one pair of turning points.. (15.10) Here rc is the turning point and n — 1. d5.2.15. so that with Eq. We also assume that V'(x) > 0.. We therefore proceed as follows. .7) k(x) = h/y/2m0(EV(x)). We now want to find a quantization condition for the threedimensional problem. or f l H " ^ ) ! . . N = 0.63b) for the onedimensional Schrodinger equation.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!).63a). (15. ..
dr = .12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.15) I ' = ^ 1 (( 1) B{x.11) rc = (15. s.13) we obtain _ X_ „ _ so that A£_ (15. (15.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .310 CHAPTER 15. (1516) (15.17) The integral on the right is the integral representation of the beta function B(x.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. (15. _ —vr" *dr. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.y) defined by (xl)\(yl)\ y ^ (x + y .14) Prom Eqs. Power Potentials eigenvalues E —• En for the radial potential (15.12) and (15.'(1*)1/2= J1F\lt)y1dt.13) dt . c (nj\irh. (15. . E A fv.\rvc = 0 . Now.
55).2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. One can convince oneself now. (15.15.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . The case v = 4 is that of the pure anharmonic oscillator also discussed by M. ^ '"Compare with Eq.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. and n — — ]7r/i.1. (12. 15. $T(z) = {z.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.1 Approach to square well with v —> oo. for instance simply graphically as indicated in Fig. i r ( i ) = r ( l + i ) = ( i ) ! . Weinstein [281]. . T(z + 1) = z\.1)!. 15.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.
For the following we set in Eq. But for our present purposes this applies only in as far as the solutions (j)(x) which are exponentially decreasing at infinity vanish at x = 0. the harmonic oscillator. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. Power Potentials In the case of v = 2.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15.21) (4nl).25) "Comparison with Eq. In the case v = 1. Physically it does not make sense for the probability amplitude to have a discontinuity there. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.> r ( 2 ) A V 2 in \)*\W \it^T& (15.oj = ^/2X/JX.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.g — —2. we obtain E„. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. x > 0. 15. since these are the ones we are interested in (cf.2) implies A s fj.41) for the . > and with Eq.61) h2 = I.1 + \)hx> = {An • l)£fiw. the case of the linear potential. (13.2.312 CHAPTER 15. The potential V(x) = X\x\ (15. (15. we obtain En = (n . eigenvalues (JV + h)tkj — (2n .mo = \. 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).24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). discussion at the beginning of this section).oj2/2. (6. (6.
e. s = E.26) For z = s := E — x —> —oo.x). E > V or s > 0. i.56) and (14. Eq.e. Ai(s) "v^ G ' cos s 3/2 _ t (15.60) applies. (13. (15.27) i.x)\x=0 = 0. (15.15. Ai(s) 1 2v^(s)V4 exp x —> +oo. 15.e.29) .57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . i.2 Behaviour of an odd wave function at the origin. dz2 According to Eqs. \(s?» (15. In the domain —a < x < + a .28) In particular at x — 0.59) applies. (13.e. (14. this wave function has the required exponentially decreasing behaviour at infinity. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . and for s —> oo the trigonometric behaviour of Eq. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. i.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig.e. i.
(2/3)E3/2 or 3 / 1 2/3 E — En — . of course. 201.e.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. p.08795 5.08181 5. *This Table is reprinted from C.52056 6.3..= ) = «..82878 1 2 3 4 5 6 7 8 9 10 2.02265 10.1 demonstrates the quality of the WKB approximation by comparison with the exact values.94413 9. The expression (15.03914 11.02137 10.04017 11. the eigenvalues En are determined by the zeros of the Airy function. i.94249 9. Table 2.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .00852 11.82814 i.. (15.32025 4. Quantum Mechanics with Applications to Quarkonium.78671 7. n = l.e. but for large > values of E Eq. Power Potentials Table 15.En from Ai(£„) = 0 2. Quigg and J.93528 12. Rosner. Co.314 CHAPTER 15.33811 4. copyright of NorthHolland Publ. It is possible. with permission from Elsevier.51716 6. . so that n — 1/4 ~ n.31) (an odd function has an odd number of zeros!).e. (15. Table 15.28) is really only valid for s large and s — oo.93602 12.78445 7..30) ! ^ / 2 . to obtain the zeros of the Airy function numerically. i. It is interesting to note that in the dominant approximation both methods agree. [231]. = im — (7r/4). (15.2.00767 11.29) implies that cos (  ^ .^ = !(2nl). L.
In order to be comparable with experimental data these investigations.15. We saw previously (see for instance Eq.g. . dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. e. Jackson [140]. i.e.** r ( # ^ e e ) o c *(0) 2 .3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs.3 The ThreeDimensional Wave Function 315 15. *Note that this makes N a W K B normalization constant. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. Weisskopf [239]. The resulting masses agree in most cases very well with those extracted from experimental observations.g. the charmanticharm meson ^ . have only a finite lifetime and decay into other particles such as electronpositron pairs. i.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. In general these quarkantiquark pairs. cos2()^— / 1 fr° 2 > w .50)) that in the domain V < E the leading WKB approximation of the wave function u{r) is given by the periodic function *>"^(JC'£)I)' ^vm=rr (1532) where iV is the normalization constant. We indicate briefly here — without entering into further details — how these quantities can be calculated.e. van Royen and V. which are inversely proportional to the lifetimes.34a) **The most frequently quoted reference for this result is R. can be shown to be proportional to the modulussquared of the particle wave function at the origin.• •) by a mean value. P. e. r <53) 1 3  dr cos 2 () (15. D.e. (14. ee. F. in Chapters 24 and 26. See also J. of course. We shall encounter WKB normalization constants at numerous points in later chapters. The decay widths T.
.38) N : sin V ^ Jr dr' 7T (15. From Eq.rcu w X{r)dr = Kh . (15. i.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .2.l N n .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. Power Potentials this averaging implies a numerical factor like 1/2.36) \i dEn irh2 dn According to Eqs.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15..2) and (15. (15.34b) VMEvy Variation or differentiation of Eq.316 In accordance with t h e relation CHAPTER 15..\ \ * n = l. (15. & (15. so t h a t approximately l N2 / Jo k(r)dr .40) VWY .39) in / .35) N2 .e. (15.1 )n . (15.37) (15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15. dE a—— / dn J0 so t h a t with Eq.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 . A = (15.
for various potentials: (a) For a confinement potential of power v. S.23) for the energy levels of the linear potential.37) we obtain: 1 (2M^)1/2 47r ft. . MiillerKirsten [147]. MiillerKirsten [28]. Bose. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. S.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0.* Using the result (15. Hence Eq. Hite and H.g. G. W.41) I*(0)2=_L^)^.l ) n . § S. W.15.44) (15.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. . (15.38) implies 317 (15. for instance.2. K.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. (15. J. 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. Itt^pocn2*"1^"^. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. . W. J. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. 3 .1 ^ ( 2 ^ „ ) 1 / 4 . The case of arbitrary positive power potentials as above has been considered by R. J. . E. 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). MiillerKirsten [29]. and hence with Eq. we have ^'(0) = ( . Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. Bose and H. the expression (15. Kaushal and H. K.
. n = 0." 1 11 L .318 (c) and for the Coulomb potential^ CHAPTER 15.j . in agreement with Eq.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).133). Power Potentials *(0) 2 oc . (15. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . q = 2n + l. ip = exp[(z — c)/2].e. Example 15. . = 1(1 + 1).75). Landau and E. (15.^ +•••. E' = E + gln(r0).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. (11. as well as Expanding U(z) about its extremum at ZQ = —1/2 and using the perturbation method of Dingle and Muller derive the expansions N)2 = ^^ 3 . rewrite the equation as ^ z + [ .L 2 + U{z)]<f>..46) and the classical Kepler period of Eq. 7 With the substitutions r = exp(z — c). Lifshitz [157]. *=^f. D.5. W. Solution: Details can be found in the literature.— oo < z < oo.g ^£+(<*Pl»riy = 0. Bose and H.. and there Eq. .I l n f ^ p .c = —a//3. M. {S = ^ > 0.1.L2 h4 = 4/3exp[(2a — /3)/f3].^ ( 3 ^ + l)55p^«(3^l) + . i. J. K. U(z) =  dz zf3e2a/?e2*. MiillerKirsten [29]. Eq = i q .. S. (14. = (I + 1/2) 2 . .2.
and in fact. A very similar potential is the Yukawa potential V{r) = g2 — . it will always notice the latter's presence. 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.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . Historically this potential arose with the realization 319 pVr . r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2.Chapter 16 Screened Coulomb Potentials 16. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units.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 clear that for ro — oo the > potential becomes of Coulomb type. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. This effect sounds unphysical. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. that no matter how far the particle is scattered away from the source of the Coulomb potential. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. it is. so that like charges are repelled and unlike charges are attracted by each other. This implies. In reality.
and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.3) where for the real potentials we consider here. Since.2 ) or (p = hk) 1 /" . and that. (16.* 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.e. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. all coefficients Mj are real and independent of the energy E = k2. . Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. from the relation Pi + P 2 + M2c2 = 0. Heisenberg [133]. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. i.320 CHAPTER 16.ii k .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 mathematical expression with an exponential is not so easy to handle analytically.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). What is meant is that the fourdimensional Minkowskian Fourier transform of the propagator is given by the relation dk (2TT) J 4 e x ^r—2 = 7Z^u o)^n k + fi2 4TT v y x 2 ikx ^ e^W ( 16 .1) We encountered Green's functions earlier. in fact the parameter [i represents the mass of such a spinless meson. 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. however. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. one frequently resorts in calculations to the expansion of the potential in rising powers of r. (16.
{l. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. (16. Bethe and T. (16. Mandelstam [185]. 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 = 0.3) in the radial Schrodinger equation for the partial wave ip(l. Potentials expandable as in Eq.16. (16. "These were first considered by C. if>oo. 2 . £+**PVV t/. i. only for the cases of n = 0. 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.e.k. A.r). Kinoshita [21]. C.k.1 in the above. J.4) where E = k2 and h = c = 1 = 2m. Squires [258]. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.3) have been considered by various authors.e. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. (16. . Wu [47]. Cheng and T. bound state eigenenergies and the 5matrix. i.5) where K — ik and A n is an expansion in descending powers of K.mo being the reduced mass of the system.§ Our intention here is to consider these potentials as a generalization of the Coulomb potential and hence to proceed along similar lines in the derivation of Regge trajectories.Q. Masson [180]. l+n+l = ^P. S S.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. . T. Frautschi [97] and E. to be of the order of 1/k. however. Omnes and M. .e. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. S.1.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.r)=0. See also the other references below. Froissart [223].e. f H. physical) values of angular momentum as functions of energy E. They are usually written I = an(E). Lovelace and D. i. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. 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. .
In the following we follow mainly the last two of these references. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. (16. (16. J. as may be seen from Eqs. Longoni and T. Squires [258]. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. 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. 'In the case of the Coulomb potential (cf. Muller [201] and [202]. Muller [204]. for all n. J.4) to z = —IKv and set TP(l. Bottino. M. 16. one can assume an expansion of the potential V(r) in ascending powers of r. Screened Coulomb Potentials as a function of the energy. W.9) the cut starts at E = k2 = 0.z) Then x is a solution of the equation VaX= (16.322 CHAPTER 16. Under these conditions the S'matrix is meromorphic (i. W. in the present case at fc2 = 0 or k2 — Mi = 0.** The energy is later obtained by reversion of the resulting series.19). < const. Regge [31] and E. < oo for all 0 < vr/2. Section 11.e.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. J. H.z) = ez2/2zl+1X(l.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. J. W.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H.k.k. Schilcher [203]. *See in particular A.ndp. tf H . A.15) and (16. Miiller and K. starting with the power r _ 1 of the Coulomb potential. . o rV(r) I dpp\V{peie)\ regular at r = 0.)/j. 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. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis.
j l) + (a + j.10): Sm(a. .j) may be computed from a recurrence relation which follows from the coefficients (16. (16.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.11b) ^ The associated boundary conditions are: So(a. as in the case of the Coulomb problem.16.5) is equivalent to a = —n.5)) a = + 323 . K.b.a + + l). MiillerKirsten [249].7) is seen to be of order 1/K.9) we obtain m l)$(a .awhere (a.i 0) = 0.z) = Ha). which means that the hypergeometric series breaks off after a finite number of terms.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. (16. so that to leading order we have VaX{0) = 0. a + 1) = a — b+ 1.10) zm$(a)= ^ j=m Sm(a. {0) X = Ha.j +(a + j + l.j) (16. Eq. and all Sm(a. (16. The function <&(a. j)Smi(a.a+ l)$(a + 1) + (a.0) = 1. (a. where $(o. (a.a)${a) + {a. anharmonic and cosine potentials have been derived in L.11a) The coefficients Sm(a.1).9) (16. We also observe that the ansatz (16. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a). all other So(a.2 Regge Trajectories where d2 Va = z^ and (cf.a — l) — a — l.j) = (a + j l. J.b. (16.* Recurrence relations for coefficients of perturbation expansions for Yukawa. a) —b — 2a. By a repeated application of the recurrence relation (16.j)$(a + j).a + j)Smi(a. Sharma and H.j) for \j\ > m are zero.a + j)Smi(a. W.
and hence Va$(a + n) = ra$(a + n). This follows from the fact that D a $(a) = 0.324 CHAPTER 16. (16.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.e.l. 0 < \j\ < i.An{K). for the coefficients M^> of the expansion Details can be found in the references cited above.' — . + (16. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. (16.[ a + l.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. We first cite the final result .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. a .13) one obtains the quantity An(K) and hence with Eq. [a. (16.j). i R »] = 2K[a.<&(a + n) on the right hand side of Eq.5) the Regge trajectories I = ln(K).7). [ a .o]i = M 0 . of course. when n = 0. 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).aj 2 \ [a.a + l ] 2 [ . [a. Evaluating the first few terms of the expansion (16. Any term /j. (16.[a>a + J]i+i*(a + 3). Va+n = Va .aj 2 (2K)A _ [a. #(a + ra) fjL$(a + n) V.12a) where [a. (16.a + j]i+l = MiSi(a. oo _ . i. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. . the latter can be written .1] 2 .a]l^a) + ^2(2K)i+1 Y.n. 1 .a\4.
However. With numerical methods one can achieve more. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .e. i.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . Thus the expected appearance is that shown in Fig. including an exploration of the domain of small energies E. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. P.1. 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. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.16.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . of course. G. Ahmadzadeh. we obtain the expansion for the Regge poles.^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex iplane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . (16.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.§ This is an important observation which indicates the equivalence of the methods. (16.2.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . Inserting this expansion into Eq. Obviously the expansion — which.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). Tate [6]. 16. .^ p [ 3 M 4 ( n . Burke and C.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] .5). Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.1) + 3M3M02 + M22n(n + 1) (16. .1 by evaluating the first two terms.
a] 2 = M i Si (a. For a resonance with a long lifetime. 16.1: Evaluation of perturbation terms Use the above formulae to verify the first two terms in the expansion of A n for the Yukawa Regge trajectories. so the imaginary part ai of I — an represents its width in angular momentum. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown.^ . one can expect a resonance with the lifetime determined by this imaginary part. (16. Hence Si (o. ai is small and A6 is large." Re I 1 0 1 2 Fig.—" + In = — . a] 3 = M 2 S 2 (a. 0) and Prom Eq.1 Typical Regge trajectory for a strongly attractive Yukawa potential. (16. C. 114. where the imaginary part of I is very small. The conjugate variable to energy is time. Similarly the conjugate variable to angular momentum is angle.11b) we obtain Si(a.326 CHAPTER 16.a)So(a. Example 16.O) = Other terms vanish since So (a. 16. In fact. However. satisfies the relation ajAO ~ h. just as a decay width T represents the width of the resonance in energy. . From Eq. ±1) = 0. For a bound state a j = 0 and the orbit becomes permanent.13).a) = b . . and the angle A#. [a. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. through which the particle orbits during the course of the resonance. and the lifetime At of the resonance satisfies the relation TAt ~ h.0). Frautschi [97]. 0). Solution: We evaluate the first three terms of expansion (16. p.2 a = In See S. (a.0) = (a.1).12b) we obtain [a.
1) + (a .0) + 0 + 0 = ( 6 . 0) + a S i ( a .11b).0) (a .al)S0(a. A" An An A" = = = (a.00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.1 ) + (6 . 1) (a .16.2o)Si(a. . Sec. 0) + (a + 1. a)Si (a. we have to evaluate (cf. (a. using the Cauchy formula (6.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . (n+l)U+^pJ .6)Si(a.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6. 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 .6 + 1 ) . One sets z = 1/r so that.6 + l)a + A.13) we obtain 0 (Mo . . 14. 1).0)+ 0 + 0 = (a1).An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .0) = (a . a)Si (a. .a)5o(a. a)Si (a.2 Regge Trajectories 327 Analogously we evaluate S2(a.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity".1 ) + (a. 1) Hence 52(a.1) Si(a. i ) again from Eq. Thus (with terms which are 0): Si(a.b)(a . (a. (16.0) Si(a.n ( n + l Inserting these expressions into expansion (16.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .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 .a + l)S0(a.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . Solution: Ignoring higher order terms in r.0) + 0 + 0 = ( a . We obtain the quantities S i ( a .2 a ) .
5) together with the expansion (16.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. The turning points in this case are at r = 0 and r = oo.* 16. (16.15).e. Solution: For details of the solution we refer to papers of Boukema.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . we can use Eq.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. ±2HK in agreement with our expressions above.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . Boukema [34] and [35]. by expanding the right hand side in rising powers of v and evaluating the individual integrals. 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 . In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. . i. Here we do not perform this procedure.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Example 16.t First. Singh [252]. I. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. tSee Chapter 11 and V.l. *J.. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .2.328 Thus here dr K' CHAPTER 16. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. Screened Coulomb Potentials Mo .14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h.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. n = 0. (16. however.
which are precisely the expressions yielding the Regge trajectories of above.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\.19) +2M 3 M 0 (3n 2 + 3 n . . J. Miiller [204]. W. One can use the explicit expression (16.17) and is therefore real for real values of the potential coefficients Mj. 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. for instance.. We observe also that the S'matrix is unitary as a consequence of the result (16. i.l. Paralleling the case of the Coulomb potential in Chapter 11.. J.g.19) we obtain the energy. W. the behaviour of the scattering phase in the domain of high energies.l)n(n + l)(n + 2) + 2M2M02 .2. the following *See e.18) of the Smatrix now to explore further aspects.* 16.17)..e. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. n = 0.0[(KZ + Mx)'6} (16. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\.14). 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 . such as. This has been done§ and one obtains . Miiller [211].4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.16. Now reversing the expansion (16.l (two additional terms are given in the literature). . H. § See H.
k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).^ 16. B. J. i.1) 10M 3 M 2 M 0 2 n(n + l)(3n 2 + 3n + 2) + 20M23n3(n + l ) 3 30M 4 M 2 M 0 (n . VahediFaridi [212] contain as special case the expansions of the first pair of authors.11) + 2M22M03(9n2 + 9n . Sommerfeld [256]. Zauderer [286]. N.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.330 CHAPTER 16.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . the transformation was introduced in the form given below by G.e. E. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. Watson [280] in 1918 and later resurrected by A.l)n(n + l)(n + 2) . C. in the following expression with an "See in particular G. 'According to S.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + 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.< 3 M 4 M 0 ( n . Large coupling expansions derived in H. Iafrate and L.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . 282.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. A. p. Frautschi [97].2)(n .* We recall from Chapter 11 the definition of the scattering amplitude F(9. Screened Coulomb Potentials expansion. 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) ^ — .1) + 2M2M. J. Mendelsohn [135] and E. . Warburton [279]. 107. MiiilerKirsten and N. W.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . p.
. (cos 9). dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i.k) — e 2i5[(k) ~" p—iirl (16. 16.. I = n + x.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.fe)P .TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.k)~l] = ~ k" PiSiW sin 5i(k) and S(l.26. \x\ <C l . (11. n = 0.22) = \F(9. ~:' sin nl (16. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).2.184) to (11. . Eq.23) f(l.k) = 2zfc L " v "'' v ..25) Setting Note the minus sign in the argument of the Legendre function.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.kmcose).1.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1.. 16. z=o in which (cf. For this purpose we consider the following integral taken along the contour C shown in Fig.''' The idea is now to consider each term of the expansion (16. Actually 5j lrr/2 as in Eqs.187)) (16. (11.k)\2.187).k) 0ikr (16. The scattering amplitude possesses the partial wave expansion F(0.24) Si(k) being the phase shift.k) = ^T(2l + !)/(*.2 The integration contour C. c = 1.iJ 11 —[S(l.
a + c —> b + d. 16. in which case c = a and d = b.3 Reaction channels and their respective Mandelstam variables. as indicated in Fig. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.k) Zi Jc TT{L . For the momenta indicated we set (with metric +.28) .—.—) s = (p + q)2. n=0 which is the usual partial wave expansion. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem.26) P((+COS0) = 2 ( 2 n + l)/(n. Thus the direct reaction may. u={pq')\ (16.3. 16. d(q') (16. In order to see the relevance of Regge trajectories in a reaction of particles. t = (qq')2. for instance.n) oo {l)1 Pt{cosO) v ' (16. 16.27) a(p) © b(q) Fig. be elastic scattering of a off b. we obtain m® = ± I dl^±^f(l.—.332 CHAPTER 16. a + d — • c + b.fc)P„(cos0). Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. 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.3. 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').
e. (16. N.0s). the dynamics is described by the exchange of quantum numbers. B.9s)\2. Khuri and S.16.4. In a Regge theory on the other hand. the variables t and u would describe momentum transfers in the crossed channels. if s describes the square of the total energy in the s channel.t. S. One now makes two hypotheses. s = 2qf(lcos0t)(16.31) i l Taking all particles to have the same mass mo.t. the quantities carrying these quantum numbers are the Regge trajectories.g.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. T See e. N.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. Thus. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).§ Since the three variables s. M. Treiman [25]. u.u) (if any of the external particles has nonzero spin. (16.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.c o s 0 s ) .^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. F. § See G.. etc. Symbolically we then have the situation shown in Fig.g. Mandelstam [184] and R. (a) Mandelstam's hypothesis: All three processes (described by the s. one then has A(s. one can write the amplitude as depending only on two. For a Lorentzinvariant normalization of the initial and final states.32) t = 4(q? + ml).t) = VtFt(t. if we had been considering the s reaction originally.(<fe)^(cos 6t). the momentum k there would correspond to qs) s = ~4(ql + ml). except for poles and cuts which characterize the reactions in the three channels. which is an analytic function of the variables s. .et) (16.The variables s. t. there will be several such amplitudes which together describe all three processes).2 ^ ( 1 + cos6 s ).t) = y^Fs(s. Chew [48].t).t.u are known as Mandelstam variables. u = . Goldberger. 16.g. t = 2q£(i . Blankenbecler. A(s. A(s.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). Ft = £ ( 2 J + l)F.t. the only kinematics we require in the following is given by the relations (e.u are not independent. L.
Lehmann [162]. in the domain of schannel physical values of the kinematical of the ichannel). Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories.334 Yukawa CHAPTER 16. So what can one do? 1 H.)e~ •fj.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. But for integral values of I.5 The Lehmann ellipse. 16. Screened Coulomb Potentials Regge oc(t) Fig.r 1+m 2 /2q 2 o ^t "" Re cost: Fig.5 for Yukawa potentials V{r) u: lm cose dr a(/j. For s — oo : cos 9t = 1 + s/2q^ — oo. 16. 16.4 Yukawa versus Regge theory. the Legendre polynomial Pj(cos 9) does not possess a cut. But > > wave expansion X)(2Z + l)F. . the partial wave expansion is not valid in the physical region of the (i. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^.e. However.
16. 16. B Iml 1 \ i All \ r\ 0 / closed . i. 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. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.26) but for a different choice of the contour. Thus consider now the same integral but taken along the closed contour C shown in Fig.1 / 2 . and there is only a finite number N + 1 in the domain of 9W = .6. Moreover. all poles of Ql > 0. sin nl [16. the contour representation (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.5 The SommerfeldWatson Transform 335 T h e answer is. to use the SommerfeldWatson transform.33) . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. contour C / / D ^ ^ . „ i.e.e.6 The integration contour C". N F(6. 16.t).^ B 1 A Fig.
(IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. p. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. In particular. Eq. we have (16. the amplitude can be represented as a sum over ^channel Regge poles.336 CHAPTER 16. for plots and discussion see R. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i.t)&B(t)sa°V.35) This result is valid for s — oo and t negative (cf. MiillerKirsten [29]. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. for high energies in • > the schannel). i. Regge trajectories have. of course. 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. if ao(t) is the Regge pole with largest real part. 42. Regge [229].32)) and finite. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. Predazzi and T. Bose and H. i. as we discussed briefly in the above. Apart from more refined details.e. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel.e. this behaviour has been confirmed in high energy hadronic reactions. K. J. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. (16. Foissart [223]. 16. . In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. It follows that under these conditions.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. Omnes and M. W.e. Limic [177] and E. at high schannel energies.
20. The above treatment of screened Coulomb potentials is incomplete. a composite particle being one with structure. The way to do this is similar to calculations in Chapters 18 and 20.g. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. and also as socalled "fixed poles" .16.1.** Whereas the former are related to absorptive properties of a reaction. H. and is composed of quarks. called "Regge cuts". MiillerKirsten [209]. W. the latter are related to the distinction between "' elementary and "composite particles". Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. like a meson or a baryon. J.6 Concluding Remarks 337 important parameter in string theory. For an overview see e. Singularities appear also in the form of cuts in the plane of complex angular momentum. Related aspects are discussed in Sec. .
This page is intentionally left blank .
since many analogous relations hold. called elliptic modulus. are not much harder to handle than trigonometric functions. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers.) which are functions that interpolate between the n. Arscott [11]. M.* These elliptic equations involve Jacobian elliptic functions like sn(a. See F.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. 25.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms.g. 2K or AK. In essence these periodic Jacobian elliptic functions. the most immediate candidate is a trigonometric form like that of the the cosine function. 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.Chapter 17 Periodic Potentials 17. 19 and p. example 14. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.0 < k2 < 1. p. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). however. motivated mainly by the regularities of the crystal structure of matter. K being the complete elliptic integral of the first kind. The Mathieu equation can be obtained as limiting cases of spheroidal wave equations and the elliptic Lame and ellipsoidal equations which represent a further level of complication. The Schrodinger equation with this potential. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. 339 . The reason for these difficulties is.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. depending on the parameter k. and are themselves periodic with period e.
(17. W.1) where h2 is a parameter and — IT < z < IT. The Mathieu equation. need not be afraid of them. J. which is conveniently written ^ + [\2K2sn2z]y = 0. those of integral order. J. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. which one looks up in books like that of MilneThomson [196] or Tables when required. Both of these domains are important for a host of other considerations. In fact. which does not possess any polynomial solutions.1. [205]. W. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0.340 CHAPTER 17. the three Jacobian elliptic functions s n z . the Lame equation possesses 2 n + l polynomial solutions. Miiller.4. (17. and to relate these to the weak coupling bands or regions of stability.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K.e. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . 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.7. 17. Dingle and H.2h2 cos 2z]y = 0. . For large coupling.2 .. c n z and d n z are handled with analogous formulas.. ''The reader who encounters these Jacobian elliptic functions here for the first time. Lame equation: H. i. or Sw&ve Schrodinger equation with elliptic potential. cosine and tangent. The finite heights of the periodic functions of the potentials permit tunneling from one well to another and thereby produce a splitting of the asymptotically degenerate harmonic oscillator levels. 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. We might mention already here the much less familiar but more general Lame equation. B. Muller [73]. 8. The first of these conditions is already tMathieu equation: R. Like the trigonometric functions sine. 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. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. each with its own characteristic eigenvalue.1. Periodic Potentials vide periodic potentials like trigonometric functions. This is a significant difference compared with the Mathieu equation. The most important formulas for our purposes here are collected in Appendix A. See also the discussion in Sec. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions.
Vachaspati [242].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. the semiaxis A > 0.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. Adding the negative term — 2h2 cos 2z to A.e. we see that in a plot of h2 versus A. i. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). i.e. It is this additional parameter function appearing in the solution of the Mathieu equation which attributes the equation its reputation as being particularly hard to handle. In classical mechanics stability is treated for instance in connection with planetary motion.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. 17. y" + \y = 0(h2). Although the mathematics literature on the subject uses the physical concepts of stability and instability. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. these concepts are rarely explained as such there. . the case of h2 small. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. 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). h2 = 0 belongs to the domain of stability. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. we compare the For interesting related discussions see M.1. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. as in Fig.2. i. Looking at the Mathieu equation for h? —> 0. 17. since the solution there is of the form cos vAz. Salem and T.e.e.^ 17. i. a bounded function. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small.17.
TT. Periodic Potentials equations ^ p + [A . y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2). (17. y+{z) .e.342 CHAPTER 17. d2y(z + 7r + [A . ° ^ [1 + 0(h2)}.e.2h2 cos 2z]y(z) = 0. d(z + TT) cos 2z We see that there are solutions which are proportional. i. a = const. sin V Az or linear combinations. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z).2/i2 cos 2(2 + vr)]y(z + vr) = 0.3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. we obtain y(vr) y(2?r) a = — — a = —^^r. y(2vr) 2 a = —7—^ = 1. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. (17. — • . We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. 2/(0)' yfr) . i.B and a. y~(z) = 6sin vXz. Setting z = 0.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z).
we choose these with the following set of boundary conditions: y+(0) = a. y'(z + n) = a[ay'+(z) + py'_(z)}. 2/'(7r) = a ^ W + ^ .oWX] . This means. Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. \y+>y\ = y+{ir)y'{ir) . This expression is equal to the coefficient of a2 in the quadratic equation for a.4).y+(7r)y_(7r) = 0 or abVXa2 .y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\. Using Eq. and by'+(?r) = —ay/Xy{n). y{z) = aty+(z) + 0y(z). 7 y\z) = ay'+(z) + (3y'_{z).{ay'_{ir) + bV\y+(ir)}a Now from Eq.17.aa][y'_{n) . whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. We can write therefore.e. the determinant of the matrix must vanish. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) . y'_(Q) = bV\ with Wronskian W[y+. yV(0) = 0. with constants a and /3.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3.y+(7r)j/_(7r)} = 0. [y+(v) . The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa].y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. 2/_(0) = 0. (17.3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. i. (17.a2b2\ a± ~ 2a6\/A .
<7__cr_ = 1.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2).455A2 + 1291A . (17. i. ** An important consequence of Eq. cr+ = 1/<T_. **This point is not immediately clear from mathematics literature.e.= — . This calculation is demonstrated in Example 17. 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. 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 .2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ . 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.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 . <T__ + <T_ = 2y + (vr) .22)) the boundary conditions of periodic solutions. Eq. .6) We can derive various terms of this expansion perturbatively with the method of Sec.1. Prom this condition we determine later (cf..5) a The parameter i/ is known as Floquet exponent. (17. T h e result for A is the expansion A = h4 2 . (17.344 Setting / = 2by/\y+(Tr). 8. _ / ± v 7 ^ V . Periodic Potentials g= 2aby/\. (5i/ 2 + 7)/i 8 vz + — .7. (17. i. This is the easiest application of t h a t method since only simple trigonometric expressions are involved.8) "Observe that for integral values of v (in lowest order of h2).+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. this implies y+(Tr)/a = ± 1 . we have <7± — CHAPTER 17.e.0 This series may be reversed t o yield the Floquet exponent. Setting cr+ = e . where the constants are usually taken as unity from the beginning. .
10) A = n2. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17.2 how the expansions for small values of h2 may be obtained perturbatively. (17. = J—^ + o(hr). Example 17. Solution: We write the eigenvalue equation A = v2. p.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small.17. 124. (17.5) and (17. (17. VA = ±n. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). (17.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 .9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.35A + 8 . These cases therefore have to be dealt with separately.e.12) S e e J. (17. W. cos TT\/A 32A(A1) 2 .3). For these integral cases of v we demonstrate in Example 17.35A + 15A2)/l8 /. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. Schiifke [193].1) 15A2 . we must have a2 — 1 and hence e 2iv/A7r = 1} i. (17. .jl2. as will be done later.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.„„ x n. 8. Meixner and F. It follows that in these cases i? = n2[l + 0(h2)]. n an integer.11) or smvz or e±ivz.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .
e.v±2) = l. siniy + 2)z + sin(i/ — 2)z. i. '. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . v) = 0. (17. = .15) Hence a term fMy„+a on the right hand side of Eq.11). me±v(z.v + 2)yu+2]. sev(z. . h2(u. The recurrence relation can also be obtained by substituting the right hand side of Eq. when a = 0 or 2v + a = 0. v . for instance.16) directly into the original equation in the form of Eq. The right hand side of Eq. Next we treat terms yv+a in j / 1 ' in a similar way. 17. W. see Example 17. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i.3. (17. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. since j / 0 ' = yv leaves unaccounted R\. Proceeding in this manner we obtain the solutions ce. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. (17. e ± i ^ + 2 ' 2 + e±i("2)z. where except.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions.e. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17. (17. (i/.2.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.2) which applies when v is nonintegral. i. .. . i. v)yu + {v.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. v . h2). (17. up to 0(h2) we have A = 0. or A = 0(/i 4 ). (17. of course.346 CHAPTER 17.2) _ __ _ (y.2)j/„_ 2 + {v. Schafke [193]. With our perturbation formalism — as demonstrated in cases considered in Sees.„ + £ h2i £ P2i(2j)yv+j (17. as shown.16) *These smallh expansions are convergent. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .!/) = 2A We now observe that DvVv = 0.3 and 19.e.(2) + . se and me as* oo i y = „«>) + yW + i. by J.h2).11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0.h2).e.14) = = = cos(y+ 2)z + cos{y— 2)z. and (v. (17. Meixner and F. ("• ^ + 2) (o) 2(2^ .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.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v.14) therefore leads to the following nextto leading order contribution (v.
g. J. (17.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. .7).e.v)\ . These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. Schafke [193]. Solution: We start as in Example 17. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. 0 = h2(u. i. + y~i) (17. W.R{ . Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . one has to deal with each integral case separately as in the following example. Example 17. v) ' ( 2 v + 2) (17.1 with A = v2 — 2h2A but with v = 1. For v = 1 we then have yi = cosz = y~\. thus determining the quantity A. however.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..17) Inserting 1 for the stepcoefficients of Eq.2 ) v(y2. (17."2).. . The higher order terms naturally require a little more algebra. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. In the case of integral values of i>.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)+. See e.. the solutions are not yet normalized. * Note. Following the first few arguments of Example 17. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. Meixner and F.u) + h4 (". 2 ( 2 i / .2—r——— (i/ + 2..14). .1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3]. fo^ + 2) . This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. 121. p.18) Again applying Eq. (17. . „ .17..
m.24c) below) . (17. Meixner and F. Periodic Potentials h4 r ••• . 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 . h4 (17. W.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn .19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). h2). Eq. J. the case of A an integer. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. Schafke [193]. and v is to be found in ascending powers of h2. 120. 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. i. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. see below!) one obtains the additional terms given by Meixner and Schafke [193]. We introduce a normalization constant c and rename the normalized solution y then yc.348 From this we obtain 2h2A = h2 CHAPTER 17.n integers. We consider this case again in an example. . With normalization (not to 1. p. 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]. In the above we considered the case of v an integer and calculated the eigenvalue A. 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 associated solution is ym+ym. 123) (there the solution is called cei(z.e. The reverse situation is later of importance.
(17.17.9b) about A = 4.Q . J. W. on the left hand side of Eq.3 1 A + 8) ( A . o 1• • • . Hashimoto [124]. (11A2 .9b) and considering the approach A — 4 gives > (A4)2 + .• 1 + 5n h 2 s 2932 Setting in cosm.2 Cosine Potential: Weak Coupling Solutions Example 17. Liang and Yunbo Zhang [189]. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . MiillerKirsten. fe47T2(A4) 4%/A(A . Substituting these expressions into Eq.1) . S. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.3 5 A + 8)TT2(A4) _ 3 / / 64(A .1) (A . so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. . (17.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . (17.9b) v — 2 + <5.('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 . 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 ..4 ) + .1)2/A TT2 +h8 (15A2 . Manvelyan. ^S.4)Av A2v A ~ 32A(A . J . ..l ) 2 (A . We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .3 1 x 4 + 8) 274233 (. Gubser and A. H.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • .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.
(17.W4W=UtWSta±I(. M. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r.1 Boundaries of domains of stability. "These conditions. i.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. We now derive these boundary conditions.w^.6. Taking in particular Eqs. = and s+w 2tW±I. 28 . Periodic Potentials Our next immediate aim is to specify the solutions for which Eq.21a) and (17.29.2 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. for instance.21b). 17.e. Arscott [11]. pp.. These are defined by specific boundary conditions. Fig. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.10) applies. may also be found. bV\ y+(0) 2/40) 07. . as also conditions (17.350 CHAPTER 17.2. 17. We verify these for the large/i 2 solutions in Example 17.4) into account. solutions for integral values of v. in F. (17.
y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. /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. which after some rearrangement can be written y+(vr) _ . ^\y±M /vr\yL(7r) .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. .21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ . /7r\j/!_(7r) y L l 7TJ . Then y+ y 351 U . o y+(fy(f)ob\/A _ = 1 + 2I&A/X . we obtain y+(7r) _ i  2y_(f). f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . .4(f) (17. 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+.17.
y > ce2n+i=qo.T + .1.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . ce± = cos 8 64 h4 5h8 b2 = 4 1 .5) that the left hand sides of Eqs. We see that there are four possibilities and hence four different types of functions..— — — • • • .352 CHAPTER 17. These are defined by the following boundary conditions. of v is even or odd. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0.• a2n(h2).2): hA 7h8 29h12 + Y l28^04. in this order. There the second term on the right is nonzero.+ .. .1). These may be subdivided into classes depending on whether the integral value n — 0.21a) and (17. IT.23) We can now see how the domains of stability and instability arise.2. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .24b) 8 h2 z —— cos 3z + • • • ..22) * a2n+i(h ). y > se 2n +2s 9o +i The functions so defined have respectively period 7r. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. (17. (17. A 2 y • ce2n=q0i. 2K. and whether the function is even or odd.24d) v 12 ' **To avoid confusion we emphasize: In Sec.3. 17. Mr7nA (17.. y > se2n+i=qo. (17. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. 17. A *• b2n+2(h2).1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. the case of ai with solution cei (of period 27r). Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2. (17. 27r. A > b2n+i(h2)... .' fh2 V2ceo = l .24c) 8 h2 sin 4z \ . was treated in detail in Example 17. (17.2 we require this relation for nonintegral values of v.21b) must be ± 1 . This is shown schematically in Fig. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • . (17. where the notation for the corresponding eigenvalue is put alongside on the right: y+(f) = 0 with y+(f) = 0 y'{\) = 0 with y _ ( § ) = 0 with with A . 19. (17.
17. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. We can see from the inverse of Fig.17. 353 .z3 ce2 = cos 2z . . of course. Miiller [73].3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . 13824 .> (17. 17.3 17.hz — 12 . J. the walls are not infinitely high and hence tunneling occurs from one well into another. In the case of the cosine potential. .24e) 17. 17. Dingle and H.2 The cosine potential. 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.2cos4.1 that the boundaries of regions of stability merge to lines.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R. W. 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. The cosine potential cos 2z depicted in Fig. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain. B. .1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. Thus in Fig.3 we show schematically how the low energy solutions for integral values of the Floquet exponent are related to the asymptotically degenerate harmonic oscillator eigenvalues. 17.17„.3.
. Periodic Potentials minima at z — ±ir/2.. 8. Thus in this case the quantity on the left is not an exact odd integer. we therefore set 9 A A = 2h2 +2hq + —. Sees. there is tunneling from one barrier to the next.2/i cos z V = 0.354 CHAPTER 17. In the case of finite heights of the potential barriers.1. (17. Taking the remaining terms of the cosine expansion into account. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1..2. and we set this equal to q.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. 4 (17.26) 17. (17. n 0.3) and is.2 The solutions We insert Eq. Ion (17. We therefore wish to expand it about these points. where A / 8 is a remainder.3.2 and 8. simply the onedimensional Schrodinger equation for the harmonic oscillator. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . but only approximately so.26) into the original Mathieu equation and obtain y" + 2h q + rrxr .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. as in the case of the cosine potential.. of course.27) .
17.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z). (17. (17.3 Schematic picture of Mathieu equation eigenvalues.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. and substituting this into Eq.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) .27) we )A = 0.17. (17.30) D A): COS « = 4z + 1 iq sin 2. 16/i/ (17. another solution.31) We make the important observation here t h a t if one solution of the type (17. z). Choosing y(z) = A(z)e2hsinz.+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. the linearly independent one.29) is known. can be . 2 (17.
94) \ — Aq4 (17. ^ ) A + ( ? . 2 A A = («. (9. of course. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout.4)A.33) (17. 2[(92 + l) + A].9) (9. Aq+4 2Jh . (9.9 + 4) = (9.e.35) not yet taken care of is set equal to zero.^ 6T [(«>« + 4 ) A ? + 4 + (Q. 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 (17.356 CHAPTER 17. ? .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. i. (17.94) = (9 + l)(9 + 3). when i = 0 — the terms not involving Aq in (17.36) Then ^4. 1 + 4 ) ^ + 4 + (^. (17.4 ] .32) cos^'+i) [\K . (9l)(93)._ 4 .35) But since and so a term iiAq+n on the right hand side of Eq.Q)Ag + ( ? .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. 2 /i D^Ag = 0. (17.\z To lowest order the solution A is A^> = Aq.4 ) ^ . ? .30) amounting to Ri ] q = .9) = 0.g + 4) . D{q%Aq+4i = 0. (17.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17. .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. With some algebra one finds that 16^ + where (9.
37) Here all the terms. (q. q. (17.q + 4) (g. A. q .38) Aq8 + 1 2 Clearly we can continue in this way. (17. Then for A = A(0)+A(D+A(2) + . . to satisfy Eq. Thus in particular A(z._ 8 (17. (17. q. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q (17.4) (q . In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. .39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. h).8)A.4.4) (q .3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. (q.g + . h).40) .g + 4)(g + 4.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. +4. .. q .g4) 4. except those involving Aq.<J+4 Aq+& + 1 2 1 . the sum of the terms in Aq in Eqs. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. Thus 0 = (g.4) 1 1 (q. Q . q ..30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.q 8)Aq+8 (o)' ^±V(q 1 4.37) and so on — left uncompensated — must vanish.35) and (17. ) ^ 4.)+ & ! = « ( . h) = A(z. (17.17. A(z.4.4 .q 4)Aq+4 q + 4 1 q + + + + + <^(. q. q.. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.39) which is the equation from which the quantity A and hence the eigenvalue A is determined.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ . h) = A(z.q + 4)(q + 4.
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. We return to Eqs. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. 2 w2 + H* (17. . it is found that D(B) z+ — ]B = 0.43) 9 2eh ." 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. 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. (17.42) (17. Before we study the eigenvalue equation in more detail.d2B w dw2 + w(l w dB_ dw d dw 1.) i ( Q . We observe that for the large values of h we are considering here.46) 2s(«.e. (17. The differential equation there is Eq.i ) We encountered Hermite polynomials earlier in Chapter 6. (6. Note different definitions in use.28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( qsin Changing now the independent variable to w{z) = iVhcOS I 7T + Z I . and discussed there.358 CHAPTER 17.67).45) {B) We recognize Dq as the differential operator of the Hermite equation.41) so that the early successive contributions decrease in magnitude. where Hen{w) is a Hermite polynomial when n is an integer. i. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw).44) where D(B) 9 = dw2 $_ !)• (17.
16hJ (17.e. i. with the equation C" + 4h cos z C + 2h ( q .. . q)Bq + (q. i..e.48) This shows that this solution is valid in particular around z = ir/2.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. i.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. the expansion for the quantity A being identical with that of the previous case. Finally we obtain a third pair of solutions..sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. (17. i. as in Eq.51) .7 T H 1 2 Z 4 <1. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).e.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ . A{wz (17.47) where the coefficients are the same as before.50) C = 0. 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.q + 4 ) £ g + 4 + (q. y — Cexp(2hsinz) and Cexp(—2/isinz). also the same coefficients in the case of functions C below.qi)Bq_4. 1 COS I .d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q.3 ) \He* ! (9+1) H.34). forms a rapidly decreasing expansion provided that \w (z)\ « Vh. (17.17. (17. which is an amazing and unique property of the Mathieu solutions — cf. The solution with contributions B^°\B^>.e.
q.w2)He'n . (17. 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.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .44) can be linearized with the same coefficients as in the case of the trigonometric solutions.54) where the proportionality factor is complex.53) This solution is valid in particular around z = —7r/2. 2 J where n=(ql).h).A Hen . (9+3) (2n 2 + 2n + 1) + . 17.n2(n l)Hen2 . A closer look at that expansion is our next objective. He'n(w) = wHen(w) Hen+i(w).7 T 1 2 Z 4 <1. (17. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.4. (17. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.4: Hermite function linearization of perturbation Using the recurrence relations (6. q)Cq + (q.h) ocB{z. These domains of validity are indicated in Fig. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. show that the perturbation remainder of Eq.52) and the coefficients are again the same as before. (17. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. 2' the remainder 72.59d) of Hermite functions Hen(w).q. We note that C(z. (17. q + 4)C 9 + 4 + (q.360 CHAPTER 17. ie 1 COS I . we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). . Example 17. each of them associated with one and the same expansion for the eigenvalue.(w2 + A)Hen.44).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)C g _ 4 .
17. (?) 1 1 + (9. e.' 361 in 2*f*1>[i(gl)]! = (q.g) 1 1 and so on. the solution y = exp(2/i sin 2)^(2). r j=r. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.l)(q ..1i .q + A){q + A..j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. 2)Aq_8} + • (17. q + 4) = (g + 1)(? + 3). The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. (q. with (.3).l)[l(q _!)]. q) = 2[(q2 + 1) + A].q 4. (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).q4) (17.)Bqi(w).i ) ( w ) 2l(. (q.g4)(q4. l)Ag4 + P2(q. ^ 1 Hv" '^ ^' ^ ' .g.0 and —4. q .57) . in the following compact form with coefficients Pr(q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . q)Bq(w) + (q. We can write the solutions obtained previously.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)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.q + 4)Bq+4(w) + (q.4) = (q . (b) The allowed moves are +4.55) M2 = M. no intermediate move to or from q being allowed.3. 2 + ^ M 3 + (17. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4.q + 4) (q + 4.17. ) .56) (94.
1) + (g + 4j. 2) 1 2 ' ' ^ ' ^ 1 (g.g + 4) 2 1 (g.i (g.e. j .g + 12) 1 2 3 (g. (17.j).362 Clearly CHAPTER 17. W. P2(q 2) Pl(g.60) M2r+l = = *R. From the coefficients (17. By the above rules we can write down the recurrence relation for the evolution of a coefficient Pr by steps from the coefficients Pr\: jPr (g.g + 4)(g + 4. the factor j removing a duplication of factors in the denominator at the junction.34) we deduce that Pr(q.Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j. j + 1) (17.j ^ 0) = 0.g + 4 ) ( g + 4. as jP r 2 (g.g + 4) + 1 1 1 and so on.j).58) with P 0 (g. g + 4 j ) P r . . r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. g + 4j)P r _i (g.g + 8)(g + 8. J. l ) . Dingle and H.j) = (l)rPr(q. g + 4j)P r _!(g.l ) .j)Pr+1(g. j) +(g + 4j + 4.j)] = 2j]jPr2(g.j)^2(9.g + 8)(g + 8. .g + 4)(g + 4.g + 4) ^3(9. Periodic Potentials PlM)=(M+i). for instance. and the contribution from a sequence of r moves from q + 4j back to g._1) = Mzil. 0) = 1 and all other P0{q. (17.j).3) ^3(9. Miiller [73]. M2 = P 2 ( g .j). j) = (g + 4j .4.g + 4)(g + 4. x)_ (g.P 2 ( g . p = (<?><? + 4 ) ( g + 4 >g + 8) .59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. B.g + 4)(g + 4. i. and in general* M2r = J]j[P2(g.j). Thus we have.
*E. the solutions and eigenvalue expansions again have the same type of symmetries.2 * +3) + ^ 2 (9g4 . Goldstein [115].62a) f R . These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. S.( 3 ^ .* We observe that the expansion remains unchanged under the replacements q — —q. q is only known to be approximately an odd integer qo = 2n + 1.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.17.284g + 57) + (17. W. a= (8/i)3(9x) [1(91)]! 1 . L. 17.3.^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. as a result of tunneling between wells.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions.5 the following proportionalities there: B[w(z)] = aA(z). Dingle and H. as pointed out earlier. In fact we demonstrate in Example 17.^ . It will be seen in Chapter 18 that in the case of anharmonic oscillators. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. h — —h. B. Its precise deviation from qo. of course. [137]. J.4. [116]. Up to > > this point. where solutions overlap. We observe that there are regions.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 .(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) .92q3 + 70g2 . . 17. Ince [136]. remains to be determined next. Miiller [73]. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods.
and determine the constants a and a. Example 17. Solution: We begin with the solution containing the function A{z).62b) *z domains of overlap Fig. We write and expand this function as follows: .1 ) ( 7T + .364 and C[w(z)] =a~A(z).COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA. 17.7T H Aq(z) = = C O S ^ .(17.32). The first term of the expansion of A(z) is the function Aq(z) given by Eq. Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.1 4 .1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . a .4 Domains of solutions and their domains of overlap._ ^ l CHAPTER 17.
22). with e. s .17.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 .63) In Example 17._i)]! w*te» 2l(91 (. (17.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.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.e.g + 4) (q + 4. these are conditions imposed on even and odd solutions at the point z = it/2.61) for the connection and Eq. the coefficient of the dominant factor c o s ( .i ) 1 (g.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. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. (6.!)(. A*1) given by Eq.284g + 57) + • The other relation is obtained similarly. \ql)/2(w) 2i(9D[i(.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 . Since the solution with A is obtained from that with A by a change of sign of z.2.46). with (8h)li''1) (3g2 .53) for the expansion. As we saw there. i. (17.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.g. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. 17.36). Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/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 . we have to impose the appropriate boundary conditions.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed. (8. These are the conditions of Eqs. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa. we obtain Bq[w(z)} H.92q 3 + 70q 2 . (17.
(17.66) then follows from the recurrence relation (6. i.62a) and (17. He'u{w) = wHev{w)He„+1{w). Magnus and F.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. 80. a ± C[w{z)}c_2hsinz^ a (17.66). With the help of the duplication and inversion formulas of factorials.g. Periodic Potentials We can extend the solutions (17. He2n+1 (0) = 0. and then replacing 2n by i>.59d). one obtains the first of relations (17. p. He'v(0) = Hev+1{0).66) ^E.22) and to correlate the solutions to those for small values of h2. The second of relations (17. .63) to the domain around z = ir/2 by using the proportionalities (17.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.e. give the following expressions for polynomials with argument zero: ~19)"(^)!. W. Oberhettinger [181]. The functions defined as in Eq. Then y± oc ^ M e ^ i n . Thus y+ = ce and y_ = se. 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. i.e.» a \dw _ C[w = 0}c_2h a w=0 = Q (17.62b).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\ ! ' .366 CHAPTER 17.
Hence expanding the cotangent about these points.( g .39 (17. .67a) g 106g 87 2 14 /i 2 (17. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .1) 7T .! ) ] ! [ .39 2~^h? (17.67c) 4 2 \dw)w=Q C[w = 0] [\(q . we have cot j ^ ( g ~ 1)} = ~\<(l . from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.87 4 2 ¥w (17. the solutions would be given by q = go = 3.69) If the right hand side of this equation were exactly equal to zero.40g3 + 18g2 .Qo) + 0[(q . (17.7.17.67c). and using the reflection formula (8. and similarly for C and its derivative.g0)3].68) Inserting here expansions (17.\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .67b) <f ~ 82g^ .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .11.67a) and (17.136g + 9) + • • • .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 .106g .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.» 2 6 /i the condition reduces to the equation cot{ .67d) Considering now the second of conditions (17.82g2 .
40gg + 18gg .70) + ^ 2 (9?o .408g0 + 1089) 2 /i +• (17.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .e.136g0 + 9) + • Clearly the last of conditions (17.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.(17.528gg + 3307Q .40g3 + 18q2 . .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.120g£ + 467g . 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 . The remaining condition (17.72) This leads again to the result (17. (17. .5.71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . i..40gg + 18gg ..70) except that now qo = 1.136q + 9) + . we use \(q)~\(q0) + (qq0)(^J . Considering now the first of conditions (17.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg ..65) leads to the same result except for a change in sign.65) yields the same result except for a change in sign. = _a a 6 _4h (17.9. Thus we have 2 (16/i) qo/2 T2 4.
+ ^ 2 (9<?o + 89o " 78g02 . J.. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) .. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. and not the separation of these as a result of tunneling. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. results from the boundary conditions. In the literature that we follow here* in all cases several more higher order terms have been given. The complete. and that for the odd Mathieu functions se qo+ i and se9o (lower. n = 0. just as any solvable problem is important for the insights it offers in a concrete and transparent form. though approximate. Miiller [73].74b) The above results for the cosine potential are in many respects important. p.3. 120. See also F. *In Sec. 9o = 2n + 1.61).2. Dingle and H.1.3 we call this difference 26E„n. (17. W. 17. B. The tunneling effect. this becomes 369 = (16/i)^+1e4h l .t From Eq. (17. minus sign). i.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . . plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. quite apart from applications.17. . In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. Goldstein [116]. 20.e. M. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. Arscott [11]. We see here that *R.88q0 . (17. The dominant contributions were also found by S. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels.A_(g0) ^ .3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.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.87) = Azpteo). (17.3.e. example 3. i..
As mentioned at the beginning. From Eq. 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. the Schrodinger equation with the cosine potential is a special case of several other more general cases. T y'_(0) y'A±*) y{z). (17. Example 17. Thus we do not reproduce every step. 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. y^{z) = Aq(z)e~2hsin *. . Naturally one expects the solutions there to be related to those of the periodic case considered here. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. 17. evident through the exponential factor exp(—4/i) in Eq.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . The explicit derivation of the level splitting (17.74b) is also important for various other reasons. The splitting can also be obtained with some methods of large order of perturbation theory.370 CHAPTER 17. The exponential factor represents. Periodic Potentials this nonperturbative effect.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). results from the imposition of boundary conditions. Then we obtain first y±(z) = yA(z)±y^(z). M±TT) . . 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. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion.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. and one can compare the methods.74b).§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. (17.6: Translation of Solutions For the largeh 2 even and odd solutions (17. yA{z) = Aq(z)e2hsin '. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). in effect. _3/+(z) + ____2/_(z). the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). 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. We begin with the solution containing the function A(z).2.
y]z=„ = [y+yL .) cos < 7r(q 1) [ = . we obtain — with Aq{z + 7r) = (1)^ ^ 2 A . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) . etc.7 r ( g ^ l ) U + ( z ) q : i s i n { . A ?q( Z = ±TT) = .( 2 ) .1 )  .4ft) cos i TT(<J + 1 ) 1 . From the equations for the sine and cosine we can deduce the value of the Wronskian at z = n in the leading approximation for large ft: W[y+.T ( ? =F 1) p + (0).Ah) ~ 32ft. (l^ifr^i^O + Cl^ifo^j/aCO cos  .2hAq(n)] . ( .[A'qM + 2hAq(n)] = iV2(q . we obtain the conditions stated at the beginning.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials. y'_ (0) = \/2(4h . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0). y+(0)~2V2. Replacing sines and cosines by the functional expressions.31).VA DV 371 1 V±.g A ? ( z = ±TT). Aq(z) The derivatives can be handled with the help of Eq. (17.4ft) ~ 8(q .4.( z ) From this we deduce for z = 0.17. The Lame equation has not been a widely known equation of mathematical physics so . 17. which is easiest by comparison with the foregoing treatment of the Mathieu equation.z ) .• Then y'+ (0) = 0.q). but also to enable a brief familiarization with the Lame equation.c o s  ~(<1 T 1) f y .7 r ( q = F l ) } j / .3 The Ellipsoidal Potential Replacing VA.%/2(g . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). where Aq(0) = Aq(0) = V2.yV+\z^ = [j/+(0)] 2 (<? . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .( ± 7 r ) = c ° s  .4ft.4 17.. = ±ir). Similarly we find y(z±ir) = = and by putting z = 0: V. and y+(z = ±TT) = V2(q . A' (z = ±ir) = (.4ft) sin j ^n(q .
the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. k) for k = 0." Fig. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). The most conspicuous difference compared with the Mathieu equation is.75) ' T h i s means double well.0. Arscott [11]. p.fi2A.4sn4u]y = 0. MullerKirsten and D.5 The potential sn2(z..0. J. so that Floquet's theorem cannot be immediately applied . 17. however. 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 .0.. 194) remarks to parallels with the Mathieu equation:". *A.5.. W. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials. H.* As Arscott [11] (p. but also cubic potentials. Tchrakian [165]). the nonoccurrence of the Floquet exponent.372 CHAPTER 17.K2sn2u .. 19. as alluded to at the beginning. But more recently it has been observed to arise in various contexts.75. Ince [138].. f E . H. inverted double well and cosine potentials (J.. This line of investigation. L. Erdelyi [85].98. Liang.0 and .* and the work of both prepared the ground for presentday investigations. and was continued by Erdelyi.. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . .Q. Arscott [11] (p. (17. § F . 194) remarks. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane. M. Periodic Potentials far.5 < z < 5.1.
cmx and dnu. the usual factor —h2/2mo.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). . . The function sn2u is plotted in Fig.e. Although the results have been calculated for the ellipsoidal wave equation. 2 .2 = n(n + l)k2.e. W.K) = qK+^±. i. the equation ^  + [A .77) where q —> go = 2iV + 1. (17. K being the complete elliptic integral of the first kind.77) into Eq.78) We follow in this brief recapitulation the description in H. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. 17.75) reduces to Lame's equation and one writes K. J. Eq. 17. MiillerKirsten.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. The range of the independent variable u is 0 < u < 2K.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions.1. is the elliptic modulus of the Jacobian elliptic functions snu. If we put fl = 0. in front of the second derivative has to be kept in mind (mo being the mass). . (17.\k\ < 1.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. Jianzu Zhang and Yunbo Zhang [206]. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.2k. In order to distinguish the above equation from that with the Lame potential. (17. (17.K2sx?u}y = 0. i. (17. we consider mainly the Lame equation. and of the derivation of the level splitting'. in the case n2 — oo.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. .5 for several values of fc. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). The first step is to write the eigenvalue A as A(q.17. In comparisons with the Schrodinger equation. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects.4. N = 0. The second step is to insert (17.
.q.77). A into equations in terms of the variables z(u) = ^—{k' . (17. dnu =F cnu .82 . i. / z J . one has to derive new sets of solutions there and match these to the former (i.e. A(u. (17.e.K). A. B and C. C replacing yl.Es(ii).e.374 where CHAPTER 17. . Periodic Potentials . dnu± cnu K Thus one can construct solutions Ec(u).81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). one pair in terms of Hermite functions of a real variable. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. A are derived. which are respectively even in u (or snu) or odd.K)=A(u. but solutions B. by transforming the equations for A. more precisly for dnu = cnu F 1 » . (17. \/8K (dmz = cn« \ ' F . C which are valid for dnu±cnu <1. one obtains again the same expansion for A. ^ (17. _ Solving the resulting equations iteratively as before. k' = Vl~ k\ \dmi±cnu/ 17. the other in terms of those of an imaginary variable. A second solution is written y = A(u) exp < / nsnudu >.80) The domain of validity of these solutions is that away from an extremum of the potential. /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.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. i. Thus in the third step two more pairs of solutions B. .q.K — > A(u)=A(u + 2K). determine their proportionality factors) in domains of overlap (their extreme regions of validity).
3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a.17.e. 64.4. Tricomi [87]. as well as \duJ2K 11 (17. i. (17. Es(2f0 = Es(0) = 0. W. G. F.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. This expansion is found to be in the more general case of Eq.63)} + ••• . £2 = 0). a of _ _ B = aA.384n2k4(q2 + 1)} q r {(l + A.6(5g4 .3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions.75). Oberhettinger and F.^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) 8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) 64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£. Magnus. 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) . who obtained this expansion for the eigenvalues of Lame's equation (i.38g2 . *A. . J. (17. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. Erdelyi. (17. p.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . W.e. the ellipsoidal equation.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. C = a~A. Miiller [205].85) The first three terms of this expansion were first given by Ince [138].
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 25.2 /c +256fc2g0(g2 + 5)} (17. 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).91) tt H. and the lower to Es*>+1 or Es*5.l n 2^+ i J + 0(«i) (17. Finally expanding A(g) ~ A(g0) + ( g . (17.4k2(q2 + 2g0 + 5)} + • • • ].go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .Ec£ 0_1 and Es2° of periods 4K. . 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 .136g0 + 9) n 3.:J {3(1 + k2)2 (9q* . 2K.Es^ 0+1 . J.90) + 3.40g03 + 18g02 . go being an odd integer.376 CHAPTER 17.g 0 ) ( ^ . W. Periodic Potentials These conditions define respectively functions EcX°. from which the difference q .78g^ .89) one obtains the eigenvalues from which the level splitting can be deduced.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .g0 is obtained by expansion around zeros. Muller [205]. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .88g0 .87) (17.j 26K2 (1 85) = A(g0) + (q .i Here the upper sign refers to Ec*> l or Ec£°.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 and 4K respectively.
V. One can see. (17. Miiller [206].17. K = 2/i. Dunne and K. the conditions (17. that calculations with the Schrodinger equation are not only easier.2h2 . W. but also more easily generalizable.86). K2 = h2 (say).f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«.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 = 0.4 h ( i e M 9 0 7 2 \. h . Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. A = 0.2h2 = A. in which they dubbed the classical configurations Lame instantons. L. + E. . (17.J1 O. this equation becomes y" + {A . (17. Thus in the case of the dominant contribution of Eq. it = x ± ^ .3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods.Ah2 sin2 u}y = 0. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A .74a). Replacing u by x ± IT/2. (17.4.e.93) (17.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.2h2 cos 2x}y = 0.* Apart from the choice of notation. 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. however. .90) one has in this limit: /l±±\ ~K/U = e . i. Ince [138]. A ./2i and hence A^ A/ A AU I2 .87) agree with those of Ince. J. 17. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. *H.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. Rao [79].* M G .
. Cho. W. Rana [287]. N. all basic potentials. This is an interesting problem. H. Liang. Tchrakian [165]. M.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation.g.e. See Chapter 25. D. i.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). S. A.378 CHAPTER 17. Ouvry [276] and Jianzu Zhang. W. Khare and U. Shiraishi [49]. also because the role played by the Floquet exponent in this problem is not yet well understood Jl Another recent appearance of the Mathieu equation is in the study of the mass spectrum of a scalar field in a world with latticized and circular continuum space. Ganguly [103]. for instance. § . A host of related elliptic equations has recently been discovered and studied.Q . 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. J .5 Concluding Remarks The potentials we considered above will reappear in later chapters. H. A generalized associated Lame potential has been considered by A. J. A. Sukhatme [148]. Gu and S. 1 1 See Z. Periodic Potentials 17. inverted double well and cubic potentials. MiillerKirsten and J. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. de Veigy and S. Thus. MiillerKirsten and D.^ The equations considered above also appear in diverse new problems of physics. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. **Y. Quian [123]. H. double well. Kan and K. See e.Y. J.W.
W. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. Wu. both of which make the calculation more difficult.Q.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. Priedberg and T. The recent work of R.. Bender and T. T. D. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. H.. Wiedemann [4] and a revised version of parts of this reference by J. J. f C . Wu [18]. MiillerKirsten and A.* 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. 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. M. W. MiillerKirsten [163]. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Achuthan. In particular the investigations of Bender and Wu. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. ' T h u s A.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. 379 . [19]. Bender and T. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. J.".Chapter 18 Anharmonic Oscillator Potentials 18. This seemingly simple problem revealed extremely rich internal structure . T. Liang and H. Weinstein [281].
Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. and thus lead to very different physical situations.1. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . The three different cases are: (1) Discrete eigenvalues with no tunneling: In this case V{z) = \\h'\z* + \\<?\z\ (2) Discrete eigenvalues with tunneling: In this case. 18. described as the case of the double well potential.380 CHAPTER 18. with the potential described as an inverted double well potential. V(z) = \\h*\z* + \\<?\z*. (3) Complex eigenvalues with tunneling: In this case.1 The three different types of anharmonic potentials. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. 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 . *~z (1) (2) (3) Fig. 18. These contributions may be given different signs.
Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. The shift of the eigenvalues is best calculated with straightforward perturbation theory. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic.175) below. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17. We begin with the latter. The result is an expansion in descending powers of h?. 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. as is sometimes hoped. The imaginary part will be rederived from path integrals in Chapter 26. . since the potential decreases without limit on either side of the centre. 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. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). 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).e. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. 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 eigenvalues of this case are given by Eq. 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. it will not be possible to obtain the exponentially small contributions with convergent expansions. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. that the general applicability of the method becomes evident. Thus we are mainly concerned with the double well potential and its inverted form. Case (3) is seen to be very different from the first two cases. except for a change of sign of c 2 .1 Introductory Remarks 381 normalisable wave functions.18. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. (18.86) below. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). The level splitting will be rederived from path integrals in Chapter 26. i. In this case our aim is to obtain the aforementioned complex eigenvalue. however with decay as a consequence of tunneling. The eigenvalues of this case are given by Eq.
h 4 2 2h1> ~1/2 2c?. The potential in this case is given by V{z) = v(z).2. (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 . (18. We take here h = 1 and the mass rriQ of the particle = 1/2.382 CHAPTER 18. 18. 18. Anharmonic Oscillator Potentials 18. a point which has to be kept in mind in comparisons.2.1 and more specifically in Fig.3) "1/2 harmonic oscillator h8/25c2 . If suffixes 1/2.1 refer to the two cases.2 18. 18.2 The inverted double well potential with (hatched) oscillator potential.z Fig.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. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.1) for h4 and c 2 real and positive. v(z) = ^h4z2 + ^c2z\ (18. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.
/i)..7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w).. The problem here is to obtain the solutions in various domains of the variable.d2 .. (18.e. \h?\ — oo and c finite. The result will be that derived originally by Bender and Wu.2. (18.. the harmonic part of the > potential dominates over the quartic contribution and Eq.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. 18. to match these in domains of overlap. i. n = 0. and a variable w defined by setting E= b* 1 + A A and w = hz. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. _i\(w) and q = qo = 2n + l.18.2. (18. (18.5) becomes .2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0. y{w) oc Di. 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 (18.1/7  Vg(w)y(w) = o(J^j.4) we can rewrite Eq.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.1. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. (18. 4 .
•A~2 hrz" 2^\ c^z V2 —— + (18. (18.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.10) into (18. as we shall see. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.2.14b) . be and A = A(q. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.2 A (18. (18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^.9) z =0 these Thus these c2z4 1/2 + • 2 (18. This observation allows us to define the pair of solutions yA(z) = .11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) . as encountered and explained earlier. h) is obtained from the perturbation expansion of the eigenvalue.2 We are concerned with the equation d2y(z dz2 where ^ 1 . these solutions are not valid around z = 0. Before we return to solutions we derive a new pair which is valid in the adjoining domains. 1 d r A!{z)±iA{z)±{ 0. We observe — before touching the square roots in Eq. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18.13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0.10) E= qh2 ^ j.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.14a) 1/2 VA( ) = A(z)exp (18.384 CHAPTER 18.4(2. In order to arrive at solutions we set in Eq.+ ^ . Inserting (18.9) we obtain 1 h4z2 c2z4 A L2 (18.
1 (18. / c z Aq{z)+0 1/2 .13) and A(z) that of the lower of these equations.zh2 =()*2"(18. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.e.15) where A{z) is the solution of the upper of Eqs.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.1 and as a verification again in connection with the solution y# — as the other solutions. Since these higher order contributions are of little interest for our present considerations. (18. (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. we do not pursue their calculation.20b) .19) We see that one solution follows from the other by replacing z by — z.16) For large h2 we can write the Eqs.34) below) — to be derived in detail in Example 18. 1 2 4 2 Aq(z) + 0 (18. i. Clearly Aq(z)..Aq(z) approximate the solutions of Eqs. (18.4 .17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18. (18. (18.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z). We take the square root by setting z2h^1/2 +.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. /H 4 Z h +~C Z 21.18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. 385 (18.
1 . (18. W i t h proper care in selecting signs of square roots we can use the solutions (18. 2 .e. z) ± yA(q.2 .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).l)A(z) 1 = tfA"(z) A .f .e. . Vq+ii = Vq + 2i. „ .20b).20a) and (18.. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. z)). h2.20a).. (18. A = . (18.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A.17) and (18.. one observes that with £>. we write > V±(z) = . Anharmonic Oscillator Potentials i. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).2 ( * ) 2 and from this or separately A"(Z) : 1 (q .:=*£ +i(5l). away from the central minimum.e.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. a2 = 2 1 . zA'(z) where the expansion coefficients are given by ao = l. * = — .v ' = o (« .• Using Eqs.[yA(q.21) Example 18.^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. c*i = 1 1 . ft = . h2..18) we obtain A » = H^rL^z) 2 z . (18. i.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q .! ) ^ . 128 8 16 A = i. to obtain in leading order the eigenvalues E. 03 = . (18. procedure. i. 04 = 5 . .386 CHAPTER 18. h2 —> —h2). A A _ 3 . X » / A q + 4 i = Mg+4i .
[1(9 + 1)]! .e.18.(q + 3)(q + l)Aqvri(q '^ ' . ^f( g + i)(±H2^ + 1 > \ — 1.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.22) is invariant under the combined substitutions q — —q.—77 [i(9l)].. The solutions yq(z) of the equation Vq(w)yq(w) = 0. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .. Rq '. _1J±w) and D_\. of course. We return to Eq. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B.. 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).23) . (18. give an equation from which A is determined. (18. w = hz. r ( CJw) = ^ .w — > > ±iw) or functions R B w q( ) . = •>— r.2 The Inverted Double Well Potential Thus a term /j. 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.22) are parabolic cylinder functions D i . In this way we obtain the next order contribution A^1' to A^°>..'2H9i) ^faDfr™) and .5). +1J±iw) (observe that Eq. and the coefficient of terms with i = 0. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . those in Aq(z). (18. when i = 0. (18.
Aj)(18.l)2/?4.4j)yq+Aj.q + 4j}yq+4j. we also have Vq+^yq+tj = 0. (18.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.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 + .25) where in the case i = 2: SA{q. As an alternative to Bq(w) in Eq. F. 115 . .q}=A + c2S4(q. (18.±4) S4(9. G. Oberhettinger and F. j=i (18. Erdelyi.( < ? .24) The extra factors in Eq.29) ' A . (18. (18. 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 .±8) 5 4 (g. Comparison with our notation is easier if this reference is used. (18. and so Vqyq+Aj(w) = Ajyq+Aj(w).123.3)y g _ 4 .q + 4j] = c2SA{q. pp.27) Now.0). and for j ± 0 : [q.3. since Vqyq = 0. Magnus. S2i(q. Tricomi [86]. Actually these factors can also be extracted from W K B solutions for large values of q. W.^ For higher even powers of w we write i w2l V<i = Y. (18.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.388 CHAPTER 18.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. (q±2)(q±S). l(q2 + l). See Example 18.0) = = = ±(q±3)(q±7). Vq+4j = Vq + 4j.
Thus the next order contribution to yq is y H = ie 2^ —z. (18. 2 . i.3 °) For the sum y{w) = y(°)(«. n — 0.34) We observe that odd powers of q arise in combination with odd powers of 1/h2.1. .35 ) . p i(9»9 + 4 J'WK7». i. .31) Proceeding in this way we obtain the solution y = yW („. (18. Equation (18. A = ^(q2 + l)c 2 + o(J^. (18.32) Evaluating this expansion and inserting the result for A into Eq.10) we obtain E(q.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. h ) = qh . . ( 18 .(!) («.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. (18.—v*+*j ( 18 . In Case (3) the parameter q is only approximately an odd integer in view of tunneling.[rp ) Y.) ty^^w) to be a solution to that order we must also have to that order [q.q]=0.18. (18.e.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1.) + j. and c 2  replaced by — c 2 .• q.e.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. so that the entire expansion is invariant under the interchanges q .—A(q + 1) . We can now write the solution y(w) in the form y{w)=yq{w) + Y. h2 —> h2. and even powers of q in combination with even powers of 1/h2. This type of invariance is a property of a very large class of eigenvalue problems.
q + 4j) = 0. (18. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. g ± 4 ] T4 ±4 and so on. we may infer that given one solution y(z). J." Since our starting equation (18.36) with the boundary conditions p o(q. and for j ^ 0 all other P0(q. q) = = 1. Wiedemann [4] . Again we can write down a recurrence relation for the coefficients Pi{Q. We thus have the following pair of solutions VB(Z) VB(Z) = B w .37) For further details concerning these coefficients. there is another solution y(—z).g±4] ±4 ±4 [g. MiillerKirsten and A. their recurrence relations and the solutions of the latter we refer to the literature.38) [yB(z)]argz=7T = [?/s(2. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. > "p.i 1 + 4j) in complete analogy to other applications of the method.—. ? T 4 ] [ g T 4 . 2 4tPi{q. W.11) is invariant under a change of sign of z. q + At] (18. Achuthan. H.q±A) = MM!i±MM+[M±8][g±8. ^VO(<L<?) ^i(?. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. i.argz=0 (18. We observed earlier that these are obtained by making the replacements q —> —q.390 where for instance v J CHAPTER 18. o.e.)]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. w — ±iw. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz.
h^ih.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.39) are with the same coefficients Pi(q.h^ih 391 (18. see e. In order to be able to extract the proportionality factor between two solutions. 18.g.4)).18.2. Thus in the transition region some become proportional. Mauss [192].z2 . like w of Eqs. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. with normalization since the normalization constants are also asymptotic expansions.The solutions yc. Integrating and expanding as follows since h? is assumed to be large. 4 . the solutions of type A being valid away from the minimum. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.g. in fact. J.q + 4j) as in ys.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.g.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?. Vc(Z) = [V!B(Z)}q^~q. .40) "Variables like those we use here for expansion about the minimum of a potential (e. Such contributions arise. (18.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. (18. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. one has to stretch each by appropriate expansion to the limit of its domain of validity. Aq(z)) does not appear in any of the higher order terms. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 .
2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.45) . We do not require this at present.42) H W2 (ql)e$W I „ . Comparing the solution VA{Z) of Eq. (18. .43) (since z — — 2 implies > argz = ±7r).0x ! [ £ ( g . 2 ^i![i(g4il)]!(2«. eh* /12c* e\z*h? (18.42) we obtain for the solution ys(z). ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . 2 ) i ' ir! i iHim 1 3 j argw < 7T.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18.44) In the solution y~g(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. > n.. (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(±\\.l ) ] ! ( . with z in ys(z) replaced by — z.w yB(z) ~ B.2 « .ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=. The function Di. we would have to substitute correspondingly the expression (18.41) with the solution ys(z) of Eq.4 i . (18.7 r . y^(z) we see that in the direction of z = 0 (of course.392 CHAPTER 18.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. 4/T 4 Prom (18.z).43) W + 1)]! w^+V A .
e.42) into y~c(z) w e obtain Vc(z) = (h*. The first of the conditions (18.46) However.9.5. i... and the second qo = 3. (18. Recalling the solutions y±{z) which we defined with Eq.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.41). (18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18. We proceed similarly with the solutions yc(z).y'_(0) ^ 0.7.11.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .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.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. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18. 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.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.50) will be seen to imply qo = 2n + 1 = 1. At z = 0 the solutions of type A are invalid. the ratio of yA(z)iVB(z) i s n ° t a constant.18.49) d yc{z) Again there is no such simple relation between yA(z) 18. (18..21) as even and odd about z = 0.2.yc{z)pansion (18. large probability for the particle to be found thereabouts. we see that in their common domain of validity yA(z) where a = =Vc(z). Looking at the potential we are considering here — as depicted in Fig. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. hence we have to .1 ) e 12c 393 Kgl)]^"1) l + O h? (18.50) and y+(0) ^ 0.
M. 1 [7(93)]! n «(°) = . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.' '.(«.(0) (18. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.5 ( 9 + !!)(*"')• i ( 1) ..(0) = and htoi)(°) ^—'. A.55) Solution: Prom the literature.(0) + ^ ( 0 ) l z—>0 2.2.( g + 3)}..57) [i( 9 _l)] ! 2 i(9i) B.^{(9+1)}. — C. with the reflection formula (—z)\{z — 1)! = IT/ simrz.g. Z (18.oi(.(0) = .( g + 3)}.(iu)]. Stegun [1].58) diu [  ( g . y c ( 0 ) .53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. ( 9 + 3)]! s i n { .l ) ] ! [ .F ? . [C. Abramowitz and I. IT 4 dio In fact.s i n { ./o^Tpif (q+i) .SxW) = \ i»fl(0) .(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}. C.fc(0) Thus we obtain the equations (18.394 CHAPTER 18. Expressions for C 9 (0).54) dio q(w) .u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(. we obtain i V7T24( g .! ( . e.1)('») + ^ r + l)]! ^. .56) D 4(.i)H = e. + 1)]!' D ' j .) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18. We leave the detailed calculations to Example 18.51) and 0 = V(0) = Jim i»xM ."*(«!) ^ ' ~ [ . Example 18.i ( g + 3)]l (18.. (18.2: Evaluation of yB(0).. one finds that [ 1 IT (q + 1)1! = .i ) . y'B(0). [ .52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.l)(°) Thus with the help of the reflection formula cited above: D B. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18.
i .6 ) } ^ — i t a n < — (q + 3) >._l) From this we derive C.+i)(°) [ _ I ( .i ( « + i ) [ _ i ( .)[i(g3)]! and ^i(. (18.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. A __f7r \4 'J We rewrite the right hand side of the derivative equation again with the help of the inversion and duplication formulas and obtain _ (^)g/4(_l)j(g+l) e /> 8 /6^ _ p (/i4)g/4(_1)(g+l) _^_ .sin{^(g+ 1)} = J . _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.sin{^( 9 .56). (18. W .49). isin{f(g + 3 . 4 r T (18.3)}. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.18. using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.+1iH]«=o = . ~ ^ 1 5(9+1) 2i(''.62) .1/2)!.(0) = *>i(„+i)(0> 2 . (18. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.46). Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .53) in dominant order and insert the appropriate expressions for a and a from Eqs. we obtain 2 i(»+i)[_i(.
1.2.. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions.64) for q = g0 = 3. ..11. 18. (18. i.. (18.63) vanishes for q = go = 1.65) Expanding similarly the left hand side of Eq. (gg 0 )^±^ y e5?. .. (18.64) about go — 3. ^ . we obtain cos((g + 3)) = .9. + 3)} = . 1 1 .go) ^ (1) It follows that we obtain for the even function with q — go = 1..396 CHAPTER 18. Anharmonic Oscillator Potentials Then the derivative relation of Eqs. 7..63) and (18.9. 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. In fact the left hand side of (18.64) the right hand side is an exponentially small quantity. With a Taylor expansion about go the left hand side of (18._ e =*.. 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.i . by the rotations E > einE = E.. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo. (18.L ^ i . . Thus we have to determine these conditions first. Recall > the original Schrodinger equation (18. z^ ein/2z.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. 18.5.e. the vanishing of the wave functions at infinity). . This is Case (1) of Fig.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. (18.^ >^_>m e J*.. we again obtain (18. equivalently.53) becomes 8 m(I(. (18.2) with potential (18. z2 > z2. .53).63) becomes (? .^ . 7 .9o) ^ cos  .( g 0 + 3)  + • • • ~ (g . the energy E is real..1).5..64) In each of Eqs.63) Proceeding similarly with the second of relations (18.
for fiz * ±00 we have y(z) ~ exp < ± i / dz In order to decide which solution or combination of solutions is compatible with the square integrability in the rotated (c2 reversed) case. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. if TT/3 < 6 < 0. in the domain — TT/2 < 36 < > 7r/2.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. i. 2 c2>0.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations.e.e.e. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. i. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. the case of complex E). Now. y{z)~exv\i[) 7T dV'V. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. r / c 2x 1/2^3 = exp{ ± i — —\.67) Rotating z by vr/2. (18. the resulting wave functions vanish at infinity and thus are square integrable. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase.e.66) z3 = \z\3em.e. 1/2. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . —}. or — < 6 < . i. 6 6 In the case of the inverted double well potential under consideration here (i. replacing sin 30 by " 0 + ^) 1 IT = .18. we set z w r = \z\eie. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling.cos 36.68) .+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. (18. i.
Looking at Fig.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.68) for c 2 > 0. (18.Z\.21) to + infinity and to demand that they satisfy the condition (18. Anharmonic Oscillator Potentials V(z) Fig.398 CHAPTER 18. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. Eq. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. 18. we match the WKB 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. i.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). Our procedure now is to continue the even and odd solutions (18. for z > Too. We have to remember that we have various branches of the solutions y(z) in different domains of z. and there impose the boundary condition (18. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. We therefore explain our procedure first. (18.3 The inverted double well potential with turning points ZQ. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. We do this extension with the help of WKB solutions.21)) were defined in terms of solutions of type A which have a wide domain of validity.Z\.e. Our even and odd solutions y±(z) (cf. This is the boundary condition also used by Bender and Wu [18].
*R.71) We now come to the algebra of evaluating the integrals in the above solutions. h2 (18. 18. (18. The distant turning point at z\ as indicated in Fig.B. z\ ~ . I [195].*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . i.3 is given by (using Eq. Dingle [70]. Dingle [70]. Messiah. p.^ T rz\ 1 + yh* .z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . f R . t To the right of the turning point at z\ these solutions match on to (r.e. (22) or A. . 4 2 T ' i. (18. 6. 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\.69) The W K B solutions have been discussed in Chapter 14. ^ 2 4 •K + —(r.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. z2h4/A > c2z4/2.2..e. 291. ^2? 1 2qc2 /2? .4. From there or the literature* we obtain in the domain V > ^RE to the left of z\ as the dominant terms of t h e W K B solutions 1/4 ^WKB^ \qh2 Z\ + \z2h4 dz  \c2z4 1/2 „2 czz 4 x exp •r(I. We begin with the exponential factors occurring in Eqs. Sec. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.^ 4 1/2 + icV 7T Jz + (18. B. 291.70) where z < z\.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . equations (21).4) for E and ignoring nondominant terms) t o . p.18. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. Vol. A J24 1 ? 2 • "^ •z h \—c z ~ — q h .70).zi)/ \ 1/4 ywKB\ ) z — g/i* .4 h + c z 2 1/4 < .2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.
i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >.40)). (18.40)) E± = exp _/^_2 f ± In A 2c2 .2 \ 1/2 .\qh2 + \z2h± . (18. (18.73) .^ i ^ ^ c 2 .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.211/2 1 \ /j. zi 2q 1 c 2 [1 ..e.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq. CHAPTER 18. Thus the above factor yields '2c2\l/2 h4 so that (cf.400 i.2. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.2. Eq.69)) rzi J<z z dz Z[&Z*]W = + ^Y /2 m '"U4 1/2 2d2 + 2c2 Since we are interested in determining the proportionality of two solutions in their common domain of validity we require only the dominant 2dependence contained in this expression.^ I 2 2 1/2 zV = F 8c2 Jz [i . .^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g . Anharmonic Oscillator Potentials E± = exp = exp . (18.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . Eq.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).
78) Inserting this into the solutions (18. > » Returning to the even and odd solutions defined by Eqs.77) Now in the domain 2 — oo we have > \qh2 .Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J. (18.18.20b).*l) (3y^(z) (18.74) Comparing these solutions now with the solutions (18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.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.79) +S_(±)exp .*l) (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.71) and these into (18.(J*i) VA 0 ) = PVwKB (Z C.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.76b) apart from factors [ l + 0 ( l / / i 2 ) ] . h2 — — h2 is maintained.\z2h4 + J Z\ nl/2 \M fZ .76a) P P r 2/1 2 " 9 ( .77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) .20a) and (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.21 1/2 .
82b) This is our second condition along with Eq..^ § ( 4 g 2 + 29) + O ( ^ ) ..80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.402 CHAPTER 18./i2) = E{qQ. (18.6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues./ ? = 0. we see that we have to demand that 5 + ( ± ) = 0. i / 3 ± . Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. .34).e. i. (18.2.5. i.68).. E(q. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J ..76a) for (3 and /3.^ j . (18. (18.65).^(q2 + 1) . Eq. (18. (18. h2) = \qh2 . (18.h2) + {qqo)(^pj = E(q0.. 18.85) + .e.h2) + (qqQ)h + (18.3.84) Expanding about q = qo we obtain £(g.81) Inserting expressions (18. (18. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .
above) (1. Since this is not without interest (e. Example 18.e.h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .o 2 + 29) + o ( ^ ft6 \ 2qoh2[ —~ 2 ( ) < • 9o/ 2 ' 2c J ft 6 /6c 2 '(27r)V2[i(gol)]r • (18 86)  The imaginary part of this expression agrees with the result of Bender and Wu (see formula (3. where the superscript {r.h6/6c2 E = E(q0. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. o 2 + l)^(4. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. S W K B ( 2 ) = PVA(z). h2).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 .g ( .4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z. ft6 ^ = e. i.2 .3.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. !*WKB(Z) ~ 2 . 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.g.l ) ] ! > + = ^ 2 . zo} means "to the right of the turning point ZQ" .86) will be dealt with in Chapter 20. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).18. h2) of the relations KWKB (Z) =PVA(Z). Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.ZQ) .~p{q.yA{z).36) of their Ref.
cf. Expanding the square root occurring in ^ W K B W ' ^ W K B ^ ) ' Igft2_I22h4+lc2z4 2 4 2 and we see that 1/2 = 7 ^ 1/2 2." / 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.zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . Anharmonic Oscillator Potentials where zo > z. the (approximate) solutions* 1/2 exp • ± i ( . 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. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .ic224 2 4 qh2 + i^fe 4 .zo)/ \ qh2 2 ZO + z2hA 4 2 ./ ^ 2 X 1 / 2 exp(ifi. 2 z 2 ) / 2 e \ ..g In other words. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. N.404 CHAPTER 18. Schwind [247].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.O. f x Zn 4) h2 2. .1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation .] { hz combined with the particular constants contained in y ^ ' ^ g . 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.
z ± ) = .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. we shall employ basically the same procedure as above or.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.DfJ»v we see Comparing now the expressions (18.41) for yA(Z).e. (18. as in the case of the Mathieu equation.g. Then .89) . But there are significant differences. i.1)! = ( 2 7 r ) 1 / 2 e ." / 4 ( ? « / 4 .. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.zo).(r.^ 1 . „ / M 1 / 2 i ._ . We can reexpress these relations with the help of Stirling's formula.18.88) + [EV(z))y{z)=0 (18.13) of Bender and Wu [19].3. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(.J ' i J . in Eq. h4 > 0.3 18. (18. (18. in fact. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0.23)) appear quite naturally. 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.kyil(. 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.yA{z) with expressions 27WKB i J'WKB ' that the factors p. 4" ) (h?z2)^q+1) for q j£ odd integers.5 "l _ o f . (j .
vW(z±) = ±Qch2. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2. (18. .z±) y = 0. (18.91) becomes T>q±(w±)y(w±) = o(^)y. (18.406 CHAPTER 18.z±).95a) .4 The double well potential.i>5.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. 18.V(z±) l{z z±fhA + 0[(z . (18.90) V(i)(^)=0.91) (18. In order to obtain a rough approximation of the eigenvalues we expand the potential about the minima at z± and obtain y dz2 We set + E .z Fig. the previous equation (18.94) (18.93) w± = h±(z .
.95a) and (18. we obtain another by replacing z by —z. (18.2 T h r e e pairs of solutions We define our first pair of solutions y(z) as solutions with the proportionality y(z) = exp \h\ f uV2{z)dz (18.96) dz2 + 5 « 4 +  . given one solution. n = 0.3...2.96). we obtain d*y (18. (18.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.. d2u{z dz2 + v+ 2~r u(z) = 0.e.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. i.100) . where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].97a) U(z) = (zz±)2 + 0[(zz±)3}.18.95b) By comparison of Eqs. (18. is again seen to be invariant under a change of sign of z. Thus again. Inserting the expression (18. (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z).3 The Double Well Potential where ^ 407 d2 •« 2 2  d8. we conclude that in the dominant approximation q± is an odd integer.87). Eq. (18.93) for E into Eq.i 4 f W » = o. (18.1. 18. qo = 2n + l.97b) Our basic equation.
1 6 ) .4.e. (18.yA(z) are associated with one and the same expansion for A and hence E. h2.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. Aq(z) = Aq(—z) = Aq(z). z) = yA{~q.103) Looking at Eqs. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. 8hw (18. We leave the calculation to Example 18. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. i.^ A"{z) + 2c V2 *A(z) (18.101b) To a first approximation for large h2 we can neglect the right hand sides. h2.e.h2. (18. (18.106) .c W + 1) . z) = yA{q.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. h2 h4 2c' with g+ = q.408 Since CHAPTER 18.105) "8 + 4 ^ . Vq = {z2+z2)— + (qz+z).^q(l7q2 6 4/i + 19) + (18. these equations can be rewritten as + (qz+z)A{z) = . Integration of Eq.101b).104) Both solutions yA{z).h2) . yA(q. we observe that the solution y^iq. (18. The result is given by A and E(q. i. ~h2. h2. (18.101a). n± = V2h2. but the solution is a different one. z) may be obtained from the solution yA{q. z).c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .
106). (18. however. .108) We wish to rewrite this expression as a sum y coefficientiv4.{z2 .4zz+q+ z\{q2 + 1)].113) 2 .101a) with A replaced by 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) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. .3 The Double Well Potential 409 Example 18. We know the first derivative of Aq from Eq. Solution: The structure of the solution (18. + Aq+2 + Aq Aq+2 . for instance by componendo z z et dividendo. (18. ± 1 . i.109) We observe some properties of the function Aq{z) given by Eq. such as periodic potentials. (18.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 .*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.z2 ) N t. . as a linear combination of terms Aq+2i.e.105) and (18.103): . .18..114a) .102). 92 ZZJf (18.110) Aq+2+Aq2 9+2 A. J = 0.Aq 1 + 2^± 2 Aq+2 (^M^f (18. Thus it is natural to explore analogous steps.2 z\) • Aq$4 £Aq + Aq—4.z\) + 2z(z . (18. = Aq+2i+2jAq. First. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (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. z • . . . The first such step would be to reexpress the right hand side of Eq.qz+)2 . (18. ± 2 . <4q+2 — z — z+ Aq+2iAq+2j (18. we reexpress A'q in terms of functions of z multiplied by Aq.qz+) [2z2 .112) From these we obtain. A Differentiation yields » = ^^w (18.+ 2 ^ ± l + 2^±i + . (18.(l + + 2  z y* .
101a) the contribution R .116) [Ag4 . A(°~) = Aq.l ) 2 A .+6 + • • • ] . q) = 0.4Aq+2 + Aq+A Inserting (18. 1_4^±^+8^±±12^±^ +.4)A„_ 4 + (q.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.q + 2) = 4{q + l ) 2 .2A g + A g _ 4 ] .2\2 (z* .aAg+S 12+ 16 (18.4A ? _2 + 6Aq .115) (z*z%) 2 i2 (z 2 .116) into Eq. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+. (18. (17.117) 4(q + l)2Aq+2 Here the first approximation of A.4 4+* + .9=F4) = (g=Fl)(?=F3). Since VqAq = 0.33) and (17.2Aq + Aq4] 1. (q. (18. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4. we obtain 5 K = ( ^ f ) [ ( 9 .e. QA?+4 _ 10 A + 6 . . (18. cf. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.<? +{q.4]j. q + 4)Aq+4 + (q. leaves uncompensated on the right hand side of Eq.108). . i.q + 6)A<.l ) ( g .410 and from this = Hence with Eq.118) 2s/2c ~^j[(?.119) It is now clear how the calculation of higher order contributions proceeds in our standard way. •••].34).4 ( g .q + 2)Aq+2 (18.115) and (18.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq. (q. Anharmonic Oscillator Potentials A 1 .q)Aq + (q. (18.112). <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.112) CHAPTER 18. _ 4 . (18.114b) . (18. where the lowest coefficients have been determined above as (9.q2)Aq2 + (q.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .(0) V2 ' 2c K + h4+Ao . In particular the dominant approximation of A is obtained by setting (q.3 ) A . Eqs.
Terms jj.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.0)=0.z)±yA(q.0) = y^(q. for i = 0.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9. (18.e. (18. i.105). where Rq is the sum of terms left uncompensated by AW. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .122) The solutions yA(z). 9 + 2) Aq+2Az+ (9. (18.(0).h2.g + 2)(g + 2.0) = ^ . g .z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.h2.g) 4z+ {q q)+ 2r (9. 9 + 4) ^+48z+ (18.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9. g .e. h2. thus yielding the next approximation of A as given in Eq. y'_(q.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. A' (0) = q/z\) y+(q. (i) _ / V2c\ [ ( g . g + 4)(g + 4.123) ^[yA(q.{^ h2.yA(z) derived above are valid around z — 0.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g. z)\ = 2^A^ h2 '>z)± VA(Q.h2.h2. *)] (18. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.g2)(g2.0)=0.94) 8z+ (9. g) 8z 4 . y'+(q.94)(g4. in the domains away from the minima. h2. h2 2y^ Age? h4 (18.g) 8z+ (g. of course.4 ) ( 0 ) ( g .g) Az.2 ) (0) (18.g ( 1 7 g 2 + 19).h2. i.101a) the contribution Rq . (g.28.124) .
18.yB(z) is obtained around a minimum of the potential.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)).91). we > . Since w±(—z) = —h±(z + z±).93) and setting w± = h±(z — z±). Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2). This means. (18. (18. Anharmonic Oscillator Potentials Our second pair of solutions.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18.yc(z) with complex variables and Cq(w) given by Eq.125b) It is clear that correspondingly we have solutions yc(z). the equation is — with differential operator T>q as defined by Eq. 2(9+) .95a) and (18.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). (18.We draw attention to two additional points. (18.412 CHAPTER 18.1 ) D (18.Evaluating the exponential factor contained in VA{Z) of Eq.95b) that the solution there is of parabolic cylinder type. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. Inserting (18.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).2A y(w±). in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. yB(z). We see already from Eqs. + 0(hl2).99) for z — z±.23) with appropriate change of parameters to those of the present case. and another VB(Z) ± 25^ch3wl + c2wi . we have w±(—z±) = —2h±z±. but w±(z±) = 0. Bq[w±{ ^ .3.126b) These are solutions again around a minimum and with yB(z). Moreover. (18.^ % [i(?l)]!24(«.126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18. (18.
(18. Here z± = ±h?/2c are the positions of the minima of the potential.i*l(«^ 4'"+^+ 4"'+v ( « . a= ^ . the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4.• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . 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.97b) ~ exp .146).3 The Double Well Potential (cf.18. Eqs..164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18. 2 An+z+e 4w+ (18.« + ) 2 (9+1) e (18. (18. Allowing z to approach z+ in the solution IJA(Z). . . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. Thus for h very large these turning points are very close to the minimum at z+. (18.172)) we add here comments on this approximation.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18. ein+z+e in+(z 22+(<?+i) .z±)dz .127) Recalling that around arg w± ~ 0.frW.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq.129) .128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.ZI and find that these are given by z+ + 0(l/h) for finite q (cf. (18. Later we calculate the coordinates of turning points ZQ. .• • " . (18. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . .^/w + ± 1 . and comparing with Eq. 24(9D[i(gi)]! hj (18. .147)). (18.172) obtained later.
they assume large quantum numbers.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . J. since.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. g .+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . We have to impose boundary conditions at the minima and at the origin. K. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.3. H. S. and this means at both minima. even above the turning points. de Deus [66] and W. Furry [102] . Rau [22]. The involvement of these WKB solutions leads to problems.7r and the solutions of type C around argz = ±7r/2. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. 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. Bhattacharya [23]. we naturally expect the wave function there to be similar to that of the harmonic oscillator. Bhattacharya and A. but it is clear that none of the above solutions can be used at the top of the central barrier. R.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. Thus 1 E . 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. 18. Various investigations^ therefore struggle to overcome this to a good approximation.(18. K.z+)]fr+V ' (18. P. The solutions in terms of parabolic cylinder functions have a wide range of validity.414 and CHAPTER 18. The pair of solutions of type B is defined around argz = 0. D. Since it is more probable to find a particle in the region of a minimum than elsewhere. basically. The next aspect to be considered is that of boundary conditions.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. S.
Zii yB{z±) a ± a yciz±) (18. 18.136) .132) Fig. y(z±)=0.5. as indicated there.5.132). as indicated in Fig. The odd wave function then exhibits a correspondingly opposite behaviour. y'_(z±) = 0. (18. (18. 18.135) Hence the conditions (18. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable.134) y'+{z±) ± 0. as indicated in Fig. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.5 Behaviour of fundamental wave functions. 18. and y+(z±) = 0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.18. ^(*±)#0 (18. y{z±) £ 0.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. an even wave function can also pass through zero at these points. We have y±(z±) = ^VA{z)±yA{z)]z^. However.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.133) y+(z±)^0.
416 CHAPTER 18.«*{=(. also (18. _ 8)} s . 1 1 .142) (<Z9o)J(l)^°+1)(l)1/2 .134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 . Eqs. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.m Using Eq. . . (18. (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. 7 . (18..138) Using formulae derived in Example 18.„*{!<.2 we can rewrite the second of (18.136) as  M .5.. +1)} = 4 m. .54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula.. .9. (1. (18. Thus sin {^ +1) } = ^^>^30fe^i.136) the dominant approximations we obtain (cf.
Since the type A solutions are not valid at the minima.3 The Double Well Potential Thus altogether we obtain 417 ^ • . 18.z\.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. Hence we have to impose at z — 0 the conditions j/_(0) = 0. y'+(0) = 0. we must also demand this behaviour here along with a nonvanishing Wronskian.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. 18. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. We deduce .3. y+(0)^0.» ^ B ^ ' . we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. We do this with the help of WKB solutions. V(z) — • z E=V Fig. (18.1 " <813 1 4)  In Example 18.6 Turning points z$.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. y'_(0)^0.
e. + £\h%U(z)=Q. 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. In the domain 0 < z < ZQ. As a consequence.147) Since the minima are at z± — h2 /2c with /i 2 very large. one finds that . (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.418 CHAPTER 18. ZQ) means "to the left of 20" .\qh\ + \h%U{z) (18.^1 25c2 + V2 2 Thus again we see that for large values of h2 the turning points are very close to the minima of the potential for nonasymptotically large values of q. h8 25c2 (18.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.145) \qh\ + \zW Using z+ = h2 /2c.148) ("f * "The superscript (I. (18. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) . 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. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0.144).e. Anharmonic Oscillator Potentials from Eq. i.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18.
3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.* 4+ . 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) .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z . (18. we have to match yA(z). )+^*+)2lr} rnD h\z'2)}z Q )\ = .149) In order to be able to extend the even and odd solutions to z = 0. (18.^ j O e .i (2q ln (18.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .^ e therefore have to consider the exponential factors occurring in (18. . 20 dz \qh\ + \h%U{z) .148) and (18.18.^ + ) 2 + igln2(zz+).151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.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 >'^.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.
we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1.126a). This is the aspect investigated by Furry [102].153) This expression is valid to the left of the turning point at ZQ above the minimum at z+.125a).152b) [\(q 1)!] V2 K7: hi Here q/A was assumed to be large but we write [\{q — 1)]! since this is the factor appearing in the solution (18.420 CHAPTER 18. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. We see therefore. Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. In a corresponding manner — i. However. The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18. using Stirling's formula (and not the inversion relation) — we have —(1. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. 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. l/2 ~h\U{z \(zz+?h\ (18.ZQ) . t h a t the Stirling approximation is amazingly good even for small values of the argument).154) . the results necessarily require adjustment or normalization there in the q—dependence. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.125a) and (18. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. of course.e.
. and is shown to reproduce correctly the result (18.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. this result is somewhat imprecise. The relation (18.18. \hl.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.157) Since the factorials [^(q — 1)]!.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. multiplied by (1 + 0(\/h\))) the proportionality constants 7 . 7 of t h e matching relations y^NKB\z) i. (18.143) T^^*)) 27T (18.153). it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.158) Applying the boundary conditions (18.5 for the calculation of the tunneling deviation q — go by using the usual (i. as the results also seem to support.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.99).e.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.n('>2o) 3/WKB(Z) =1VA{Z)I (18. Returning to the even and odd solutions (18.157) is used in Example 18. .103) and (18. (18.159) 7 i(«i) r(9^3) ^2~ j(9l) .154) with the typeA solutions (18. However. linearly matched) W K B solutions. 27T1/2 9/4 1VA{Z). Comparing for z —• z+ the W K B solutions (18. (18. we obtain in leading order (i.e. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.155) (2z.e.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.K?+1).104). (18.
(18. F. n + uY 2c „ ^6 8V2c2 (18. 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 (18.161) we can rewrite the integral as r I2{z) = = dz^{a2 . D.t The elliptic modulus k (with k' = 1 — k2) and an expression u appearing in the integral are defined by b2 1 U kz = ^ = l + u so t h a t = u=S^q hi = GV2q. 3G2 12V2c ^See P. formulae 220.149) near z = 0.163a) (18.k'/2.iV2 ! i . .164) / 2 (0) z ca [(1 + k2)E(k) .3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. Friedman [40].422 CHAPTER 18. Byrd and M.. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. G2 K [i 4.05.z2)(b2  z2).148).19. p. 60 and 361..163b) The integral hi?) . 213. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18. . p.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. (18..
2. . n.. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . .± ' 3G 2 ' T ( ^ (18. /2n + l 2 ± — . [4] misses an n—dependent power of 2 in the result. ± .1 " e " O T (18 169)  Y^V / 4 tfo.. which implies small G2.1)]' Gfo. 1.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. . (G\ .3 )l ! . The nontrivial expansions are derived in Example 5.3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .18.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . ^ ^ . [167].§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. 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 .) = 2 \.( ft 7. where the result is shown to be (with q ~ q0 = 2 n + l .. n = 0..1.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . .165) in agreement with Ref.
143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.106). and obtain ( * .1.172) into Eq. .3.175) ~E0(q0..»W.h2) and hence the splitting of asymptotically degenerate energy levels. (18.157).424 CHAPTER 18..i ) ] ! e ~ s A ' .6 Eigenvalues and level splitting We now insert the replacement (18.174) .143) with qo — 2n + l. we can therefore impose the boundary conditions at z = 0 by making in Eq. .172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z. (18. (18. Inserting the expressions for h+ and z+ in terms of h. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.. . (18. 2 . (18. .n = 0. h2) the split expressions 29o+1/i2(—W2 h6 (18. we obtain for E(q0.h2) + (qq0)^=. (18.. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. with the help of Eq.173).^ W ^ / 4 t o . (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). 18.3.159) — 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. 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.3 concerning the exponentials. (i*™) We obtain the energy E(q. 18.
e.157).^ 2^ h* .7 =( e N n+1 . The result (18.. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 .176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.e.174) and (18. 2h4' i. for finite h2. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency. i n U n / ! Thus AE(q0.106). K.2 = 2 .. 1 u2 c 2 (3 9 2 + l) ^ « b .2.h2)E+(qQ.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.e. the level splitting.143) with (18. the pure WKB result of Banerjee and Bhatnagar [14] (i..h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. (18.1.n = 0.h2) = E_(q0..fr2) is given by Eq. i. which evidently supplies some correction terms. i.e. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k. 1 7 425 2 .178) "In S. . T \ *n ! 1 ' (18.18. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.V2^\ \n+ i . the following expression for large q ~ 2n + 1.q2 7.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .3 The Double Well Potential where £0(<7o. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .e. (18. Combining Eqs.
(18. . supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. Then the splitting A ^ W B of Eq. we would have obtained the result with E. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». Furry [102]. A>0.6) to yield an improvement of WKB results for small values of n. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. and A there with c 2 / 2 here.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. (18.176) together with Eq. as is also explained by Bhattacharya [23].180) h8 E0(qo. Bhattacharya [23]. the Furry factor corrected WKB result follows automatically.181) "W. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. however. h = 1.6 \9o/2 e~h&l^\ with (m 0 = l) (18. H. h2)\mo=i = .3)). We see therefore.426 CHAPTER 18.2X2 V(z) = \z2tL) . as we do here. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0.h2 <> — h2. Eq. (18. This. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. h4 and c2 replaced by 2E.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).178). K.4/4 here." Of course. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.^ 2 + ^h2 If in addition the potential is written in the form A / .. we identity the parameter k there with ft.2 / 7. 2/z4 and 2c2 respectively (see Eq. ** Comparing the present work with that of S. (66). 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. (18. that if this symmetry is taken into account from the very beginning. 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.
provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions.^ *See E. §J.181) therefore implies the correspondence rj1 <> A/2. J.c2 = A/2. Patrasciociu [110]. Jentschura [293]. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. m 2 <• p 2 / 2 .182) implies AlE(l. formula (4.34) and (E. (18. Banerjee and S. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. ^To help the comparison note that in J.W.11).18.§ 18. J. Liang and H. first paper. ZinnJustin and U. P.^ Thus for the case of the ground state Eq. J. See R. We considered above only the symmetric twominima potential. so that by comparison with Eq. The comparison with Eq.* A similar correspondence is also found in the case of the cosine potential.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods. Mansour and H. Jentschura [293]. however.15). M. K. . these are presumably not of much interest in physics (for reasons explained in Chapter 20).3 The Double Well Potential 427 a form frequently used in field theoretic applications. D. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. Friedberg and T.—Q.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. D. MiillerKirsten [186].88) h4 = 2/x2. Lee [98]. Gildener and A. (2. ZinnJustin and U. (18. and g2 is given as g2 = r]2/m3.3. Bhatnagar [14]. (18. D. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. M. W. Equation (18. The asymmetric case can presumably also be dealt with in a similar way since various references point out that the asymmetric case can be transformed into a symmetric one. Eqs.h>) * 2V2^^^y/2e^3/3A. H. the level splitting is (with h = 1) This result agrees with the ground state (go = 1) result of Gildener and Patrascioiu [110] using the path integral method for the evaluation of pseudoparticle (instanton) contributions.
[131]. [19] — that the expansions considered above are asymptotic. Mahapatra. The double well potential. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. 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. 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. like those for anharmonic oscillators. can be found in an article by Coleman [54]. Very illuminating discussions of double wells and periodic potentials.g. Perturbation theoretical aspects are "See e. [132] and Loeffel. Simon and Wightman [178].428 CHAPTER 18. has also been the subject of numerous numerical studies. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. though not exclusively numerical. Martin. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. we cite papers of the Uppsala group of Froman and Froman [100]. B. B. In fact. N. Santi and N. 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]. mostly in connection with instantons. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. as some relevant references with their view we cite papers of Harrell and Simon [129]. Ashbaugh and Harrell [12] and Harrell [130]. 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 ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. in both symmetric and asymmetric form. As references in this direction. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. A reasonable. Pradhan [182]. for instance. . P. There is no reason to view the anharmonic oscillator differently. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. as he discusses. 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].
+ (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. Example 18. Bender.. We have 4 ~k' 2u(l . Saxena. 2 a.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18.f + • • •) 2u\ ./4 E{k) = 1 + • l n In (18. X +2 + ' ^ (18. k! (18. (18.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. Cooper and Strottman [221]. We obtain the expansions from Ref. T h e r e f o r e it is convenient t o deal w i t h this here.186) (18. Srivastava and Varma [24] and Bhatacharya and Rau [23].\G\ + • and 1k2 =2Gv/2q4. which is assumed to be small. Gutschick.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . Datta. Thus k' and hence k' = v ^ ( l .183) S o l u t i o n : We have k2=1^. (18.2G^2q + 4G2q .189) =2u 2u2 (18. 1+ u : Gy/2q.G2q + .u)1'2 = V2u{ 1 . Wave functions of symmetric and asymmetric double well potentials have been considered in the following reference in which it is demonstrated that actual physical tunneling takes place only into those states which have significant overlap with the false vacuum eigenfunction: Nieto. [122] as 13 1.192) 'F 12 16 . q ~ 2n + 1.188) 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 .18.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14].187) We now reexpress various quantities in terms of u. Biswas. fc ~ 1 .^ + • (18.
3u(6u .u)} u{u(l .u 3 ( u .3 y .13)}.195) With (18.3 u ( l .3u(6u . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .3) .4.3 « .u){24 . (18. (18. 8 + 6 u 2 .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .3u(6u . 1 3 « 2 = .u){24 .194) and (18.193) 2u(l .1)(1 .3) 1 + In ^ .195) we obtain E(k) 1 + ln uK(k) .3 u .3).u)} .{l u ( l u) / W 2 + 2} .1)(1 .18u 2 + 39M} + 12u 3 (u .3u(6« .196) Now consider the last line here without the factor 2.«) + 2} + V [ .2{«(1 .u)(u .u) + 2} + i u ( u .13)} + 12u 3 (u .430 and CHAPTER 18.u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .{ M 2 ( 1 .1)(1 . .u){24 .u ( l .u) + 2}] 1 H 1 o u(u .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 . up to and including u2): u(u .3 .6w2 + 13u} + 12u 3 (« .u ) (^){« + 5^ 1 )} + 5.^ L ) H[(i _ U ){4 + 3u(l .13)} + ±u3(u .u){4 + 3u(l .3) = = ~ ..u) 2 {8 .197) .« ) ^ { 2 4 .a = (l«)(jl 2u/ 2 ) .1)(1 .e.13)} + ~u3(u .« 2 + 3u] In [ .— ) . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.u)} + — u ( u .u) + 2u} + i u 2 [ 2 + (1 .194) ln '^J + 2 In i/2u )+H l + 2u(l . 8 + 9u 2 .3) .u) 2 {24 .u 13 \ 1 ln(± k> fc'2 +  (18.u ( l .6 in the denominator and in the last step pick out the lowest order terms in u (i. 8 . (18. ) J 4 (18.
J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.it)} .18.128) and (18. . / to + 3GH~ \ 8 .196). (18.148) (for particle mass 1/2 and in dominant order) is given by H. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .3). Eqs.u H I .156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf...128) and (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .^HfHGH Example 18. (18.2 l_ln(1 JL\ V {2. Inserting here from the above expansions the contributions up to and including those of order u2.2 (18.2{w(l u) + 2}] = u(3u2 . From (18. [(1 . where w±(z) — h±(z — z±).196) and (18.3 The Double Well Potential Now consider the bracket [.197) we now obtain 4 1 • E(fc) .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18.uK(fe) ~ 1 + In I — = Mu2(3u..2 /o 2 431 (18. 2 . VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .zo) HWKB normalized _ ( h* y^eM. V^LJ 4 2 16 ^ . + iu^Zu2)^~u+ — +0(u3) l .199) We now return t o the expression on the left of Eq.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).173) t o be evaluated. i.e. ' 8G 2 q = 4 2 q ln f 26/2 \ o . l2 = ^ ( l + uf/2[E(k)uK(k)}. i.198) 9 «.200) Remembering that u = Gy/2q.] in (18. (18.195) with (18.4u .e. we have.3) . (18.4u .«){4 + 3u(l .
7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. (26. (18. normalized = We see that the normalization constant is independent of a quantum number. Then ((r.** oo / dwD\. ° \Edzy/\EV(z) V(z)V4 VWKB. and obtained as Eq.f hi\ VWKB. (18. We start from Eq. oo . 93.l)]![i(0 . Oberhettinger [181]. Eq. by using the periodic WKB solutions below the turning points. Magnus and F.Aw) = s/2i 2("!) V (41) and [(9 .147). 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 [ .*o)/ ^WWKBOO (18. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf.zo). (18. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. pp.35)).e. Example 18. Solution: The turning points at zo and z\ on either side of the minimum at 24.204) .g. W. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.e.143). 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.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+. i.hA/2 = h^/4 this implies . 1 7(l.432 CHAPTER 18. 82. (18.158). i. 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 .146) and (18.3)]!23<«D [5(91)]' >A./ . are given by Eqs.
r + ^=co S [r. from z\. z{) we expect tt (18. 7T\ + ^ 4 7r\ dz .V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) . approach the minimum also from the right.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. A. Then at any point z € (zo.e.g.2.iK h\U{z) + / 4 (18.J^Wj +(18..6.e.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.g. 1/2 . Messiah [195].209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.( . Sec.18.211) 7 z l . (18.2.. \h\U{z) 1/2 . JV = 1.3 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 ..e.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.. i.^r + + .133) at the minimum z+ and obtain the conditions: 0 = — sin 7 and 0 = — cos 7 Hence £ + 1/2 dz(±qh\\h%U{z) 4_ ± — cos 7 />G^ 2 2 +i4c/(z)) 4 (18. i.212) tt See e.132) and (18.207) (18. of course. 6.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = . We could.. this implies sin cos sin cos Thus e.210) Choosing the point z to be z+. . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. from ZQ. i..205) We now apply the boundary conditions (18.3.
Banerjee and S. in fact (with our use of qo). The formula thus agrees with that in the literature.434 CHAPTER 18.5. (18. 2IO/K. and cot< (q + 1) '!}» for q = 50 = 1.216) Inserting this into Eq. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). K. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. S.215) we can insert for 7 / 7 the expression given by Eqs. D.Zi / r.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. . ._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. 7r"y in agreement with Eq. i.205).2 . the level splitting is given by AE(q0. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). 7 .149).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 (18.148) and (18. Bhattacharya [23].215) which results from the factor of 2 in front of the sine in the WKB formula (18.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . Bhatnagar [14]. Landau and E. In Eq. P. (18. and L.3. W "Cf.210) assume a form as in our perturbation theory.214) {«+"i} 0 for q = go = 3. .143).e.146))..213) The integral on the left can be approximated by 1. M. (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. ' /^i . (18. We now see that Eqs.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18.209) and (18. Lifshitz [156].. The prefactor 2\/2h2/n is.215) we can expand the left hand sides about these points and thus obtain . (18. (18. 5. where u> is the oscillator frequency of the wells.215). 1 1 . . K. they become c o t { ( . (18. 9.90 — =F— for q0 = 1.27 9 . + l ) j } ^ 1_ 27' (18.
435 . H. W. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. Giittinger and H. 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. Aly.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. in an early investigation Case^ showed that potentials of the form r~~n. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. Case [45]. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. n > 2. M. J. f K . However. also in P. are not as troublesome as one might expect. 'For a review from this perspective with numerous references see H. the investigation of singular potentials in nonrelativistic quantum theory was motivated by a desire to obtain a better understanding of the (then presumed) singular nature of the nonrenormalizable weak quantum field theory interaction. 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. Morse and H. MiillerKirsten [9]. W. M.* However. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Feshbach [198].1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state.Chapter 19 Singular Potentials 19.
J. 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. S. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. Hashimoto [124]. MiillerKirsten and Jianzu Zhang [226]. It is therefore natural that we study this case here in detail. Singular potentials arise in various other contexts. Maldacena [41].436 CHAPTER 19. (19. D. W. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. In the concluding section we refer to additional related literature and applications. S. Callan and J. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. Recently.2. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. for instance.2 19. J.Q. J. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. . S. some also permitting further research. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. Liang and Yunbo Zhang [189]. H. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. MiillerKirsten.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.e. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. § There one could visualize this scattering off the spherically symmetric potential as a spacetime curvature effect or — with black hole event horizon zero — as that of a potential barrier surrounding the horizon (shrunk to zero at the origin). W. G. that with the hyperbolic cosine replacing the trigonometric cosine). Park. R. N. D referring to Dirichlet boundary conditions) in 10dimensional string theory. In fact. Manvelyan. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name.2) § C . 19. K. H. Gubser and A. Singular Potentials theories in mind.
. H. (19.4) first for small values of h2 and then for large values of h2. J. 19. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . r = 7 e z . J. and as we have in mind in Sec. It is clear that we draw analogies to our earlier considerations of the periodic potential.g. Eq. Liang and Yunbo Zhang [189]. We observe that with the replacements p2 z^iz. Manvelyan. 8.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. W.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17.19.3 below) g2 is negative and hence h? is real for E > 0. (19.( I + 0.3b) 7 = % j. This has advantages compared with higher dimensional cases. (19. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. In the following we study the solutions of Eq. The following substitutions are advantageous: y = r1/2cf>. using the method of Dingle and Miiller of Sec.7 that we employed also in previous chapters. ig h \j g (19.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation.mo = 1/2 and h = c = 1. (19.28) below). \[\ in the correspondence is nonintegral.Q. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. and so =4 5 and n ez = r = ~ = e^4fir. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. h = ei^y/k^. 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. This is rarely possible and therefore this case deserves particular attention.3a) In the case of the attractive potential (as in the string theory context referred to above. h2 = ikg. R. MiillerKirsten. It may be noticed that since I is an integer.
The zerothorder approximation = Zv[w) (19.11) We follow here largely H. where £>„ : = —_ + i _ + J l .4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .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. (19.Hv (w).^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. The solutions Zv(w) are written Jv(w). MullerKirsten and N. 8. Aly.i l .438 19. — Zv+i).1. . (19.2.2 A / i 2 . H. VahediFaridi [10].Nv(w). (19.5) so that Eq.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. Eq. H. First we make the additional substitution w = 2/icosh z. where these are the Bessel function of the first kind.e.I .. (19.9) leaves uncompensated on the right hand side of Eq. —Zu — (Z„_i + Zu+i). w 2 . W. (19.6) terms amounting to 1 „ 2 d2Zu A V (19.2 CHAPTER 19. dwA w dw { wA ) (lg. i. Singular Potentials Small h2 solutions in t e r m s of B e s s e l functions We now develop a perturbative procedure" for solving the associated Mathieu equation in the domain around h2 = 0.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. Proceeding along the lines of the perturbation method of Sec. (19. and H„ (w).g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.7. J.
u + 2)*Z„ +2 ]. .u + 4fZu+A}. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.2 ) * ( z / .13) therefore lead to the first order contribution jfU = h2[{v.17) Thus a term fiGu+a on the right hand side of Eq. v + 2)*Z„ +2 (19.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19. v .16) ( I / .13) (19. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately).13).6). (19.19.2. (i/. v + 2)Gv+2]. z . and this means in (19.u (19. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.10) in terms of functions G„ defined by Gv+a = —nZ v+a .14) (19. = §.2 ) * i ^ 2 + (!/. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . it is more convenient to use these relations in order to rewrite Eq.. The terms (19. when a or 2v + a = 0.20) 2)*ZV_2 + {y. However.2)*Zi.i/4)*Z 1 / _ 4 +(u.21) +(i/.i/) = 2A. where the starred coefficients are defined by (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. (19. wz The expression (19.!/+ 2)*4°J 2 ].v + 2)*{u + 2._2 + {v. v + 2)*(v + 2. I / ± 2 ) = 1.2)G„_ 2 + (y.15) (19. (19. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19. of course. where We observe that DUZV = Q. v2)*{y2.12) (19.10) is now particularly simple: i?i°) = h2[(v. . v .10) as a linear combination of various Zv. v)Gv + (y.18) Now </>(°) = Zv left uncompensated Rv .
23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . etc. we obtain t h e expansion . v) +(i/.2. . .. . . Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . in Eq. Adding these terms and setting the coefficient of Gv equal t o zero. .26) iy — 2.4 + 58. i/ + 2j). Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e.2 + 2 9 ) ^ 64(^ 2 . + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9.j^0 (19. P2*o(0) = 0. j=i.2)(v + 2j .v + 2y{v + 2.v + 2)*(i/ + 2. i/ ± 2)* (i/ ± 2. (19.g.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . and p 0 ( 2 j ^ 0) = 0.I/±4)*. i/ ± 2)*. v — 2)*(v — 2.1169)/i 12 . i/)] + • • • . we obtain 9 h4 = a 2(al) 3 2 (13a . p4(±2) = (i/. v ( 4 5 a . we obtain 0 = h2{v. v + 2)*{v + 2. i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.440 CHAPTER 19.v) + hA[{v.22) where p2(±2) p4(±4) = = (i/. Evaluating t h e first few terms.455a + 1291a . 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . p 0 (0) = 1. Ri.i/±2)*.v)] (19. (19.I/±2)*(I/±2.17)). (19.l)(z/ 2 .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.25)/i 8 32(al)3(a4) . v + 2j) (19.v) + {v.24) P2i2(2j + 2){v + 2j + 2. subject to the boundary conditions P2i(2j) = 0 for  j  > i. (I/. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. .4)(»/2 . ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . .v2)*{v2.
v)(f>v + {v.cosh 2z)4>v = = 2h2A(/>„ .i .33) = = cosh(^ + 2)z + cosh(^ — 2)z. (19.h2[(j)u+2 + 4>v_2] h2 [{u. .29) sinh vz or e±uz. v .An(u + n).4). apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. (19.cosh 2z)4>.30) £>„&.19.2. we can rewrite the latter as j ~ . — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. (19. (19. The solutions 4> of the modified Mathieu equation are now completely determined.32) (19. 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 .^ + e±(v~2)z.v + 2)<t>v+2]. and is therefore real for both cases g2 > 0 and g2 < 0. (19.31) It follows that Dv+2n4>v+2n = 0.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. = 0.v2(j> = 2/i 2 (A . 2 .2)<t>„2 + (v.34) . Thus to O(0) in h we have (jr ' = 4>v — cosh vz. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. ±2.7) into Eq. so that J2 (19. Thus v is real for small values of  h21. £>„ = — . we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . 19. Substituting (19.
4 Notation and properties of solutions We now introduce standard notation as in established literature.h2i *=1 J2 j=i. (19. Singular Potentials (I/.h). For rigorous convergence and validity discussions of any of these solutions — our's here differ only in the method of derivation with the perturbation method of Sec.37) where p M (±2) = (I/. Schafke [193].h).442 where (I/. > These solutions correspond in the periodic case respectively to the solutions and me^.38) (f> with (\>v = exp(vz) — Meu(z. • (19. F.J/±2) = .h). 19.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. 8. 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. .J7t0 p2i(2j)<f>u+2j. (19. v ± 2) etc.1 . Meixner and F.7 — we refer to Meixner and Schafke [193]. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.h) = <(>„ +Y.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. In order to avoid confusion arising from the use of different equations. The use of the symbols (v. The solutions in terms of cylindrical functions are written *J.I/±2)*. W. M. Arscott [11]. etc.I/) = 2 A . CHAPTER 19. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. > (f> with </v = sinh vz — Seu(z. we prefer to adhere to one equation with different solutions.2. 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.
443 (19.h).h2) oo = Y.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. see this reference p.g. Magnus and F. Also from the expansion of Jv{z) in rising powers of z.19.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant.h). W. 4>v = Hv (2h cosh 2) * M^\z.43) and hence similarly M^\z + inir. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. i. (19. 'See e. W.40) does not. h). (19.40) Cev{z.41) i=l We see immediately that Meu(z + niri. as Mev{z. 16.e. h) = exp{vniri)Me„(z. h) are therefore proportional to each other as a comparison of Eqs. h) = exp(inwn)MP(z.h) and oo i M^\z. (19. h). this normalized solution possesses contributions exp(i/z) in higher order terms. (19. (19. as we remarked earlier.h) = exp(vz) + Y. .44) (19. ' (z. For the relation of Hankel functions used below.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). Oberhettinger [181]. whereas the — as yet — unnormalized solution (19. (19.MP(z. h).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) . 4>v = Hi. Meixner and F.42) The solutions Meu(z. Schafke [193].h). Mi. h). — oo cv2r{h2)e^+2r>. <f)v = N„(2hcosh z) MJ?\z.45) T Below we frequently write the normalized solution as in J. h) = av{h)M^ (z. i.e.(2hcosh z). 17.42).h).h)±Sey{z. {2h cosh z) • MJ?\z. Me„(z. p. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. As emphasized in Chapter 8.44) implies. which.
46) H^z) = e^H^(z). the first relation is obtained as follows: 2M<£1 = M^l+M^l. Singular Potentials where clearly MeJQ. Meixner and Schafke [193]. p. h). H™(z) = e^HfXz). Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).e.h) on the other hand can be shown (cf. ^ . (19.h) Using further properties like (19.8. oiv{h) = (19. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. These points are important in our derivation of the S'matrix below. i.49) Inserting this expansion into the modified Mathieu equation (19. .47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. p. For later essential requirements (i. s (19.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4.48) The series expansions of the associated Mathieu functions M„ (z.e. h) = M™(z. 2M™ = eiuvMi3) + e~ivlrM^\ . h) . As mentioned earlier.h2) = ] T cv2r(h2)e{y+2r)z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.h) can be shown (cf.444 CHAPTER 19. 130) to converge uniformly for all finite complex values of z.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0.h) . Mi 3 . Meixner and Schafke [193].43). MJ>>(0. The functions Me±v{z.e.g. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. we have the following for a change from v to — v. r=—oo (19.50) E. i.exp(TM"r)MW(.h2): oo Mev(z.
(19. eg = c~v = 1. h~2[\ .T22r[\ A .(/^.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . p. Eq.[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. 117. Meixner and F.51) we have 2r (19. see J.^ C 2r+2 (19.^ p 2 2 ^ or c 2r+2 2r2 1 (19.^ + 2r + 2 ) 2 ] . .19. 122. (39). . \ . Schafke [193]./ i 2 C ^ . + 2 r ) 2 ] . + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(.(v + 2r)2} / z .^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . p. \) /.53) /i2[A(^ + 2 r ) ] . 2r (19>51) 2r For ease of reading we give the steps in rewriting this. W. Thus now c 2r .54) 2 C2r 4 ^ h.* For 'Actually.52) in agreement with Meixner and Schafke [193].2 [ A .
55 ' Example 19. MiillerKirsten and N. and also R. H.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. Aly.C + l) 3 (i/ .+0^"> < 19 . + i). (19.. J . VahediFaridi [10] (this paper contains several misprints which we correct here).12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. .l ) 19.446 CHAPTER 19. Manvelyan. J.5 D e r i v a t i o n of t h e S . but follow our earlier reference H..8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected.m 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.Q .55). t For this purpose *We do this in the manner of the original derivation of R. W. Then di 2 1 t" + 2(^1) + • • • .2.1 for the evaluation of a typical case): hb + 0(hw).^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10).27) and truncate the continued fraction after the second step. Liang and Yunbo Zhang [189]. Spector [257]. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . J. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. (19. M. Here we insert for A the expansion (19.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. + 0(/.( " + 2)2] ~ h2 h. H. H. MiillerKirsten.[v^.+4)!ll_. 128(i/+ l) 2 (i/ + 2 ) ( ! / . . Solution: We put r = 1 in Eq. W. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .
59) which tends to zero with r. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). Magnus and F. In the case of the repulsive singular potential here.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. 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..5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. Then y reg = = r rV2M®(z.19. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. (19.h) 1/2 w) (19.3a) and (19. 22. . The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. (19. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.j7L0 P2i(2j)exp i[v + 2j + 2 2 .57) ^ 1 . Oberhettinger [181]. and then the continuation of this to infinity. p.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 .56) where by Eqs. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. I / + 2 2 l\TT +E ^ i\ E j=i. we have W r = ip + ut — 7r/4' . and considering the propagation of this wavefront. W .
(19.62) Also with Eq./i). but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.#. Magnus and F.j¥=o Ki(2j)exp(Tmj) (19.j^Q p2. '2"4 . Meixner and Schafke [193]. 17.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i.60) Using the above asymptotic expressions for the Hankel functions. The series M^ (z. . Oberhettinger [181].h) = exp(Mr^)il4 4 ) (z. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.V + § 9 2 > 4hh* . h) through the entire range of !&{z).A1"—2 z '+"2 A~~ n or rA .h) can be shown (cf.3a) and (19.(2j)<gH • (19. (19. 4%r g >0 4r^r ' rg See W.A14A. 178) to be convergent for  cosh z\ > 1. AV . Singular Potentials so that when t — oo : r — 0./l ~ 2 ~ 2z + " 4 > 0. Now from Eqs.3b) ln • 7T Q"*4' r fg ° Vk (19. '(z.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. 2h and the square of this expression implies 2 2 ~ ' A.448 CHAPTER 19. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. p. p. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). This means that the origin of coordinates acts • > as a sink. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. We require therefore > > the continuation of Mi.
Originally we chose the sign of m / 4 as in Eqs. (19. i Im z z = In r/r0+i7t/4 .38) or (19.3a). Since the real part of z changes sign at r = ro. (19.rl)(r2 . . These solutions converge uniformly for all finite complex values of z (cf. i.63) Thus there is a gap between the two regions of validity — i. (19. This interchange in the > solution My interchanges y reg with y+. 19. is satisfied only for r < r_ < r__ or r_ < r + < r. (r — r+)(r + r+)(r — r_)(r + r_) > 0. 130).1.1 The domains of solutions. and in order to maintain this symmetry.62).2 The Potential 1/r4 — Case of Small h2 Hence (r2 . 19. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. Meixner and Schafke [193]. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. p.19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig.49). the domain r_ < r < r+ — as illustrated in Fig.3b) and (19.ijt/4 Rez . we have to assign different signs to the imaginary part of z in the two distant regions. 449 This condition.e. which has to be bridged by using another set of solutions. A suitable set is the pair of fundamental solutions Me±v defined by (19. r 7r for 0 < r < ro. Thus we choose .r2_) > 0 with r£ = rg(2 ± >/3).e.
as just explained. i.[ r l/2 M.v{z)\ . At r = ro the real part of z vanishes and then switches from negative to positive.66) a = (3 . to r^2[AM^ + BM^\ A. at this junction we require also the imaginary part of z to change sign.(2)]z=w/4 + 0Me.450 CHAPTER 19. Hence we match the right hand sides of Eqs. (3 = a . and we determine these coefficients from this equality and the additional equality obtained from the derivatives. h) = Me^u(—z. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.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_.e. Thus. i. h) (as one can check or look up in Meixner and Schafke [193]. . (19. 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.z=—in/4.^)] dr z=in/4 dr (19. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. (19.B^0.v(z)\ —m/4) + ^[r^Me.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r.l/2Me_i/(T. Singular Potentials Then starting from the region r ~ 0. 131). Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. p.
70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19. (19. A and B. dr BM^(r)].g]:=f(z)*&g(z)WZ) dz dz Thus.68) From Eqs.64) the relations M&Hz) jz[M^\z)] and from Eqs. (3.70) by Me'_u (the prime meaning derivative). (19. (19. with the replacements of Eq. = aMev(z) + 0Mev(z). and we are left with relations expressible only in terms of z.71) We now determine the coefficients a. (19. ^ ] = ^ + ^ ^ .2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. Then. (19. (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}. (19. (19.69)) cancel out in view of the nonderivative relations. Thus we obtain from Eqs.19.69) Thus if we replace in the derivative relations of Eqs. (19.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . multiplying the first of Eqs.65) and (19.3a) and (19. (19. the nonderivative parts (from the first term on the right of Eq.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. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. we use the notation W\f.65). (19. and the second of Eqs.68) the derivatives with respect to r by this relation. for instance. (19.70) by MeU and subtracting .3b) we infer that (z = l n r + const.
we obtain immediately W[Z. p.g.46).. and Wronskians W[M^\MIJ)] = [i. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19. h).(0. W[M^\Mlf)]W[Meu.47) we obtain (av defined by Eq. TV W[Mev. (19. (19.73) We now use Eq. M^4)] + W[MJ>3). and the second is obtained similarly: W[M?\Mev\ W[Mev. Me.Me^} (19. we obtain the first of the following two equations.48) (or cf. 7T [1. (19.j] given in Meixner and Schafke [193] (p.452 CHAPTER 19. using Eqs..u.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. i. i. .e.. Meixner and Schafke [193].4] = . 170/171).Mev] = = = av.71) we obtain in a similar way A B = = W\MP\ Me.e.46)) W[Mev. B = J 2 L _ e . (19.M^} W[Me„. 7T (19.^ 47 9? MJ>4)] MeV] [3. or obtainable by substituting e.] W[Mev.75) With these relations and Eq.\ \ ( av 2iwKav ^ v .Mev]' P l ' From Eqs. (19.M^]= IT e~iv*av.58) and considering large values of \z\.3] = [1. = IT <*„. .77) av I 2ism VK \ ctv av ^Thus.A] = W[Mi3\z).57) and (19.74) Moreover.^ ^ a „ Q _ . (19.4] = . we use the following circuit relation which can be derived like Eq.* « * r r z H . W[MJ?\ M^] W[Mev. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a. (19. av{h) = Me. Singular Potentials the second resulting equation from the first resulting equation.M^] W[Mev.u}W[Me. the leading terms of the respective cylindrical functions.M^] J™aV. (19.Mevy W[M?\Mev] W[Mev. h)/M^\o.76) . 7T W[Mev.
82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#)./ ( ./) = (l)1/2R2^2i^)ei^)U^. B=tfD'"".85a) . Z ) e ^ ] .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).46) together with the relation Mev(z.f c .80) 2^fc (1981) where \7r/ i t sin VK /(*. Z ) e . (19.(ite™)" 1 { 8 J ' f(k.h). eii/7r/2 (19. (19._ [ / ( f c .78) we obtain Ay reg « _ . The superposition of an incoming plane wave and an outgoing radial wave is written and reexpressed in terms of partial waves for r — oo » with z = rcos 6 as: 1 tCT" "1 OO e *« + /(*)!_ ~ 5 > ' c < f c P . (19. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. Then f' .h) + BM^\z. as we recall from earlier considerations (e. in Chapter 16).19.g. (19.83) The Smatrix element is therefore given by ifc*"* .h) = Me_u(z.79) where the second expression follows from Eq.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). (19.l) ' We note already here that with the substitution R = ei7r7. (19. f c r .
Fig.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. V(r)=g2/r Fig. 9?. we can define respectively as amplitudes of the incident wave. 19. and with the left hand side following from Eqs. (19. We can rewrite Eq. R — — = 2i sin ivy.2 The repulsive potential and the various waves. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). Thus in terms of the (3) variable z. h) (see Eqs. (19. R 2i sin TXV. we can write this (multiplied > by 2ismvTT. cf.86) .56).2. 19.58)) in the limit r — 0.454 we can rewrite the .58)) 1/2 2r e~ 2hir cosh z. (19.56) to (19. Singular Potentials r(i+k) sin 7r(7 + u) (19.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later. and recalling that yreg is proportional to a function Mc> {z.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. (19.Smatrix in the form s i n 7T7 Si = CHAPTER 19. (19.
l=—oo 1 1 lite*"* .In Meixner and Schafke [193] (p.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. this expansion is given as 1 oo MJ> Hz. if R becomes a pure phase factor and eil/7r a real exponential. Eqs.e™*/R\2 ' ' . (19. i. however not here. Liang and Yunbo Zhang [189]. (19. Manvelyan.e.e. unity minus reflection probability = transmission probability. Gubser and A. MullerKirsten. The latter is what happens in the S'wave case of the attractive potential.** 19. / i ) = Y. 180). J. Hashimoto [124] and R. therefore we cite it from Meixner and Schafke [193]. M^l)(0.( i J e ^ ) " 1 ! 2 = R2 + ^ . H.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . (105) to (108).28)) and R = ey real. i.88) where the factor Me„(0. i.19. (p. Jv+2r.h) The function Mo (0.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. 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. The expansion is: 4 r ( / i 2 ) M S ( z .±)2 + (2sinj/7r) 2 . y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. " S e e S. 178). W. in which v is complex. unitarity is preserved. and then exploit it for the evaluation of the quantity R..Q.89) . h) serves the use of the same coefficients as in the other expansions. A derivation is beyond the scope of our objectives here. J.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz).e.84) of the ^matrix.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. (19.2.79).m a t r i x Our next task is to evaluate the expression (19. (19. (19. By inserting in Eq. (^)l4lu{h2)Jir(hez)J±u+i+r(hez).h) is (cf. i.e. Eq. S.2COS2VTT = (R.6 E v a l u a t i o n of t h e S ." \R~Ji\2 1  tO _  2 i s i n VK\2 [  . (19. but in the 10dimensional string theory context. This implies basically the evaluation of the expression R of Eq.
6 2 ) m 6 (^_i)3(z.3 9 ^ . Here in the case of the repulsive potential.94) . i.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.49) from which we obtain for z = 0: Me„(0. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ .28)).g.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq./>) = Co> 2 ) £ 7 " = — OO j ± > . h2 is complex but v2 is real (cf.2_4)(l/2_9)iv2y) + <AA. we obtain M&(0.. Eq. (19. (19.456 so that in particular for z = 0 oo CHAPTER 19. Singular Potentials 4r (h )M$(0. We require the power expansion of the Bessel function of the first kind which we obtain from Tables of Functions as I)EH^(T)Thus.90) and choose r = 0 (one can choose e. r = 0 and 2.g. setting r = 0 in Eq. The formula (19. (19./i) with the help of Eq.90).49). As a matter of introduction we recall the definition of the coefficients with Eq.55). ^ (19. (19.e. one obtains *. h) in many different ways. u»^. r = 2 as a check). (19.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h). by allocating different values to r. e. /=—oo (19. (19.91) Inserting here the coefficients given in (19. /n ^ . Next we evaluate M^1}(0.
.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .1) 2u(4u + 15i/ ..9) V2 1 fh ±v 1+ 2 + • (19..2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.95) These expansions imply for the quantity R: I/! 2v ri1 i .96) With this explicit expansion we can evaluate the S'matrix.19.93) and the coefficient expansions (19.111) / / i 5 4 3 1+ 4i/ 2 + +' (z. From expansion (19.IV 2 2 2(v =F 3i/ . one obtains M£!(O.1 ) 2 ( I / + 1 ) ( V 2 .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).7) F " (i/±l)2(^^l)(i/24)V2.4)(i/ . In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential. We observe that this approximation involves v and h4 and hence is real.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ . . _2_(h\2 .2 _ ^ 4 ( ^ 2 _ 4)2 (19.2^ 2 ± 59^ .l) 5 (19.12i/2 + 64u ./I) h u . 4(z/ T H ^ 3 .97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy.] 2 \v i + {v2 .32i/ . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu . + # "+" ( I / .
7 CHAPTER 19. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. Then cos 2TT/? = 1 .101) is > zero. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).^ i 4 2+^ # 1 / •.2. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. (e iun e~iv*^/R*2)' (19.28) that v is real. (19. 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. Eq.R*2 ^ * #2 \ 11 tf2 R* (19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.102) (19.28)). (19.103) Then SiSt (1+5) i . In this case R — R* and so 1/i?2 ~ Ofji4).84): Si ST 1 (11/R2)(11/R*2) .eiun/R2)(eiu*n ~ (19.104) We now consider the case of a — 0 implying that g of Eq.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/? }]"' (19.98) (cf.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. (19.458 19.101) where / is complex and g is real. and sin 2TT/3 = ft/. + ( i + s w (E 4 . (19. (19.9 / . 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. This is the case of real .
107b) = Ah2[l + 0(h4)}.Q. (19. In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the threedimensional case we are considering here. (19. MullerKirsten. ( I l l ) ) in agreement with a result in S.3b)) h2 is real for E = k2 > 0. W.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].28).e.97) this can be written A 4TT2 [v\{v . H. Hashimoto [124].102) we obtain Sis. Magnus and F.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . S. 1 29/ ^ (1 .577. Here in t h e case of the attractive potential (as remarked after Eq. p. 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 (19.61n2 .2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .1)\} 2 h \2 iv 1 + Au {y . (19.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.^ ) ~ 2 2 459 1+ 9 / • & ( ! .SiSf » 4 sin7rz/ R (19. We can also evaluate now the amplitudes (19. Hence with Eq. f W . i.61n/i} + ~9~ 0(h8) (19. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). = 1 .l ) 2 2 \) +0{h«) (19.6C . Oberhettinger [181]. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .107a) In this result* we now have to insert the expansion of the parameter v given by Eq. J. J. Manvelyan. At = 2i[l + 0(h2)}. "This is the result effectively contained in R.3. Liang and Yunbo Zhang [189] (there Eq. Gubser and A.106) and evidently violates the unitarity of S. (19. With t h e help of Eq.19.107c) Here h2 is actually Vh4.
where the part of interest here is based on H. 19. H. H. W.4 Fig.1.108) For the large values of h2 that we wish to consider here.26) and hence set a1 = 2hl + 2hq + A{q. completely different solutions. Aly.460 CHAPTER 19.3. J. we again make use of the replacement (17.h) 10. MiillerKirsten and Jianzu Zhang [226]. 2. This is therefore a highly interesting case which found its complete solution only recently. S.e. Singular Potentials 19.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . Wannier [278]. H. (19.H.$ q(l.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. Tamaryan. Park. . 2 2 a = [l + 2 (19. h) for large h and upwards I = 0.a }tp = 0. VahediFaridi [8] and G.3. i.3 19.3 The function q(l.109) ^We follow here mainly D. Miiller and N. 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. N. K. We are again concerned with the associated Mathieu equation given by Eq.h) (19. W. J.
yJ\ — 1 and cos TTV . The behaviour of q as a function of h for some values of / is shown in Fig.21b). Therefore the second term on the right is nonzero.e. (17.l + 2y+(l. With the boundary conditions y+(0) = l.3. (17. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v. we defined the Floquet exponent v and observed with Eq.110) a where y+(7r) is the solution evaluated at z = TV. See also Eq. which is even about z = 0. i. (19.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. J4(0)=0. We begin.y} where VF[y+. which therefore cancels out or can be taken to be 1. (19. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.19.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. y(0) = 0.63). . and a is its normalization constant. with an essential mathematical step. 19. The replacement (19.h*y_(l. (19.5) 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.3. however. 19.§ a W[y+.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. y± oc A{z)e2h sm * ± A(z)e2n sm z . We also found the boundary condition (17.2 T h e F l o q u e t e x p o n e n t for large h2 In our treatment of the periodic Mathieu equation in Chapter 17.113) 'We are considering nonintegral values of v. y'_(0) = 1. to which we have to return in this subsection.h*y (19.21b).111) we have W[y+.e. i. (17.
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 . W. Dingle and H. which is \/2 at z = 0.A(z)e2h sin 2 sin z sin z (19. See R. This paper contains the normalization constants of large/i 2 periodic Mathieu functions. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h.32). and hence with Eqs. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . (17.118a) "We emphasize again: For integral values of v the normalization constants are different. (17. Miiller [73]. (19.39 2 14 /i 2 + (19.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. 1 1 Or see R. B. Hence we require (obtained from Eq. Dingle and H. (19. (19. B.e. i.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). J. 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. K [ (z)}2hsinz (19.114) a a Our first step is therefore the determination of the normalization constants No.43) we see that z = 7r/2 implies W(TT/2) = 0. (17.117) a a Referring back to Eq.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. W. We now obtain the expressions needed in Eq. Miiller [73]. J.462 CHAPTER 19.112) from evaluation of Eqs. in the second expression 2h is obtained in addition from differentiation of the exponential factor). (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 .119a) This formula — derived and rediscovered in Park.h . (17.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. 210.19.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 .118b) The factors a. a in Eq.62a) and (17. Tamaryan. .i ( g +3)]! 1 24/i {2hl>2). (19.87 2 14 /i 2 [ i ( g .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .2g + 3) 27/i ( 3 g 2 .119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v. (19.1Q11QM cos irv + 1 ~ •==— e .! ) ] ! [ .117) are known from the matching of solutions in Eqs. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. (19.3).62b) (see also the explicit calculation in Example 17. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.! ) ] ! [ . p. (19.
3.z) i/>(q. Then straightforward integration yields > inq 4 „. —h. each with a specific asymptotic behaviour. (19.2 we insert the expression (19. z) = ip(—q.z) = Aq(z)exp[±2hiswh z] ^~°° exp ±z/te ( ')eT^/4> V cosh z e x = Aq(z) exp[±2hi sinh z] ** ~~°° ^ h e ^ V cosh z e ^q/\ (19.h.109) remaining unchanged. h. . we can obtain the linearly independent one either as ip(—q. h.h. for h — oo).z) = A(q. h. Vcosh z \ 1 . With the solutions as they stand. (19.123) We again make the important observation by looking at Eqs. z). of course. —z).h. . h. h.h.124) Since this function differs from a solution ip by a factor k(q.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. the expression (19.3 Following the procedure of Sec.z).h).h) =?==.122) Correspondingly the various solutions ip(q. z) are TP(q.3. We define these solutions in terms of the function Ke(q. it is still a solution but not with the symmetry property ip(q. 17.464 CHAPTER 19. —h.z).122) that given one solution ip(q. In the following we require solutions H e ^ ( z ) .4.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. dA 1. z) can then be written as . i = 1. —z). Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.z)exp[±2hismh z\. tp(q.h.120) The resulting equation for the function A(q. h. z) = ip(q.120) to (19. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ .2.108) and set ip{q. h(q. h.? s m h z) Vcosh z (19.h)iP(q. V— 2in (19. .z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. „ 1 A „ —A cosh z——H ..e.109) into the equation (19.3.z) or as tp(q.h.
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 .3 The Potential 1/r4 — Case of Large h2 465 Instead. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr.19. (44) of W a n n i e r .e. in performing this cycle of replacements the function picks up a factor.fi2) can be reexpressed in terms of Hankel functions./h). Me„(z. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. * ** = > rojKe( ". 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 ./i 2 ) M (19. (z. (0.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . . 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 . as we saw earlier. we h a v e 77 = 0—> 6 = 0. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.h2).h2) = OvMPfrh2). i.126) For large values of the argument 2/icosh 2.WJ " C o n v e r t i n g t o n o t a t i o n of W a n n i e r [278]. i. the Bessel functions contained in the expansion of the associated Mathieu function M{. This is.<5±1) (19 125)  (the expression on the far right in leading order for h2 large).e. It follows t h a t S e t t i n g s i n h 77 = t a n 0 .e. achieved with the help of the Floquet solutions Me±v(z.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 .127) sin z T /'1 M TT\ — <>{ 4 .45). (19. 2j<. >• * F $ 0 * e". av{h2) = ^(0J ^ Mi i .^ s i n h 7?)]g. [ \ . In order to be able to obtain the Smatrix. (19. . 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]. We observed that these satisfy the relation (19. : 7r/2.
(19. (19.q. (19.466 CHAPTER 19.q.130a).h) Be^(z.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs. 3te»0.h. HeW(z. (2) (19.h). (19. (19. h. exp[—ikr — in/4\ kr Re{2)(z.h) (3) = = = = Ke(q.130a) Re^(z.q. Singular Potentials Note that the sine part is nonleading! Retaining only the dominant term of this expansion we have (with z — 2h cosh z) > Me±l/(z.g.q./i) Be^(z.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.q.g.q.vr .h) He (z.130b) r_0 rll2exp\g/r (ig) ' 19.z).129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H.h) = e(z) exp ihez . 3te»0.q. In this way we obtain in the domain .q./i).4 T h e c o n n e c t i o n formulas We now return to Eq. z): Re(2\z.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.eW(z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .h).3.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.130b). He (z. KeW(z.
h) is asymptotically small (cf.q./i). Changing the sign of z./i) — sin 7T7 Re^(z.li 2 ) Oil. and thus can be matched in this region. t With Eq. q.130a).h2) a. we obtain in the domain tfcz <C 0 the relations: Me./i 2 ) = Me_j. g . q. (19. (19.g.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . (19.g.134) in the region where the function He^(z.134) Considering now the first of relations (19. these are exponentially increasing there.132) into Eqs.130b)).h2) Me„(z. h) — sin 7T7 He^(z.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19.q.g.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19.(z.q. h) = Ke(q. h. (19. h) = Ke(q. we see that the solutions (compare with Eqs. h) = sin 7r(7 + v) He (1) (z. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.(z. h.129)) He (2) (z. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .133) In a similar way one obtains the connection formulas — sin nv HeW(z.135) sin irv 'This means. z) are proportional there./i).i ^ / 2 H e ( l ) (z> ?) (19. and He(*> exponentially decreasing. / i ) + sin7r(7^)He(4)(2. (19. Eqs.132) These relations are now valid over the entire range of z. (19.131) where the second relation was obtained by changing the sign of v in the first.19. (19. — sin 777 H e ( 3 ) ( 2 .3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. Substituting the Eqs.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. (19. .h) sin irv = sin7r(7 + ^)He (3) (2. z) and He (3) (z.h). q. q.
133)) — sin TXV He^4' (z.5 D e r i v a t i o n of t h e ^ . (19. is only the dominant term for large h2} 19. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . and therefore remarks after his Eq.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. The latter is defined by the following larger behaviour of the solution chosen at r = 0.119a) and (19. sin TXJ.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19. (19.1 .85b) obtained earlier.135) is the equation corresponding to Eq. Singular Potentials The factor on the left. which in our case is the solution He^4^. Wannier did not have the relation (19. (19.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . i. q.119a) or (19. Wannier [278].( . Thus here the S'matrix is defined by (using Eq.(19.3.119b).Sin TXV.139) ''Equation (19. (19. . H. Squaring Eq.135).133) we can now deduce the 5matrix Si — e2t51. cos TXV + sin TXV.e'i7ri 1. of course. The quantity 7 is now to be determined from Eq.135).468 CHAPTER 19. (19.e.l ) (l„—ikr e From this we can deduce that Si (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV .119b) as „4/i 1 + COS TXV ~ rrl ( g . where 8i is the phase shift. (19.m a t r i x From Eq.137) We see that this expression agrees with that of Eq.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. (60) in the work of G. (76): "It is not likely at this stage that an analytic relation will ever be found connecting (what we call) v and 7 to (what we call) (I + 1/2) 2 and h2".
e. the real part of v must be an integer.e.141) COS TTU irvj — i s i n WR s i n h TTUJ.19. . (19. as a comparison with Eq.137) together with (19. Si xe ~ te ilir COS KV V cos2 irv — 1 — e (COS^Q COS ITU ig7r 1 + e'iqn ielln cos TXV 2 2 C O S TTU (lp/2)n (19.140) From this we obtain the absorptivity A(l. 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^.138) we obtain. A{l.4 The absorptivity of the S wave (attractive case). h) of the Zth partial wave. 19.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.8h COS From Eq. (19.139) could actually be neglected) implying  cos m>\ = cosh nvi. i.e. imaginary part uj (so that 1 on the left hand side of Eq. (19.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. This is different from the behaviour for small values of \h2\.109). T h i s is r e a l for UR a n integer. for the attractive potential. (19. 9/2 (91) with the approximation q ~ h obtainable from Eq.28) shows. we can approximate the equation for q ~ h (i. since cos TVU is large.§ 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. i.
"G. such as phase shifts. in further contexts beyond those already mentioned. 11 J. Limic [177]. Paliov and S.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. J. D. L. A. C. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi ." as well as other aspects. Lii. 19. F. H. Apparently this is one of the very rare cases which permits such complete treatment. Esposito [88]. E. The diminishing fluctuations of A(l. (19.k) in the approach to unity are too small to become evident here. 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. N. Jones [78]. Cvetic. A. I r a n [61] and M.** 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. Treiman [271]. D. N. small h2) case have been considered by some other authors. This is sketched schematically in Fig. and G. 19. Handelsman. Dombey and R.107b) and (19. in fact. Partial or other aspects of the weak coupling (i. **A.e. H. YuanBen [285]. Bertocchi.P. Cvetic. Shabad [248]. Pao and J. Lew [128]. Furlan [20]. Fubini and G. 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. Y. **R. A. W Some of these sources can be helpful in further investigations. S . Challifour and R. Rosendorf [225]. Tiktopoulos and S. VazquezPoritz [61].470 CHAPTER 19.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. (19. Pope and T. Apart from the papers already referred to. . N. Eden [46]. H. Masson [191].4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r .4 (there are tiny fluctuations in the rapid approach to 1).142) we can see the behaviour of the S'wave absorptivity as a function of h. S.4 .139) 2Tr(16h)q /32\ f e Thus with Eqs. B. and. Lii and J. In essence. one may expect corresponding results for other singular power potentials. see also M.
e. *R. M.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. J. Dingle and H. B. We concentrate in this chapter on the large order behaviour of asymptotic *C. W. also convergent expansions are known and a lot of literature exists on the equation.* In fact. Dingle [71]. Muller [74]. This is indeed a remarkable connection. 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. Wu [18]. B. [19]. T.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. We shall also see explicitly. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut.* However. f R . 471 . 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. 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. Bender and T. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. the asymptotic solutions in different domains all have the same coefficients.g.
This approach leads . 5. Fig.2 The function ur = r\/pr vs. r for p = 4. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.472 CHAPTER 20.7. as evident from Example 17. 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. 19. 20.8. 20.2. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.6.2.1 The function ur = pr jr\ vs. Since the method is also applicable to convergent expansions.2 and below. 8. 8. 10 12 14 16 18 20 Fig. Presumably this has never been done so far.1 and cases in Sec. r for p = 4.6.
J2 2n h E (20. 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. for instance. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). even if each En is known. 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.3) n=0 Thus for any given value of g . (20. in the power expansion of the exponential function). It can be seen from these. as we discussed earlier in Chapter 8. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. For an authoritative discussion of these aspects we refer to Dingle [70].4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. i.2) or. 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.e.1 and 20. 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. and the expansion is said to be Borel summable. The behaviour of such terms is illustrated in Figs. and the integral E(g2) is called the Borel sum. 20. equivalently for g = l/h. i. the function E(g2) is given only approximately by Yln=o En92n. . The formal perturbation expansion of a quantity E(g ) . In some cases the answer is affirmative.2. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term.e.1 Introductory Remarks 473 automatically to asymptotic series. T V lim h2N 2 n EQ?) .20. this term represents the size of the error.
. In any case. (20. (20.6)). . that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). (20. In fact.4).5) then with the integral representation of the gamma function F(z) = (z — 1)!. it is important to know the large order behaviour of the asymptotic series so that the question of its (exact or only approximate) Borel summability can be decided. (and E0 = 0). We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. . El . (20. 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. Q. Thus if CO F(z) = aiS(z) + J2 anzn.e.474 CHAPTER 20.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. . In this ideal case the integral of Eq. Large Order Behaviour of Perturbation Expansions Eq. .6) The power series of F(z) converges inside the unit circle about the origin at z = 0. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . n=0 (20.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.1)? Asymptotic expansions originate from integrating a series (i.n _ En+\ n! En+1 ^ (!)"«!. In general this is not so easy. . J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . n=0 (20. 1 we obtain oo F(z)£(*)" = — .
3 and g < 0. for < ^ 0. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. $s E can be calculated by the WKB procedure.. Weinberg [228].. 20..11) V. if the potential is V(r) = +grN. § where 7(ff) = To exp  .. (20. N = l. (20. 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. . Popov and V.e.2. (20.3 The potential V(r) gr».9) The question is therefore: How can we determine 9 E(h). This is the case if the real potential has a hump of some kind. and E represents an eigenenergy.2. ? When ?s E ^ 0. or can we determine Ar independently? Fig..1 Introductory Remarks 475 except for possible subtractions./ \p\dr\. e... and with g = 1/h and it is found that§ En = 4 n ) ~ \lis)..g.10) is considered in detail in this work.20. we have a quantum mechanical problem where probability can leak away. The potential (20.N = l. M. 20..10) r as shown in Fig. i. S.
Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. Popov and V. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. (20. as we expect on the basis of Eq. Solution: The solution can be found in a paper of Eletsky. An interesting application is provided by Example 20. However. Thus in the following we apply these methods to our typical examples. Weinberg [83]. S. 8.11). the eigenvalue expansions of the cosine potential and of anharmonic oscillators. of course. (11. (11.1. V.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. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. 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. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available.134)). L. 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 .476 CHAPTER 20. .113)). Popov and Weinberg. Example 20. Here n is the principal quantum number of the Coulomb problem (see Eq. the BreitWigner formula (10.7 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. as was (and still is) not the case for t h e equation 1 V . the perturbation method of Sec.^ The logarithmic factor in Er is a novel feature of this case. 20. Eletsky. the cosine potential being t h e most completely investigated will also be seen in the following.9) for Ar. (14.76) (see also the comment on the comparison with the Kepler problem before Eq. (20. M.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. Equation (20.11) is.73) in the W K B or semiclassical approximation.
This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods.j) are known. J. Dingle and H.58) and (17. 3)? . (17. (cf.j) of functions ipq+4j. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. (17.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. i.60).l(q. Eqs. 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.13) In the special case j = i the boundary conditions stated after Eq.l ) ( g + 4 j .3 ) P i _ 1 ( g .61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.20.26) and (17. j .e.M2i+1 for large i. (17.Bq.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.j + 1). (17.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.h) = 2h2 + 2hq+j.J) = (g + 4 j . W. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.Cq.14) Thus these coefficients can be obtained once the coefficients Pi(q. % = Aq. are given by i . (20. Eqs.j) a n d M2i can be deduced.l ) +2[(q + Aj)2 + l + A]Pl.55)) X = E(q. obey the recurrence relation jPi(q. M2i = 2j2 J[Pi(q. Miiller [74]. Eqs. (20.2 A = Mx + ~M2 + 7^WM3 + •• • .12) (20. (17. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const. . or equivalently of the quantity A.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g . h). B. The recurrence relation (20.13) represents effectively a partial difference equation.34)) that the coefficients Pi(q.
l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula. see also below) .18. also written lCj.91 l)]![i(g3)]!} 2 { [ . Using this relation and the duplication formula lw('\W±^.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.16a) (20.478 CHAPTER 20.•91 {[l(? + i)]![(? + 3)]!}2 (20. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.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. Hence exact Borel summation is not possible.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20. . In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions. (20.l (20.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.*. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.
This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. we emphasize a few points *M. At the risk of repeating what we explained earlier. W. n = 0 . we follow here a formulation in the context of our earlier considerations of the Mathieu equation. in the calculation of the level splitting as a result of tunneling. MiillerKirsten [200]. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary. but with different boundary conditions. when this is g0 = 2n + l . i. f H . . Stone and J. 20. . . J.* However. Previously. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. l .3.5 \ \ / \A Fig. The potential we consider here then has the shape shown in Fig. 2 . remains unchanged.3 Cosine Potential: Complex Eigenvalues 479 20.4 The cosine potential shifted by TT/2.e. A means to achieve this was found by Stone and Reeve.4.2 solutions obtained earlier.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. Reeve [262]. .3 20. \ 1 \ \ \ \ \ 0. the difference q — q$. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. 20. We retain the previous boundary condition at one minimum. this difference was found to be real.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/?. .20. 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.
20. (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. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear).4. but achieves exactly what we want. and allows its imaginary part to tend to infinity.4. Previously this situation was the same at the neighbouring minimum. The physical wave functions there are therefore the solutions which are real and exponentially decreasing.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. a boundary condition which at the face looks somewhat abstract. 20. Setting w = 2h1/2z.20a) As stated.4 (around z — 0) to the one on the right (around z ~ ir). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . but t h a t with complex argument there. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. 8 the equation can be approximated by d2ip(w) (20. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . Effectively we consider the tunneling from the left well in Fig. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. (20. Thus one chooses not the solution with real argument at the minimum. . X + 2h2 = 2hq+—. we shift the argument by n/2.2h2 cos 2z]ip{z) = 0. as we found these for this potential. i. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them.480 CHAPTER 20.21) <u> + ( i « . which means replacing z by z±ir/2.Ah2z2 • • • }^{z) = 0. Here we choose t h e second way.e. 20. We can distort the potential there or alternatively impose a different behaviour there on the wave function.T K ^ a D_i_{q+1){±iw). Now we change this situation at the minimum on the right.
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. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. . 20. Thus. that at z = 7r. 17. Since the large/i2 solutions valid away from the minima were the solutions of types A. (20.22). 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. whereas here we start off with a solution even or odd about a minimum. by a different method.e. 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. we see that ips ~ cos \j2qhz for z ~ 0.23) . 2. A derivation is given subsequently. 3 . we now choose an harmonicoscillator type of real solution around the left minimum of Fig. such that this deviation becomes imaginary. whereas here we construct these from oscillatortype solutions. A (which we can loosely dub solutions of WKBtype). the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}.63). Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. (20. we naturally constructed the even and odd solutions with these (cf. . . (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. 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.e.24) Considering t m O i n Eq.20.4.3 Cosine Potential: Complex Eigenvalues 481 As explained above.> 0 for z > ±ioo. Eq. . so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. however. We consider here explicitly only the case of the ground state around z = 0. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. .
2.27) requires the differentiation of this expression with respect to the index v. Magnus and F. Oberhettinger [181]. However. Magnus and F. The arguments of the functions appearing in Eq.27) differ by an angle IT. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. §See e. p. From Tables of Special Functions'. p. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. (20.3. Setting v = 0 in T(—v) = (—v— 1)! gives infinity.30) 'See e. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.•w) + 2ir T{u) <"+!)% Dvi(iw).482 CHAPTER 20. We see this as follows. this can be handled in a few lines. W. 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.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/.g. .(. Thus the only nonvanishing contribution comes from differentiation of Y[—v).g.25) qo) E2 0: (20. we know from the solutions of types B and C in Sec. and then evaluation at u = 0.28) Extracting Du{—w) we have A. tp(v).h(2h1/2 J £2=0 (20. 92. (20. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. W.24) are linearly independent around z = 0 for q ^ qo.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. (20. (20. 17. Oberhettinger [181].w) = e T Dv{w) 2TT n») AvViDv^iw).29) Equation (20.
e. We now go to the WKBtype solutions.e. Magnus and F. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. those of type A.32) which we can use here since all other problems have been resolved: d D {w) dv v (20.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. dropping from the derivative part of Eq. which are valid around z ~ 7r/2. we obtain 4>s D0(2h^z) D0(2h1/2.31) dT(u) r*(v) dv 1 i/>(v) ^ o l. Magnus and F. W.g.//n and ij>{y) = C . we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. This result is » not easily generalized to <jo > 1. .+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. Thus calculations for qo ^ 1 are not given by M. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. Oberhettinger [181]. Differentiation of this equation and subsequently setting v — 0 yields. For u — 0 one obtains the result (20. including the derivative of DE2ufl(2h1/2z)). (20. 3 .33) = IV2~TT.29). r(i/) We now return to Eq. (20.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity.)e" w /4 2 l ^Dx(iw) (20. Oberhettinger [181].3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.4. (20. 92. p. i/^0 483 T(v) It follows that d dv [T(u)\ (20. i. pp. .. Reeve [262]. .20. and "Prom W.31). 5 7 7 . Stone and J.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .27) exponentially decreasing contributions. "See e. .
i.484 CHAPTER 20. (20. (20. x . (20. smf (20.40) From Magnus and Oberhettinger [181] (p. We observe that in the matching domain (i.37) Comparing Eqs. by there replacing z by z + ir/2.36) cos  . i.36) V'WKB cie~2hehz' 9h  .cos z — 2/icos z ipA^ Tcos  and i n A —. 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. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left. near z = n the boundary condition (20. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . 2 c2e2hehz UJ2 .e. we obtain the required solutions from the earlier ones.29) and (17. (17.— 721 .38) in its limiting domain towards 2 = 0.34) and (20. y COS c\ J (20.41) 2TT . that in approaching z = ir.34). where we can then match these to the solution ips. (20. We now return to the WKBlike solution (20. Recalling that we have shifted the potential by 7r/2. cos z/2 ~ (20. (20.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h.sm I We now want to match this solution to that of Eq.e.e. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.40).35) We construct with constants c\. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). Then V'WKB becomes (with cos z ~ — 1 + z' /2.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. Eqs. To this end we set z' = z — ix.. Thus in the present case and with q ~ qo = 1 we obtain „2ft.39) Now.36) and consider this at the other end of its domain of validity.
1 2 4 1 2 8h ^^2 2 (20. . (20.20. J Hence Eq. i. (20.46).34) we obtain E2 = i%\ = ±ih22Qe~8h.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. with c\ = exp(—2h) from above. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. (20.^.44) = DQ{2h / z) ± (2h7r) / e.2/.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.43) we obtain. We derive the general result in the next subsection using our own method.1/V)_le^_ Z With this result Eq.43) Comparing now Eqs. The general result stated by Stone and Reeve [262] is therefore presumably guessed.39) and (20.45) Comparing this result with tps of Eq. 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 .40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2.38) can be written TAWKB (20.46) Repeating these calculations for the general case (i. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . (20. (20.e. ^'mliw <2a47) For qo = 1 we recover the result (20.e. (20.
486 20.2 CHAPTER 20.z.48) These conditions provide the required quantum number qo = 2n + 1. (20. 20. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .5. and expanding cos 2z around the point z = — TT/2. Achuthan. i. i.46) we consider a half period of the cosine potential as illustrated in Fig.5 The cosine potential from — TT/2 to n/2. •+(=)* V(f) 0.3. Considering the original equation.e. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. J. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . W. 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.e. MiillerKirsten and A.20a). .E curre it \ j. 20. Wiedemann [4]. At the minimum to the right of the barrier shown in Fig. We have to impose two sets of boundary conditions. Eq.Ah2 ( z + . (20.e.()#0.*. the result is formula (334) there. H. V(z) . Fig. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. •_(:)*<>.5. at z = TT/2.) + 7T V> = o. i. 20. (20. On the other side of the barrier at z = —TT/2 we consider different boundary conditions.
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.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.20. Proceeding as in our previous cases and substituting the expression (20.r/2 ~ e i(2hq) V 2 . three pairs of solutions of types named A. 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. (20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.21)). we have IPA(Z) „2hsinz Aq{z) + O il>A(q. (20. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.21) into the Mathieu equation we obtain.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .e.51b) 2i(«1)[i(9l)]'.z)=tpA(q.h. For the continuation we use again the WKB formalism.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. as shown earlier. Recapitulating these from Chapter 17.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}.z). we demand that ^(*)L<.h. (20.50a) 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. (20. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. i. B and C.
We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.=F2 2\1/2{28h2y/4e~Ah l + O (20. and the second around z = 7r/2. i. \vw). Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function.^ i .52a) = Cq[w(z)}.55) Extending these solutions to the domain around the minimum at z = 7r/2. the relation q .57) where go = 1) 3.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20.qo . (20. we obtain 1>±W = \ a a (20.e. 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. Here the upper sign refers to the even states and the lower to the odd states.488 CHAPTER 20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) .48) and proceeding as in Chapter 17.56) Applying the boundary conditions (20. 5.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. ^ W H . we obtain our first transcendental equation. (20.52b) The first solution is valid asymptotically around z = —vr/2. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2. We return to this condition after derivation of the . (20.
49) and we can apply our boundary condition. . e x p ( . Our procedure now is to match these to exponential WKB solutions.58) Ignoring the correction A / 8 . formula 280.67). C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °.61) [Ah2 cos2 z2hq]1/2dz. Some nontrivial manipulations are again required to uncover the generation of important factorials. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.Ah2 cos2 z ~ 0. Byrd and M. 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. as we shall see with the result of Eq. These then provide complex exponentials like (20. p.5 9 ) The WKB solutions to the right of z+ (where V > $IE to the left of the maximum of the barrier) are given by ttw . r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .e.2hq] / dz) We know that the solutions I/JA{Z). i. (20. Jz+ ' T In principle the integral could be evaluated exactly — see P.[ [Ah cos z . 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. we see that the turning points at z± are given by 2hq .20.01. and we obtain the proportionality by appropriate expansion. Thereafter we extend the latter across the turning point down to periodic WKB solutions. 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. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . F. (20. Friedman [40].3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. D.Ah2 cos2 z ip = 0.
sin2 z = l . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).59) 2h COS Z+  ~ 0. we write cos2 z = l .» _ i y v .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.) +0 z+ IT Inserting this into the integral I. 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 /•«"(. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20. We have for the following expression contained in (20.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.cos2 ( z+^) = (z+ .62) as follows.490 CHAPTER 20.64) .63) We handle the logarithmic terms in Eq. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. (20. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. (20.
63) and (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . <$&?(*) 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 . 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.50a) and (20. 1/2 (20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .+ qj V2 4 A.64) into Eq. . We ignore this here. (20.20.66) The WKB solution therefore becomes (approximating the denominator of (20.60) by (4h2 cos2 z)1!4)./ * [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\"  . . .67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.62).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. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan . (4^2)i/4 [ i( g _3)] ! V/ l (20.3 Cosine Potential: Complex Eigenvalues Inserting the relations (20.50b). on using in the second step Eqs. (20.
(14.50)).69) l(8M9/4e2^ \2 [ i ( g . Eqs.+) + B^) HAh^/\ir.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .?/ 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.l^l !T ^^* T2 7 r ( ^^ /.49).1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .67) and (20. We now return to the even and odd wave functions defined by Eq. (20.i ( 2 / i g ) 1 / 2 ( z + .49).. 1 2 1 ±2 .™) . Thus with IPA(Z) and TPA(Z) taken from Eqs.492 CHAPTER 20.T 7 77 r )V2 /4(2 .U ^K^. we obtain T _ ( ± ) = 0.z)}]. S1H (2/ KZ ) / (z f 2 (2^). (14. (20.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). Large Order Behaviour of Perturbation Expansions again using the Stirling formula.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20. io(8/*)g/V^i6ft2)V4[±(g3)]! .2 [ ? ) Imposing the boundary condition (20. i. we obtain l(8/i)g/ 4 e. i 27r(8^/ 4 2/t /4 ' r(±) .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. where _ 1 + [±) *) + £ (20.55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+.KW4^8 4 .
72) We observe that with the square root as on the right hand side both sides are complex.26). Eqs.57) and (20. we obtain E(q.3.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . (20.57) by the left hand side of Eq. Ex = 2h2 + 2hq0 + A 1 (<? .h) + i l + O '4{[^ol)]!}2 E(q0.3 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.qo) El Ah' (20. (20. (20. i. (20.72).21) expanded about q = qo.72) is our second transcendental equation. who derived it for qo = 1 and guessed the form for general values of qo. 20. In both cases we wrote (cf.h)+2h(qq0) (16h)i°+le'Sh E(qQ.74) This result agrees with that of Stone and Reeve [262]. for q an integer both sides are imaginary. (17.h) = = E(q0. (20.h) + (qq0)( BE — 90 E{qQ. We now have the two equations which have to be satisfied. (20.25)) E = A = Ei + E2.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .e.75) .73) Inserting this expression into A = E(q. Equation (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.47).20. h) of Eq.h)+i%Ego. the result (20. Eqs. (20. Combining these equations by replacing (28/i2)<?/4 in Eq.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. in fact.3 ) I = }(»« !7rl/22I(.
the discontinuity across its cut from zero to infinity. Bogomol'nyi and V.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. we can recalculate with the help of Eq.W [£(901)]! < (20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj.9) the coefficients Ai of its perturbation expansion.74b). its tunneling properties and the large order behaviour.3. Achuthan.I g  ± l ( 2 „.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. 7 8 ) i=l V . We had A= or _2ft2 + 2ft. 20. B. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. (20. (17. A = . f E . Setting ng AEqo:=2iSfEqo. by others. (20.e.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. H. i. J. The same formula can be derived in the case of the double well potential. . we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.494 CHAPTER 20. A. Wiedemann [4] and Example 20. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level.+  . 25Eqo=8h[ 2\1/2 .) {I6h)™/2e4h 1 + . W.2 below. Fateyev [26]. Eq.80) *P. cf. MiillerKirsten and A.
(20.1).. MullerKirsten [200]. the integration is seen to give (20. (20. The integral is recognized to be of the type of the integral representation of the gamma function.l ) ( < ? + 4 j .14).13) out again: jPiiQj) = (g + 4 j . 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. W. there is nothing special about this case. But as Dingle remarks. B. Dingle [74]. 20. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. We also observe that these terms do not alternate in sign. a = J + \<1> (2084) . i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. j . (20.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . For convenience we write Eq.3 ) P i _ 1 ( c ? .j) + {q + 4j + l)(q + 4j + 3)Pi1(q.19) for the large values of i under consideration here.20.j) satisfy the recurrence relation (20. Using again n = (go . so that an exact Borel summation is not possible. *R.l ) +2[(q + 4j)2 + l + A}Pi_1(q. J. f H .81) We observe that this result agrees with the result of Eq.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. (20.74).j + 1). (20. J).83) One now sets ftH = ( 2 % P * ( 9 .
4 ) ^ .h) = ^2pi(a) with p. (20. Whittaker functions WKjfl.K + ^ ] L + K^)WK. the recurrence relation (20. d2WKh dz2 + 1 4 K z zM2n WKtll = 0.^(Z).27h pi(a) a = 4j(4j + 2 g .e. (20.90) See M.85) Considering now the generating function of the perturbation coefficients. g < 4 j . A. i. (20. formula 13. nx pi_i(a + l).1/2(z) (a1)! (20.(a) oc .91) + Wa+lil/2(z) = (1 .1) + 2 P i _i(a) + Pi_i(a + 1). i.e.a R . Taking dominant terms. (20. 4j(4j + 2g + 4) + / . Stegun [1]. (20.89) Replacing here K by a and \x by 1/2.83) can be written j .85) implies p(a .h) 2+ 8h a p(a.ll{z)= L . which are solutions of the differential equation known as Whittaker's equation.h) +p(a + l. z2 (20.87) Taking terms of 0(h~l+1) of this equation.85).1) + 2 (a — 1) a .l/2l Wa. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.e. Now. . (20.4. e.h). a.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.g.496 CHAPTER 20.86) one can convince oneself that Eq. we obtain (2a . 0 4j(4j + 2g) pi_i(a . (q + 4j) 2 ~ 4j(4j + 2g).z)WaA/2(z) Setting V. we regain Eq. p(a.31. Abramowitz and I. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . Pi(a) i. 1  1 / 2 (4 (20.l.
> (20.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. Magnus and F .h) ~/(fe) ' (a1)! . > > ^See W.94) Hence to insure that p(a.4 Cosine Potential: Another Method of Calculation we obtain Vai. h —> —h.\)V .i/2(z) + Va+lA/2(z) = 497 2+ Va.(« 1) 2 ].88) is known.j — —j. 89.92) with Eq.20. ^Observe that if one solution of Whittaker's equation (20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). h) is expressed in descending powers of z = 8/i. h) is expressed in descending powers of h.86) have to be chosen such that p(a.(« . a (20. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged.h) = \~\Pi(&) with pi oc h l.87). In the present case this means the replacements q — —q. i. we have to remove a factor (recalling a = j + (7/2 of Eq.i/2(z). v .93) (a1)! ' where f(h). (20.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. Dingle discovered the solutions^ p(a. (20.g(h) are functions of h which in view of Eq. . Oberhettinger [181].84)) {8h)q/2( Ah for j < 0. such that p(a. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f .e.(« . p. (20. g(h)(20. (20..92) Comparing Eq.
1.96) and a similar expression for j < 0 with /i — —/i.j _ g / 2  1 / 2 ( . Dingle and H.83) and is given by 1 1 .94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1.95a) Inserting (20.1/2(8^) ( . J.l ) ] ! 11 R.+ i f or 3 < 0 (20. we obtain the result of Dingle: (20.95b) Then for j > 0 we obtain from Eq. 2. (20..94) here we have a series in descending powers of h. W.498 CHAPTER 20.8 / t ) for j > 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 s. (20. Comparing > > > this result now with Eq. h\ oc (8h)q/2e~4h ( _ . x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . .. i). Correspondingly we set for j > 0 / 1 A pla = j + q. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2. This expression can be obtained from the recurrence relation (20. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. n = 0. (20. Muller [74]. g — —q. B. j — —j. ^ ..).
20.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. B. of Eq.14). l{Q.157(4). S. (20. E* J'=I i+M 2 [2i + 9 .l ] ! {[.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20. Miiller [74].100) **R. 5.3)]!2 2 '+^.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20.97c) We assume that in a leading approximation we can cancel the factors (i + \q).{j + \<l).l ] ! (20. i.98) We evaluate the sum by approximating this to a formula given in the literature ** r \ f R\ __{r + Rl)\ J (rl)\(Rl)\' E j=0 i.e. M.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. W. note the misprint there: (r — s) ought to be (r — s)\. Ryzhik [122]. p. is M2l = 2j2j[Pi(<l.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. formula (30). (20. Thus ° 9 (i + \q) 2«(i+Igl)! p .1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . A special case of the formula can be found in I.e. Dingle and H.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. Gradshteyn and I. formula 0.97d) Then the coefficient M<n of the eigenvalue expansion. . J.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.+i?_l]!}2 (20.j) (i + lg)[2i + i ( g .l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. (20.0) = 1 as desired (although not really enforcible for large values of i).
34) that £ is an expansion in ascending powers of h 6 /c 2 . Eq..o i • . The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity... i.5 20. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. n = 0.1.e.5.]\.86).4)) £ = ^ = ^+ w and y= ^ > ( 20 . 20. We refer back to Eq. since \{q — l) + \{q — 3) = \q.102) Thus the result is M m 1 2«Fi + g .86) and set (cf.l l ! with the large i behaviour in agreement with the result (20.. Eq.l. [^ + i ( g .3 ) ] ! _ 1 Hence M or.2.e.500 CHAPTER 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..81) for q — qo = 2 n + l . the result of Bender and Wu [18]. i.e. i.. .e.i ) ] ! [ ^ +  ( g . [19].Am i .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. i\ M i large 1 . Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. i.105 ) the latter because we see from the eigenvalue expansion (18.e.101) This formula allows us to cancel all factorials [i.' 1 i • (20. (18. (18.17).
h2 _ h222K+llHr+K+ll2{l)r+1 r+ K .4) of Bender and Wu [19]. (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.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. i.108) Proceeding in this way.77) can be verified.107) we obtain (i)r r 71" JO yirZ£{y')dy'.109) or. .5. and hence inserting here the imaginary part from Eq. The result establishes the series unequivocally as an asymptotic series. in fact as one with Borel summability in view of the factor ( .110) This is the result in agreement with the large order formula (4.2 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. Once this has been obtained one can insert the imaginary part into the Borel transform and obtain the behaviour of the late terms in the eigenvalue expansion. (20. This can then be compared with the level splitting we obtained for this potential.86). 20.l ) r .106) With £= E r=0 Mr (20. the discontinuity across its cut. i.(20. We leave this to the following Example. in terms of K where go = 2ftT + 1.e. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.e. and then the relation (20.20. (18.
. *R. Propin.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. 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. Nakai. E. Cizek.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. Liang and H. . J. t The relation (20. H. R. S. The relation (20. S. J. V. *Some of the literature in this field following the work of C. M. M. J . as in later chapters. Wiedemann [4]. Graffi. A relation similar t o (20. MiillerKirsten and A. K. Cizek.Q .* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). MiillerKirsten and A. Silverstone [62]. Large Order Behaviour of Perturbation Expansions Example 20. K. Bender and T. H. of International Workshop on Perturbation Theory at Large Order (Florida.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . V. J. Achuthan. Propin and H. J. it can be used to obtain the other. MiillerKirsten [164]. Grecchi. 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.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. G. W.502 CHAPTER 20. t p . Damburg. R. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. J. (357) to (371). R. Harris. Achuthan. Grecchi.^ 20. " P . E. Silverstone [52]. [19] can be traced back from articles in the Proc. Harrell. J. Wiedemann [4]. J. Gram. W. Eqs. J. S. W. Solution: For the solution we refer to the literature. Wu [18]. J. Damburg. J. Harrell. J. Paldus. in t h a t when the quantity on one side is known. Its usefulness has also been demonstrated there. + 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. J. 1982) [139]. M.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. Paldus and H. T.
our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. Coulomb problem) we refer to existing literature. 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. 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. However. which explains also difficulties. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. Hibbs [95]. The formalism is not so useful for boundstate problems (in this case the initial wave function is. For the application of path integrals to the hydrogen atom (i. A brief introduction is also contained in R. R. Probably the most readable introduction. 'See in particular H. Kleinert [151].e. however.Chapter 21 The P a t h Integral Formalism 21.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. Schulman [245]. can be found in the book of B. P. This formalism is particularly useful for evaluating scattering or transition amplitudes. Feynman [93]. A standard reference is the book of R. Feynman [94]. in general. P. The path integral is. We illustrate the method here by rederiving the Rutherford scattering formula with this method. A further wellknown reference is the book of L. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. unknown). P. Feynman and A. S. since in such cases the probability of this system choosing a path far away may be considered to be small. ^ Our interest here focusses on quantum mechanics.* 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. However. Felsager [91]. 503 . *R. Chapters 2 and 5.
tf\x. 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). be ip{x.ti\tp).tf\ip) .3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e. U. (21.ti){x.x.504 CHAPTER 21. Obviously K(xf. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.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.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). =J dx(xf.1) Let the wave function i/> at time t = ti.ti).2) where K is a Green's function. (21.tf) = / dxK(xf. (21.e.ti)ip(x.t). the initial time.tf. i.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. we can compute the action S1.5) Sn = X / fc=l mo (zfe . */ = U + (n + l)e. (21.e—»0 Sn.x).e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. ti) = 5(xf . The Feynman Path Integral Formalism 21. ti). *In Dirac's notation we write this equation later (xf. x. we wish to know* ip(xf. 2mo dx2 (21.
21.tf). Here and in the following we need one or the other of the following integrals (cf.1 A path in (x. 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. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. the Fresnel integral (13.t).2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. as will be shown to be possible below. 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.28)): f dr\ exp ia a) ' (21. 21.6a) .1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation.ti) to its final position at (xf.21. The usefulness of the method therefore depends on the fact that some paths are more probable than others. of which one is illustrated in Fig.
8) Note t h a t in spite of the n integrations. dx exp z—{a.m0 . for example.506 CHAPTER 21. ie f me\ 2a\a~) 1/2 (21.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.xk+x oo —oo oo .t ~Xk+l> + xk+i 2 ) +{xk ..5)): G(xf.0 ^7 (x f c + i .z f c _i) 2 } } .xk+i) exp . tf > U.ti) '•= f g m ) (N(€\]n+i dxL. one has j ^ dr1eifa+i^ /£ = .ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.37)) dxk exp d(xk . J—oo dxneiS/h — • 0.xk_i) dx exp i — { 2 a .6c) In the evaluation of these integrals we replace a by a + ifi.Xi. n > 0.tf.7) Thus we expect t h a t — in view of the multiplicative y/e here in front of the phase factor — in the limit e — 0 > oo />oo /•oo / dxi oo / J—oo dx2.x f c _ i ) 5 \m0 J (21. dxneiSn/n. 2 + 2a.6b) —c ia o drjr] e x p—V € (21. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.s f c _i) + (xk+i . (21.mQf Xk L ~2e^ ^ . 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. (21.. One therefore removes the troublesome factor by defining (Sn being the sum in Eq.(x fc+ i ..2 + (x + x f c + 1 . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.Zfci) 2 } I +{xk+i x*. We then integrate and finally let /x go to zero.
In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. i.2). Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2.Xi.g. however prefixed with a new factor.e.t In the following we demonstrate that the function constructed in this way is. 21.Xi. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. p.5). G(xf.e.U) = / JXi=x(ti) = K(xf. the Green's function K of Eq. and will therefore not be attempted here. (21.10) can be looked at as the product of a succession of free particle Green's functions (or propagators). (21. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. (21. See the square root in front of the exponential in Eq. Different constructions of these families of paths usually lead to different measures. B. 11 For discussions see e. as in our discretization in Eq. (21. and so on.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. Felsager [91].5). We shall see later (with Eq. p. These integrations are indicated by the horizontal arrows in Fig. Felsager [91].1. However. summation over all possible paths. It is neither a variable of integration in T){x} —> Ylndxn.21. (21.7) for n — 1) to give a time interval. 183.tf.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. nor the limit of integration of some variable Xi. (21.2 The Path Integral and Green's Functions 507 Eq. which all vary from —oo to oo. i.tf. § See e.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure.tf. it may be noted that with a different philosophy concerning the summation over paths.ti) for tf > U.Xi.ti) For tf > ti we now write* G(xf. .g. 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. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq.26)) that — ignoring the potential — the multiple integral (21. (21. 183. B. in fact. we arrive at a path integral without the measure factor N(e) above.24).
. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. 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.tf + e.xi. We have G(xn+2. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. one with the full action S.ti.74)). 187 . It is for these reasons that — unless required — the measure is frequently ignored.1 and allowing n to go to infinity. however. we obtain the same as before except that there is no factor N(e). I We employ this method later in Example 23. Then the measure drops out. (23.508 CHAPTER 21. Expanding this in a Taylor series around x. 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. Then we have on the right the factor G{x+rj).3 in the evaluation of I a specific path integral (see Eqs.Xi.ti). Consider a time t = tf + e for position Xf = xn+2.Then we can construct a recurrence relation as follows.ti) = 1 j^p: f°° cte n +iexp } • fm0.tf. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. pp. 21.U) imo 2 G(xn+i W)Idr. so that the piecewise linear paths fill the entire space of paths.exp 2e n ieV(xn+2) G(xn+2 + V. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals. Each of these linear pieces is completely characterized by its endpoints. Felsager [91]. the discussion in B.72) to (23. (21.tf + e.188. we have See. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. the other with that of the action of the free particle SQ.Xi.xi. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.
It follows that ip(xf. d and this means d i—G(x.15) We see that G(x.xi.tf.tf + e.Xi. tf.6c).ie 1 d2 2nV0dx^G{X.Xi.ieV(x) + ri< —G(x.21.t Thus + 0(e2 (21.tf.tf.— p G ( j . U) .U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x. Retaining only terms up to those linear in e.tf.tf.U). tf.Xi.tf.tf)= dxG(xf.tf + e. (21.U) +••• \ 509 G(x.xi.ti) for tf > ti (21.Xi.ti) 2 —l€ — ^2 + V(x)G(x. ti) satisfies the timedependent Schrodinger equation. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.ti) +V(x)G(x.Xi.6b) and (21.(21. U) = e^G + 0(e 2 ) dtf 1 d2G(x.ti) N(e) or G(x.tf''Xi. tf + e.xi.tf. .xi.16) .tf + e.ti) = 1 d2 + V(x) G(x. 2mo dx2 (21.xi.G(x.tf.Xi.ti) +\if(^G{x.tf.6a).ti) .2 The Path Integral and Green's Functions so that G(x. Xi.xi. U) + .Xi.xi. tf.ti)ip(x.14) = G(x.xi.tf.Xi.ti)\ .U + 0(e ). we obtain: G(x. tf.ti) = j^rr dnexp I ^rj1 .tf.13) We integrate with the help of the integrals (21.x.Xi.ti) = G(x.
Xi. (21. denoted by KQ. as this is also called.U).*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. (21.15) is a wave function (cf.tf.510 CHAPTER 21. K0 = 9(tf .x. Eq. 21. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation. (21.8)) Go(xf. (21. We have therefore established that the Feynman integral (21. In a problem where V is small it is convenient to calculate K with a perturbation method.16).x.x.ti) = = 9{tfti)—G + G5(tfti) .tf. is the Fourier transform of the nonrelativistic propagator. 21.1)).x.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. Hence our original function K is the same as G. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. and the Green's function KQ\ S0= Here (cf. The function G which satisfies the timedependent Schrodinger equation (21. now called So.tf.U.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.ti) = 5(xj — x).U) = G(xf.tf. as may be verified by inserting this into Eq.U)G0.U) = / T){x} eiS\ [fdt \mx2. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. or K(xf.20) .10) is indeed to be identified with the timedependent Green's function. 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.3.17) Differentiating K with respect to tf we obtain —K(xf. (21. Eq.U)6(tf . The result.
. n + i .37) — by explicit solution of the differential equation which the Green's function satisfies./ .ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.z / c .tf. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.Xi.Xi.tf. We obtained the same expression previously — see Eq. f c .7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21. . exp where Xj = xo and Xf = xn+\ and there k = 1).ti)9(t).Xi.21) fe=l and t = (n + l)e = tf — U and h = 1.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.tf. (21.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.ti)= iX (o.. (21.tf. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.21.tf.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.1 we show t h a t oo /. we have Go(xf.24) ( s / . (21. Then Go(xf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.ti) Ko(xf.ti) = = G0(x. In Example 21. 2t Go(xf. But t = tf — ti = (n + l)e.23) We now insert our expression for In into Eq.t) K0(x.3 The Green's Function for Potential and hence Go{xf.x 0 )" n + 1 (in agreement with Eq.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.Xi. exp 2 (21.Xi. (7.23) for Go and obtain Go(xf.tf.i ) 2 .22) • \ n/2 .Xi.25) T h e quantity Ko(x. (21.8).
xo)2 + (x2 .)+(1"+2.6a). (21..x „ + i ) 2 = (x„ + 2 . Solution: The result can be obtained by cumbersome integration.ti. as / JXi f r D{x{t)}eiS°lh . If the potential V(x) of Eq.512 CHAPTER 21.22).5) is carried along. (7. For n = 1 (see below) h This result can be verified like Eq. we have instead of 5*0 and KQ expressions S and K.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. or else by induction. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.10) in a product form. 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 (21.54)..xi)2 = 21 xi ^2o(„. (21. Then oo 2 / (21.26) with n integrations and n + l factors KQ. (21.t0) (21.e. (21.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i. Example 21. ^ (xi .x „ + i ) 2 = ( x n + 2 Xi + Xi .)!} J — OO . (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .7) by direct integration with ^ . the latter being a Green's function with interaction like that for the harmonic oscillator.f_^y e. In our proof of induction we assume the result is correct for n. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .1: Evaluation of a path integral Verify Eq.tn+l•xn.25) we can rewrite the Feynman path integral or kernel (21. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .2y(xn+2 .x ^ 2 + y2 ._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.x0. i. Eq.tn)••• K0(xi.22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:.
n + 2 . (21.21. ) H — (xn+2Xi) n+2 n+l n+2 y2 ..27) we have (with a = t/2m0) .6a).2.27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). A(n + 2) 21.2 y ( a .2y(xn+2 . We pass over to momentum space with the help of the following two integrals. i.e. (21. The second integral is an integral representation of the step function.a .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.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.i) H —3/ .max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a.Xi) + ( x n + 2 Xi)2.3.i) . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . (21. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. (n+l)A n n j(a. 21. and + 2 2 l n + 2 ( x n + 2 .29) Hence with KQ from Eq.i j ) H n +— z 1 . Using Eq. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 . a dp exp [ipx + iap2] = exp — i(21. it .25): Ko{xJ) = = (S) e{t)(^X'2 y J . e > 0 .n+2a. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .x2rriQ 1/2 toij_. The first. This quantity is also described as the nonrelativistic free particle propagator. (21.Q .
The first order correction contains one factor of the potential V.*' where 2mo — r. The generalization to three space dimensions (required below in Eq.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. This is the type of expansion most people describe as perturbation theory. i.e.Q (21. 21.2 Contours of integration. Thus in first .t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£.28) this becomes K0(x. ( 3 b) ° The expression E+ V — it 2m. an expansion in rising powers of the coupling constant (contained in the potential). Replacing 9(t) by its integral representation (21. the second order two factors of V and so on. (21.514 CHAPTER 21. 21. The Feynman Path Integral Formalism Im t \t>0 Rex Fig.51)) is fairly selfevident.
t) JSo dtV(x.)} (21.. Eqs.4 Including V to First Order: Perturbation Expansion h=l.^.^){iV(a.^ J_00 V(xk._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 .21.5) a n d (21.ti.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x. . . dxi ( m0 V V27rie/ m0 \^He \ exp fe=..U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.t/.(xfc .t)\. \2meJ n+l J.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.t) the first order correction is (G\ ~ G — Go) Gi(xf.i. .Xi.32) Since — we recall this for convenience G(xf.U). (21.i.tf. (21. li f f 515 order (cf.xi.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.tf. ^ = 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.. } e °° f l f •/ oo /"oo F ( a . (21. dxr. x (n+l)/2 \ lim . .. dxn exp i ^ —.Xi.33) xG0(xi.t)] n+l .xfc_ OO J— OO J—OO L J.i V e / D { a .+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.
37) Jti The last two contributions do not describe the time evolution from ti to tf. (21. The factors of "V" originate from the series expansion of the exponential exp ftf .tf. They must therefore be deleted. t') + • • • .xi.U) xV(xi.ti)K0(x.t)V(x.t2)V(x2. Xi.t)K0(x.t2)V(x2.tf.xi. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. in writing down the expansion exp i dtV(x. i.516 CHAPTER 21. dx2Ko(xf.x2. (21.e.t) =li dtV(x. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.36) {if •j.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).t2.t.x. t) Ju Thus the term containing n times V contains a factor 1/n!. OO /"OO / oo dt / J—oo dxKo(xf.ti) t2)K0(x2.i / dtV(x. (21.t2) xKo(x2.ti).e. = i dt2 / dx2Ki(xf. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.x2.tf.t) (21. i.xi. / Vdt+ JIfVdt)( fti = Jti Jti ftf ) \Jti pti Jti I' Vdt' + f f Vdt' Jti rtf Vdt' J t\ j Hi Vdt rt\ Vdt' Jti Vdt Jt\ Vdt'+ Vdt+ Jti + \ S Vdt I Jti f Vdt'.ti). .35) We also observe here that the problem of timeordering arises only in perturbation theory. Jti Consider the term of second order. t2. This is cancelled by n! different time orderings.t) ! Jti r*f dt'V(x.I dtV(x.
transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. tf. For various reasons — such as illustration.ti).38) .x. x"^ X x"2.Xi.tf.ti) i dt = K0(xf.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf.t) is V(p.t)K0(x. we are concerned with quantum mechanics.Xi.Xi. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.tf. of course.t)V(x.tf.t)Ko(x.x.i ] P I dt J dxKn(xf.E).xi.t2 K o (x 2'l2 . (21.39). Such diagrams in the context of field theory are called Feynman diagrams. 21. The diagrams represent mathematical quantities and hence are designed on the basis of rules. . (21. called Feynman rules.40) The rules for the present case are given in Fig.21. 21. x r l i 'V(x.ti) (21. E) which is given by (in onetime plus onespace dimensions) V(x.t)= f dpdEeipxiEtV(p. U) (21.Xi. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. or in momentum space as mostly in field theory. Here.ti). The Fourier transform of V(x.t)V(x.t. These may be formulated in configuration space as below.ti) = K0(xf.tf.t) ) K K K0 'V X K o Fig.3 along with the diagramatic representation of Eq.39) dxK(xf.t.Xi.3 Feynman rules and representation of the Green's function.tf.
^(x. with p = Kk. 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).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. and dn = dnxdnydnz. (3) The transition probability per second is = \A\2 . i. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.= l. The plane wave function normalized to one over a box of volume V = L3 is.Xi. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. T = duration of time for which potential is switched on. (21.tj)K{xf. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. for the allowed values of k: kx = 2imx/L.ipf have to be properly normalized. (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.7 = e i k ' x _ i : l i with f i/}*i/jdx.tf. here with E — k 2 /2mo.ti). to one particle in all of space.41) We have to pass over to three space dimensions.k = 2wn/L. There the wave functions ipi.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit threemomentum interval** = V/(2irh)3. The Feynman Path Integral Formalism 21. (21. We recall for convenience from **Note that from the wave function we obtain.e.518 CHAPTER 21. mo V since ip is normalized to one particle in volume V. * k 1 • .t) = . . dn = (L/27r) 3 dk. i.ti)il)i(xi.e.
47) . u = ^\H'kfp(k).4)... Then for stationary states M^t) and itm^ = f HUt'V^t'dt'.21. «= — . H = H0 + \H'. (10.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. Eq. i. Then H'ki can be taken out of the integrand in Eq. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.— .k.44) and f_oo can be replaced by fQ. (21.e. and with these the following perturbation ansatz: ih^* with an(t) = HV.44) = u(Ek. The transition probability per unit time is . P(*) = j r (2145) (cf. 47T 1 3 7 (21. Proceeding in this manner one obtains Fermi's "golden rule".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. (21.t) = e~^lT\ r where r = Ixl. # = ^an(i)VnM) n = a^(t) + \a^(t) + .117) and Example 10.= t7Ei4 1} wi 2 k It is frequently assumed that the only time dependence of H'(t) is that it is switsched on at t = 0 and switched off at time t > 0. V(x. in which T is a large but finite time.
ti)'ipi(xi.tf)ipi(xf.51) Inserting this into Eq. (21. We can relate this amplitude to the amplitude ark of Eq. Ai = dxf / dx.44) (here in the scattering problem with initial and final energies equal). we obtain A1 = clS(ti tf) / dxf(i)V(xf. (21. also called Born approximation. so that 0(t) can be dropped) K0(xf. dt dxip*f(xf. t. ti) = co5(t — U)5(x — Xj) etc.tf)ip*(xf. t)K0(x.tf. This now has the form of the amplitude (21.48) where from Eq.Xi.39) OO /"00 / dt / OO J —OO dxK0(xf.x.tf. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.* / . t. The threedimensional generalization is (taking tf » T and ti <C —T. Thus with the ansatz Ko(x.tf.f. (21.x.t) = ^ J*ie*<*'*><<*'*>.tf). (21.ti). (21.t. x.44) by contracting the propagators to instantaneous point interactions.t) (21.30a) K0(xf. In our previous onedimensional considerations we had with Eq.tf)K0(xf. (21.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .ti). (21. Xj. £.. (21.52) .tf.520 CHAPTER 21.tf)Ki(x.i x (i)V(x. The lowest order approximation of A.50) Here x — xj — Xi and t = tf — ti. (x*.). The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.yLi.49) i.^ / .t)V(x..e.tfx.>/>. we have to replace the onedimensional quantities dxf etc.xuti) = ^ J°° d p e i p x _ i ^ * .t)K0(x. T a large but finite time. by threedimensional dxf and so on. i.f.
).56) (21.p ) .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)exp A(p. £).2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21. (21. 2 „ x v~ m o (21.55) 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 . k2 ipx + i—t + ikx — i 1 2mo 2mo (21. this becomes A(p.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p.57) we have A(p.r) = = y .~ .T) = x exp L z/ q/2 k2 521 f dx. Then integrating over q and q' we obtain .p 2.59) . Then we require the following integral.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 . which is evaluated in Example 21.k.T) = JdtJdx(^\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 .k. .i ( k 2 . (21. At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .T) = ^JdXJdt%*t2/T2exp i ( k . P2 .54) Now inserting the expression for the potential V(x.iaT2/8: L With 0 0 • .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.k. (21.21.
e. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).P ) 2 + M2]2 4 VvrT2.< ^ £ \M VTTT2. k .62) We want to consider the limit T — oo. write down the probability per unit time which is ( 8 \1'2 _a2 16^ .522 CHAPTER 21.3. dE = PjV = ^ E ± . This expression assumes that the integration over time t is normalized to 1. i.„ 2. e 8{k) = lim 7=*—^ (21.k.T)2/T.„. (21.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .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. ( 21 . / / P 2 _ .63) We establish this result for T — 1/e in Example 21. Using Eq. 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .^ . (21.60) Now.65) But E=*. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T. T ) ~ . we obtain . e (21. 66) o .59). k2 .\ 2?7lo / 1 (21. { I i _ F T l } .T2f 8 V/ 2 .64) Inserting the result now into our expression for the total cross section." + M ] p) ' "•"• ^2 . <rtot. y 2 [ ( k . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. .63) we have: Probability per unit time = 77777. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . otot contains the factor \A(p. Using the result (21. mo m (21 .51 — 2 /w2 y ~ —.a ( — ) ^ .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2. In our case this has not been done.
and inserting the expression thus obtained into the final result of Example 10. so t h a t (21. /Yukawa = ) Solution: We choose the 2axis along the vector q. giving the potential a coupling constant.p ) 2 = 2 k 2 ( l . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21. esfn (21.^ .21. p 2 = k 2 ..5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .e.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .65) with the cutoff M > 0.j . r\ poo Yukawa = = ^ Jo 2TT / r e~Mr y_i dre~Mr d(cos 9)e^r sin qr = — G / cos e = 2TT / J0 —e~Mr iqr { eiqr .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. This result may also be obtained from lyukawa of Eq.67) and (21. (21.4. setting M2 = 0.68) Inserting (21.70) . (21.^ . (21. we obtain f mlV2 dQ v / 2 m ^ 3 2 7 T 3 a 2 atot r) 3 fcV 2 4fc 4 4sin 4 (0/2) ' JW) da _ d& = i.69) This is the wellknown Rutherford formula.M £ M * = / m0kdn = m0^2m0E (k .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 — .67) T h e delta function here implies energy conservation.e. Example 21.= e " f c a ^ o 2v^fa /4a = lim £^o —.cos 9) = 4k 2 sin 2  .59) by replacing there q 2 by ( k — p ) 2 .68) into Eq. 2mo / dft. i.j .e .* 9 ' d r e . —Mr 4w = . Then (x = r) foo „2jj.
a.70) and (21.71) with respect to k yields the same as °° 2 .^0 a^C 2^/TTO.* . We consider again nonrelativistic quantum mechanics.71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . a simpler method suffices. 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). We write the state vector of a particle of threemomentum p in the momentum space representation: p). We have x> = I J ^ l p X p l 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 . (21. The two representations are related to one another as we saw in Chapter 4 by the following expressions or Fourier integrals: p> =  d x  x ) ( x  p ) .524 The function required is CHAPTER 21. (yx) = 6(y . Since dxe" 1 1 1 = « / . Differentiation of (21.75) . In configuration space the state of the particle is written x).. Va we obtain g(k) = Je~k V a = 0.oo J / )eikx. Direct integration /4a .x) (21.72) co From (21. however. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. ' i a = lim e —0 pk 2 /e2 e^TT 21.6 P a t h Integrals in Dirac's Notation It is instructive to devote a little more time on the Feynman formalism by rewriting it in terms of Dirac's bra and ket notation.q). '<*> = . (21. In the present case. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. (21.
so that \if>t=o)s = H>)H. Thus (qPp) = p(qp) = p(27r)3<5(q . i. The time development is given by the Hamilton operator H.83) (21. M^ip(x..t). we must have d/dt{x. which describes the propagation of the particle from its initial position x. The equation 1 = J ^p)(p.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. i.i'x. t) = (x.p).77) (21. at initial time ti to its final position xy at time tf.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. Then the states {x..x.i) = e iH4 x. pp. (21. (21.t).t> =iH\x. 4) is to satisfy the Schrodinger equation. (21. x.82) (21. t' — t. (21.t)} constitute a "moving reference frame" in the sense that ip{x. See also L. (px) = e~ipx.160.t) = e* H t x). 159 . H.76) These relations imply the completeness relations 1 = J dxx><x.i) = K(x'. .21.t). The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H. Ryder [241]. i. i.e.e.t'x. Similarly we have a position operator X with Xx) = xx).If '/'(x.t\ = iH{x. (21. We consider first an infinitesimal transition.e. (with ft = 1) d/dt\ip) = —iH\tp).6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . Hence for the infinitesimal transition (x'. Then ( X  ^ . i.e.e.82): The general Schrodinger equation has the state vector to the right. so thattt d — x.We define x.^ l x ) . t\ipo)s.£) = ( x V ^ ' . x' — x infinitesimal.t\.We now introduce the Dirac bra and ket formalism.t) = —iHip(x. d/dt\x.80) (21.t'. ) H = (x</>o)s. as had to be shown.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p).81) Comments on signs in Eq.0) = e""x).t).t) = +iH\x.
V 2 < 5 ( x .e.x = V(p .x).i p '.91) . Pq) = J 7^3Pp)(pq) = J dppp>*(p . "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.p').x y(x)e i p .88) Thus P has the operator representation P= /dxx)V(x. (21. i.85) the amplitude or expectation value or position space representation (x'#x) = .* .85) The momentum operator P satisfies the relation P = /(2^3PP)(P' (2L86) so that it projects out the momentum q of a momentum state q).y<&(p'ix')(x'ivix)<Xp) = f dxe.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).x') + V{x)5(x' .P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«. (21.. 2mo Since (21.90) (P'IVIP) = y<&.89) Using this and previous relations we obtain for the Hamiltonian (21. The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P '  e ""'"'' )  P ) < P  X > e. (2187) (21.526 CHAPTER 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 '. (21. 527 (21. the signs are reversed when these operators are applied to the ket vector \x. (21.p) + i{t .92) We now return to the expression (21. ti to x j .t) (cf.21.82)).93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.tj+i\x.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 .fx.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p . But e'ipx + 0[{t . i.84). (x'.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. We now consider the transition of the particle from Xj. i.i'x. tf subdivided into infinitesimal transitions (xj+i. EQ — Hi .e.j.p') + V(p . We have <x'. for the amplitude of an infinitesimal transition. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.t') \^(27rf5(p 2mo +V(pp' where we used (21.p') • • • j W < 5 ( p ' .tj). t).p').92).
e.94) and thus obtain a phase space expression.i { f l " ( p i ..x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf. d x n . = 1 X X<jPfc We now replace H(pfc. dx.tf\Xi. t n _ i ) t„2><xi.xk (21.+ V(x f c ).X n _i)] h) (21. . i.. (21. x exp [ .dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .tfei) fe=i .ti) rx/=x(i/) d x i dx2 . We now insert for each of the infinitesimal intermediate transitions the amplitude (21.. then corresponds to the sum over all paths. x f c ) = ^ .27) in which we make the replacements x —> eifc.x i ) H h pn • ( x n . .e.98) tk — *fcl # ( p f c . d x i .x 0 ) + p 2 • (x 2 ..t 0 ) + H(p2. zmo T h e n t h e integration over all p ^ can be performed by using Eq.a — —e/2mo and divide b o t h > sides by 2TT.xn)(tn in_i)]. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/. d x i ( x n . d x „ _ i . i. T h e integration over the n—1 intermediate positions.tn respectively. x i ) ( t i . .. x i .i/xj.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa .t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21.528 CHAPTER 21. ..99) .. . .tf\Xi.xic) by fc — Xfej H(pk.2. .tixo. (Xf. .tj) = / ••• / d x „ _ i d x n _ 2 .97) H hH(pn. x n which the particle passes at times tQ.ti.. t n  x n _ i .i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .
fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . G.tf\xi. Truman [254].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. (21.8) as N(e) so t h a t the integral does not vanish for e — 0. / {xf.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. (21.H(x.e.101) Then Eq. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. to / m0 \ 3 / 2 las.e.100) over V{p}= L J ^ rwooll V (27T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O. Tokarev and A.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21. (21. i. Smolyanov. i.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. > Thus the phase space representation of the pathintegral* (21.] ex H ^I (.. (21.21.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21.103) is another form of the pathintegral for which the integration with the help of Eq. G. A.98).ti) = J D{x(t)}D{p(t)}expi [' dt\pi. .p)]\.
H(xfe. i. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk.x) . operator (21. . pfc) = Pfc • xfc .105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. (21. (21.iJ(x fc .106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).t'}.8). We have now obtained the path integral representation of the matrix element ( X / .e. Thus finally we have Feynman's principle: (Xf. (21.ti) where nl D{x(£)}exp '. p .8). t to x. Garrod [106]. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. pk) = L(xk. £ /  X J .ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation.tf\xi. nl (xf. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. xfc). I hi dtL(x.104) where for the particularly simple Lagrangian L under consideration here m0±l . x.x) (21.530 CHAPTER 21.V(xk) = m0±l . t. 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'. £ J ) with no operator in between.tf\xi. this agrees with the corresponding discrepancy in the onedimensional case of Eq.
i p '• x .t'}. Thus <xV'xx'. (21.*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(./ ^ I < « . .76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). (px).* * f dp = j (2^ p) ^ d .5 l^ t 'lp / > e . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".109) .* * .108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.v ' x ' and thus {x".6 Path Integrals in Dirac's Notation For the verification we use Eq. With the wave functions (21. . (21.21.0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'.t"\x5){x5\x'.t"\x\x'.77) and thus have 531 *!*') = *M = * 7 ^ I P > « . . ••• 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 (21.107). With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. x fdxG  ^ P <p"x3)<x3e^V) x (21.
The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK. We now proceed from there along parallel lines.0 dp x„ / (27T): 1 .£'> = x"[«5(x" .85) (xVxx'. We now consider the transition of the expectation value of operator x from Xj. tn2) • • • (xi..i ( t " .£j) = / ••• / dx n _idx n _2 .*/<x"x5)(x5tfx'> + . (21.t') (xV'xx'.tj) at equal time intervals e.t') dp . Inserting here the integral representation (21. t0). (21. Thus with x^ = xn.• dx1 (x n .113) Inserting for each of the infinitesimal intermediate transitions the amplitude . = x"6{x" . x / / e .£/xxj.x ' ) (2^ i^+vM + D «p(x"x') (21.ti = to as before. . we obtain with (21.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • . This is the expression corresponding to that of Eq. ( x " .i{t" .i t " ( x "  #  x 5 ) ( x V ) +.t"\x\x'.112) where x" is a cnumber.tj+i\x\xj.£j to Xf.94) for the case with no operator in between.90) we can rewrite this expression as (x".t n _i) x ( x n _ i .532 CHAPTER 21.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') . we obtain (here and in the following the limit n —> oo at the end being understood) (x/.i(t" .£ n xx n _i. £ n _ixx n _ 2 . (21.x') . ] . .t'){x"\H\x') = X + •••} (21.x') . tixx 0 .f ) i f e i p .110) Recalling Eq.111) 1 .iit!' 1') 2mo + vtf) + <5(x"x').tf — tn and x$ = XQ. (21.tf subdivided into n transitions (xj+i.74) of the delta function.
t"\x'. . Xn _i X„ xi dx. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.. dyi l xi x 2 .3 " n d x f c .{^( P i.t"\x'.t') = (x" + 5x".2 .114) x exp [i{pi • (xi . dxnn _i Xi X2 .t') sider the variation S(x".105).x n _ i ) ] .x 1 )(t 1 Z & ) +H(p2.. x n _ i x n reduces to the single factor x^.7 Canonical Quantization from Path Integrals (21.{x". .« • > We reconsider therefore the transition amplitude (x". we obtain* (xf.t"\x'..7 Canonical Quantization from P a t h Integrals We saw above that in the path integral method the evolution of a system is expressed by a basic functional integral like (21.2 .y (27F (27F •" (2^(2^)3 exp[. (21.t"\x'.and hence may be put in front of the integrals to yield (xf.iyxxj. n—>oo *•»• 21.x2)(t2 ~ h) + • • • + H(p ••• + p „ .21.t"\x'. (21. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 . (21.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq. tj) = xf (x/.10) or (21. the string of factors xi X2 . ..114).t') .115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .tf\x. x (21.x 0 ) + p 2 • (x 2 .. We demonstrate first that the operator representation of P given by Eq.89) is consistent with the conjugate expression dL (2L116 * .. For simplicity we consider the case of one spatial dimension.( x „ ..t') = and con 5x"Vx„{x".xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. U).ti) 533 rxf=x(tf) = / • • • / dXidx. tj  xj1 x».\Xi. tj \ xt..112).t') (21.x .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.
we separate the metric factors from T){x} in Eq.(x".118) — = (21 5x"(x". Thus in this case of classical mechanics dL_ d fdV.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). x) Jt Jt dL.tt).t') =  6— / 1 V°{x}t . Hence w « = (t).t'): 1 fx" 6(x". (21. " (t).t'). dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. dL r /•*" . and we demand the validity of the equation of motion. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x.t"\x'. Then.t"\x'. <9L' dt —dx + — 5 x ox ox f . with N = N(t" .x). (21.118) JI(x)/h . The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.534 CHAPTER 21.10) and set D{x} = D°{x}/JV(i/ . (21. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. jt/ 37 ( £T. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.l&cdi. For the purpose of enabling some formulations below.122a) = 17) 5x"Vx/. . dx dt \ dx _ (21.i/(x)/fi ) j ^ ) dtL(x. t".t i_(dL_ x't' h\dx J t„ (21.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.
t')(x'.Xf = x(tf).1 (A / ).122b) By comparison with Eq. 535 (21. Wightman [264].126) R.125a) (2) the completeness relation holds. (21. F. § The evolution of the system is expressed by the Feynman functional integral (xf. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti).7 Canonical Quantization from Path Integrals or JL(x»y\x. x' (21.1>) = ^{x".1 (A)iV..125c) Corresponding to our relation (21.x). we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. (21.tf\xi.21.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21.t"\p(t")\x'.115) in three space dimensions. I(x) = / JA dtL(x.t'). . The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory. S. 6 (21.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. ^2\x'. / Jxi V°{x} / Jx' V°{x}.t'\ = l. (21. Streater and A.125b) and the measure property (3) N~\A + A') = A^.124) where with A = tf — ti.
t"\x'.114) (xf.129) = ^ y With Eq.t').tf\T(xix2)\xi.tf\x(ti)\xi.ti) = x' 'Y^{xf.tf\xi\x'.ti). Eq.t')=x"(x".. Thus (xf.ti) = —!— p V0{x}xx • • • xneu^h.t"\x'.t\x'.t') (21.t"\x"\x'. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.130) ~i5x7jX"{x". We consider the case n = 2. Then the relation follows with the help of the properties (1).536 CHAPTER 21.122b) we can deduce the nontrivial canonical commutation relation arising here. (21.t') + x"^J^{x".t'){x\t'\x2\xi.128) where tf is later than ti and t\. (21.x](x. (21.t').t') = ih(x. Thus if T denotes such ordering.t\x'.ti) f ' V0{x}xlX2eu^lh..ti) = Xi(xf.127) We can now write down an expression for a timeordered product of operators.t>). (21.127)) (x".133) = ih{x". (21.t').. (21.tf\T(Xl • • • xn)\xi.t"\x'.t"\X'. we have the relation (xf.tf\xi. tn lie correspondingly in consecutive order between U and tf.132) The generalization to higher spatial dimensions proceeds along parallel lines. . We have (cf. But for t = ti we have (and equivalently for t = tf) as we can see from Eq. We can rewrite this expression as the commutator relation (ITU*" x"^](x". (2) and (3) above. (21.t"\x'.t') \ i ox" i ox"J or in commutator form [P.t"\x'.. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21. (21.131) = ih(x".
in the context of quantum mechanics are not convergent. p.g. Such an expansion parameter is usually known to be or assumed to be small in some sense.g. i. it has some finite. T. Expansions of this type arise. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. for sufficiently large r. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. W = e x p ( l / z ) . Such expansions are strictly speaking. If the expansion is mathematically well defined. In physical applications such cases are rare.e. i. in the sense of rigorous convergence tests. the series converges. Consider e.e. An essential singularity is really an ambiguity rather than an infinity.g.g. such as the fine structure constant of quantum electrodynamics. z = a + i/3 with a — oo or f3 —> oo or both. see e. although the first few terms indicate a convergent behaviour.e. nonzero radius of convergence in the domain of the coupling parameter. so that successive contributions to the first approximation do not invalidate this approximation. 102. N. divergent.e. Watson [283]. but asymptotic. integrated over) beyond its radius of convergence.Chapter 22 Classical Field Configurations 22. E. Whittaker and G. Besides. The vast majority of perturbation expansions discussed e. > 537 .1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. It was therefore natural to search for alternative methods of expansion. i. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. as we have seen in Chapter 8. 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.
e.e.h2. A further challenging task was the development of methods of quantization of theories which incorporate classical. and the corresponding expansion is called the "loop expansion". and in general this change is accompanied by certain constraints. On the contrary... cnumber.h°.h.538 CHAPTER 22. 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. higher order corrections are of orders ±. cnumber first approximations. although this terminology is not precise. the latter's topological properties if any. However. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27..e. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. The classical.h3. In searching for other means of expansion. There. These methods are often described as methods of collective coordinates. 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).) l/h2.. 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. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. before we reach the stage at which quantization can be considered we have to deal with numerous other aspects such as the stability of the classical approximation.g. A method which achieves this is in particular Feynman's path integral procedure. such as soliton theory. 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. In the following we consider various typical examples. This type of expansion is decsribed as "semiclassicar. The fundamental extension of the method of canonical quantization (i. such > a series begins with a contribution proportional to (e. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". The study of expansions of this type has turned out to be extremely fruitful. and has led to insights which previously seemed unimaginable. i. One therefore faces the problem of quantizing a system which is subject to constraints. nature of the dominant approximation does not imply that this describes classical motion. i. Many of these aspects are interesting by themselves .
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.1) ' T h e reader who wants a highly advanced overview of instantons. 22. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". d^] = d^d^ . 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. (22. and also to see these in a somewhat broader context (by comparison with higher dimensional cases).* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. vortices and kinks. 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).V(4>).g. V{</>) = m§0V + A0(<^)2. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration.22. monopoles. However. 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>. it is important for a better understanding of the considerations which follow. . Tong [272]. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. This means theories with field densities which are therefore infinitedimensional. modified by going to imaginary time). may like to consult the article of D.2 The Constant Classical Field 539 and in any case deserve detailed study.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. to keep this case in mind. or even a solution to a modified field equation (e. 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[<t>] m 2>o o Fig.e. £)) is bounded from below we must have Ao > 0.e.2) We ask: Are there classical (i. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. § (22. B. Felsager [91]. . 22.£). as illustrated in Fig. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. cnumber) solutions <j)(x) = 4>c = const. 22. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. The EulerLagrange equation is d„ dC d(d. m l + \04>*4> )$ = <). and hence here [d^ + m20}(l) = Xo(f (22. An even more familiar example is the GinzburgLandau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape).1.1 Different potentials for different signs of m§.3) See e.S£/<90(x. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x.540 CHAPTER 22.g. of the equation of motion? For these we must have all derivatives of <j) zero. 433. V ) . p. i. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. Classical Field Configurations where d^ — (—d/d(ct).
5) . 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. with 0 = 0. 22. The Hamiltonian density H is defined by the Legendre transform H[<f>. we have 541 0. w A iP 3ml 5* "XT" (22. TT = —J.e. 22.2 The spontaneously chosen phase.</>2)plane.ir]=7r<fiC.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. not the solution x(t)). aq> For (b = const. the only such solution is the case we wish to consider. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. and so ft[<M=m6:(^* 1 + Ao(0*0)2.e. (22. the equation has this invariance. But for Ao > 0. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. TTIQ > 0. Fig. i. We can convince ourselves that <fic is the field associated with a state of minimized energy. H = —C = V.2 The Constant Classical Field For Ao > 0. mg < 0.22. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution.
Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. and we wish to investigate the behaviour of the field cf> in its neighbourhood. the one for (3 = 0. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. For every value of the phase (3 we have a different constant cnumber field configuration 4>c.7b) In Eq.542 CHAPTER 22. e. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. in a transverse direction.e.tp2. &H/d(fi = 0) if [m20 + ^>]</> = 0. (22. >° for m 0 < °> . 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.Then m\ m\ = ml + A0A2 = m2.7a) (22. (22.3) obtained above. 22. (22. Classical Field Configurations the density Ti is minimized (i. i. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .2) and separating real and imaginary parts. and partly by climbing up the parabolic wall on either side.3). (22. . along the trough of V (which we can call a longitudinal direction). we set cf>(x) = <f)c + ri(x).3ml = 2ml = ml + TA 0 A 2 = 0. V?) (22. Suppose we choose one such configuration.6) Inserting this into Eq.e.e. This is the condition (22.g. i. as indicated in Fig.7b) the constant on the left and the coefficient of A on the right of Eq. i. 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).2.e.7a) vanish on account of Eq. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 .
has to be removed in order to allow a well defined perturbation procedure (i. though not identical phenomenon. (3=0 Fig. Thus the Green's function is similar to that in electrodynamics. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. r'. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. It is known from there that the vanishing mass of the photon together with the transversality of the electromagnetic field implies that only two of the four components of the fourvector potential A^ are independent. the existence of individual perturbation contributions).7b). ^ = h eip / \ ^2 l . which. t. 22. The elimination of the . (22. whereas ipi(x) is y ' " " " " " ^ . 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 £. t') = d(r . then a massless boson exists. of course. tangential to the classical path.22.r')S(t . Goldstone's theorem applies to fully relativistic field theories.3 The spontaneously chosen phase seen from above.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. This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction. irrespective of the question whether the perturbation series as such converges or not. There the wave function with an associated vanishing eigenvalue is called a "zero mode".e.e. We can see the problem here by looking at Eq.t'). i. 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. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory.
Soliani [237]. MullerKirsten [291]. Lohe [179].3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. . We make one more assumption concerning V which.e. t Potentials of this type have been discussed by M. spinless field $(x. A. W. we shall need only at a later stage. To insure that the energy (see below) is positive definite we require V[$] > 0. Khare [17]. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. J. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. 7711 = — 1 ( 1 6 Minkowski manifold). Behera and A. 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. 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.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). however. *S. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. Zimmerschied and H.* We consider the theory (later: quantum theory) of a real.544 CHAPTER 22.t In particular the sextic potential can be considered along parallel lines. equates to zero the coefficient of this wouldbedivergence.* Potentials of higher polynomial order in $ than $ 4 can also be considered. Later we wish to develop a perturbation series about a classical configuration of $. i. F. C.g. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. (22.d^} = ~d^d^V[^) = ^2^(^j V[$]. The Lagrangian density is taken to be C[^. see e. N. instead we shall consider static and timevarying solutions in important models of one spatial dimension.£) in one spatial dimension. 22. and we choose the metric ?7oo = + 1 . Rebbi and G.
For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. (22.12) we see that they satisfy the Newtonianlike equation <j. In a quantum theory the field $ is an operator defined on a space of states. (22.g] = \ v m For instance.l]. The field $(x. i.e.§ Unstable classical configurations are also important and will be considered in later chapters. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. (22.l}.11) (22.e.12) where the prime denotes differentiation with respect to $ .t) has an explicit timedependence it is a Heisenbergpicture operator.22. The first such condition to be imposed is that the energy of the solution of interest must be finite.9) with quartic potential is defined by ^V\g$\ = ^V\g*. 1].13b) We are not interested in just any solution of this equation. t) is an observable (which in general — in field theory as compared with quantum mechanics — does not have to be hermitian although we assume this in the present example). i. From Eq. ^ ^ * + V'[*]=0. the socalled ^theory = ^V[g$. The other condition which we impose (and study in detail) is that of stability. at least in the present case. . the reason for this condition being that we visualize the classical solution as representative of a lump of energy. — m$ 1 — cos Im 1 I ^ $ (22. Since $(x. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. (22. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. such that J > — — — = 0.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.13a) We write such fields <j>(x).«{x) = V'(<p). but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series. (22.
17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. HI tanh m(x>k) 9 Fig. (22.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. . We observe here already that the integrand of the integral in Eq. d. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. n = & . 22. <9$ i. (22.T^ = 0. (22.e.15) The zerozero component is equivalent to the Hamiltonian density H. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.4 The wellknown soliton of $4theory.c = <£z + .T = <I>.14) where the overdot implies differentiation with respect to time t. i.2\dx 2 ' 1 fd& + V[$].7r) where IT is the momentum conjugate to $ defined by dC ? r = .546 CHAPTER 22. the classical solution will therefore have to be such that this integral is finite. (22. dC • Too = r?oo£ + =r$ = C + vr$ = H.e.17) Since we have to integrate over all space.
21) This sineGordon configuration is depicted in Fig. (22. O±m(xxo) (22. See J.e.18) Here XQ is an integration constant.3 Soliton Theories in One Spatial Dimension 547 As we shall see below. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". m ( x .13b) for the specific potentials (22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . < = — tan [e >  ° ] 9 2 rrm/g — Fig. 22.x n ). Rubinstein [240]. i. . In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .10) and (22. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. 4m . the $*theory.4.5 The soliton of sineGordon theory.20) (22. > . (22. In fact.1 . — XQ). (22.5.11). Again XQ is an integration constant. 22. it is not difficult to find classical cnumber solutions 4>c to Eq. 22. in the case of the first.18) is depicted in Fig.777.22. "The name sineGordon is derived from the analogous KleinGordon equation or theory.
9 2 4>{x) = i. and hence 4/71 9<t>{x. ± 1[tanh.1 "»JtKxo) 9 1 .13b) which we can write d \W? = !"(•».* ( m' m 2 r)] ^2" . Classical Field Configurations The solutions (22.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . In the case of the sineGordon theory defined by the potential (22.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . With we obtain g(/>(a. <^>(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 . .18) and (22.20) are obtained from Eq. Inserting the potential (22.e. 4m In tan tan 5 l/e±m(*xo)tan^£^)U. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ).o) =m7r. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. or \(<P'? = V(4>)+ const.548 CHAPTER 22. (22.
The expressions (22.4 Stability of Classical Configurations The concept of stability can be complex. Suppose now that we demand stability in the sense of minimized energy. the curves rising monotonically from negative to positive values are known as kinks. the discussion of Bogomol'nyi equations) in a simpler way. however. Jackiw [141]. Then we have to demand that the functional derivative of the energy E(<p) be zero.17).19). A state of stability is therefore associated with minimized action and/or energy. this does not require the action to be minimized. 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.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks".17) implies dx Jdx dx n£)' + ™ ( ' • # \ dx J S(f>(y) \dx and with partial integration we obtain f dx d fd<f>\ dx\dx J  5V(</>) 5(x y) + 5(j)(x) Tx5{x~y) . . (22.23) See for instance R. the others as antikinks. we obtain these expressions later (cf.21) for the energies can also be obtained from Eq. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. instead of doing this now. (22.22. 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. (22. The relation (22. However. 22.22) and that its second variational derivative be positive semidefinite. SE((f>) 5cb(y) 0. in the onedimensional Ising model. for instance.e. i. (22.
in fact. with eigenvalue w2. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.e. and the EulerLagrange equation retains this property.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. the integrated contribution vanishes. dx (22.a / 0 is not the same. (22. The eigenfunction 4>'c{x) is for this reason called a "zero mode".13b) with classical solution 4>c. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy).24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. In fact. (22.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. The Newtonlike equation (22.e. This is. i. the function 4>c(x + a). a very general phenomenon which can be established as follows. Eq. 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. (22. = 0. (22. Since it arises from the violation of translational invariance by 4>c. Thus the vanishing of the first variational derivative yields again Eq.13b) retains the invariance although the solution (f>c does not. The original Lagrangian was. (22. in fact ip0(x) const. we see that the latter has one eigenfunction.e. if 4>c{x) is a solution. it is also called a "translational mode". i. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space.25). of course. written down in Lorentzinvariant form.28) . we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.550 CHAPTER 22. its eigenvalues w\ > 0. in particular the invariance under spatial translations. if we apply the generator of spatial translations d/dx to the classical equation. i. namely (p'c. dx2 + V"{4>) (22. Classical Field Configurations Since y ^ ±00.A. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22.26) Comparing this equation with Eq.13b).
17) we obtain oo dx[V(4>c) + const. i. Then the vanishing of the first functional variation implies that 5S = 0. As stated above.e. (22. (22. and that S[UC] is finite (or equivalently the energy). lim [V(<f>c) + const.31) Inserting this into Eq. we are interested in classical cnumber field configurations of finite energy.e. (22. i. it is an expression like (22.18) and (22. this integral has to be finite. 0 = S'[UC] = S'[gUc}. We see therefore that the occurrence of these zero modes is a very general phenomenon.].30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. Thus TUC is a zero mode. (22. / oo Thus for E(4>c) to be finite. = 0. (22. We can choose the potential V such that the constant is zero.22.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. and we have seen that Eqs. 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. In the onedimensional case under discussion we can rewrite Eq. in fact.] = 0 or x—>±oo lim V{(/)c) = 0 for const. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. (22. We shall see later that these zero modes lead to undefined Green's functions for the semiclassical perturbation expansion unless one or more suitable constraints are imposed. (22. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC).33) x—>±oo .20) are such configurations in two particular models. We return to our considerations of stability.27).29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.32) = V'(</>).
10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig.c o and another for x — oo. ± 2 . This number is defined like a charge in field theory by the spatial integral of the time component of a current. 22. .e. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. ± l . In the sineGordon theory > • defined by Eq. oo J—oo y The topological quantum number q is defined by 0.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. Classical Field Configurations V Fig. In the case of the 3>4theory we can define a conserved current k^ by (22. .6 The minima of V in $4theory at x — ±oo. in view of the antisymmetry of eM^. n = 0 . V{4>) = 0 at different values of </>). (22. 22. » In the <fr4theory defined by Eq. (22. (22.34) jfc" = e^dv(l since d^k^ = 0. For a configuration to have finite energy. . . the classical field (bc must approach one minimum for x — .552 CHAPTER 22. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. 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. the potentials have degenerate minima (i.35) and it is seen that in this case q can be ± 1 or 0.1). Thus the classical energy or vacuum configuration is not unique.6. .
The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . ± l ± 2 . 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . A property is. 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. zero is not a meaningful topological number. termed topological. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. continuous values.A2)2 + 0 except for A2 = 1.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). . We can see this as follows.cos(27rnA)]. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . . In the case of the sineGordon theory the minima of the potential (22. . For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22.11) are given by ( so that for x — ±oo: > — J. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense.7T (22.38) 4 V(\<t>c) >m ~2[l . Tfl \ (22. and we call this "topological stability.22.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. .36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. Again we see that E(\<f>c) is finite only for A = ± 1 . First. n = 0 . Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. in general.37) for the onedimensional case here. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. one may ask how topology comes into the picture.
425 . Eqs. (22.41) dxy/2V{$)4>'{x). the energy is minimized.31) to (22. Eq. pp. The inequality is saturated. (22. (d) the topological quantum number iV arises in a completely classical context.554 CHAPTER 22.42) Thus the energy has a lower bound given by the right hand side of this inequality. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. W.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents).^ = N.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). 22.43) **See e.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (22. We observe (a) the topological current k^ of Eq.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. 00 (22.9 the foregoing arguments in a higher dimensional context. Thus with one space dimension higher there. Classical Field Configurations involves the difference An = rioo — n. H.e. (22. MiillerKirsten [215]. J. (For a broader view we consider briefly in Sec. (c) the time component k° is a spatial divergence.428. its integral is nonzero only on account of nontrivial boundary conditions.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. .g.40) (22.13b) or ^ ' ( * ) ] 2 = V(</>). From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. 00 (22. if oo / dxy/2V{<l>)<f>'{x). the factor A = exp[ix(#)] varies over a complete circle). 22.17)) f 00 > 0. i.
555 (22. We discuss this in the following Example.21). (22.40)) this occurs when 4>'(x) T V2V(4>) = 0.44) This first order equation which saturates the inequality (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— . (22. 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.^ Clearly it solves the second order equation or (22.19).22. m 9 V V / and with x = g4>/m this becomes P IA\ CTnin(0) = = /^ J /^ V ™3 f2W A I A • 2X n—7T I ax\/2ll — cos x) = n—7T / ax* Asm — Z 3Jo 2ir r 9 Jo V 2 2m 2 cos — n — n^[2(l)+2] 2 ~ m3 8m 3 = n—rrin agreement with Eq. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. Here (V(c/)) > 0. (22. . 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. Bogomol'nyi [27].39). In the sineGordon theory we can consider.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. for instance. n E . Eq.n > 1. B.
z)plane and only such that their elastic strings remain taut the position of any one of the pendulums is determined completely by an angle <p(xn. Chapter XI.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .46) **See e. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. Initially the pendulums are suspended freely in the gravitational field. Construct an expression for the energy of such a classical system and its Lagrangian.g. ka —> K. 22. rotate) only in the (y. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. } n= — oo roc —> / L J .7 The pendulum analogy.556 CHAPTER 22.t) <j>(xn. z)plane and only such that their elastic strings remain taut.2 V dt + k[(f>(xn+i. all being identical. H. with mass m and connected with short strings to neighbouring bobs. Classical Field Configurations Example 22. so that oo > u. from the yaxis as shown in Fig. rotate) only in the (y. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. Assuming that the pendulums can move (i.e. Solution: Consider a string along the xaxis as described. 22. .1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.t) measured e. t)).e. Fig.cos</>(x„.g.t)\2 +mgr{\ . Assume that the pendulums can move (i. The kinetic energy T of the bob at position xn at time t is (recall " m r 2 r / 2 " ) 1 —mr 2(d<j>{xn.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. We can now write down an expression for the energy of such a system of (say) n pendulums. Goldstein [114].7. At equidistant points xn = na along this string we attach strings with pendulum bobs. xn+\ — xn = a) 00 a —> dx.t) 4>{xn.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i.
13a) and (22.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22.t): {w With the ansatz r)(x.25) obtained earlier.t) = <l>c(x) + r}(x. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. wk would be imaginary and so the factor exp(—iwkt) in rj(x. and so would invalidate the procedure which assumes that rj(x. (22. the topological quantum number. If some w\ were negative. 22. (22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. t) of Eq.22. (22.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.47) into (22. 22.t) = ° (22'48) (22. ' Equating coefficients of exp(—iw k t). t) is the fluctuation (field). is therefore also called "winding numbed.50) This equation is identical with Eq. Hence we set ${x.t) is a small .7 linking the pendulum bobs). (22.47) where n(x.49) exp(iwkt)ipk(x). Eq.13b) we have for small fluctuations rj(x. (22. The equation is called stability equation or equation of small fluctuations. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. (22.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt).25) which we obtained as the condition of classical stability for eigenvalues w\ > 0.49) would imply an exponential growth in the future t > 0 or in the past t < 0. Inserting (22.t). This number. of the continuous curve in Fig.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs.e. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). which represents an expansion in terms of normal modes. We can understand what stability means in the present case. It is instructive to obtain this equation by yet another method.
cos46» = 2cos2(2(9) .558 CHAPTER 22. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. . Inserting (j)c of (22.A \ Inserting <f)c of Eq.e. If that is necessary.2 m 2 + 6m 2 tanh 2 m(x . approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. and need 9 = tan" 1 j e±m^Xo) 1. the zero mode. For x — ±oo the quan> tity acting as "potential". V"(<b) = 2m2 + 6g2(f>2. This means that the stability is not universal but only local in a certain sense.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22.20). Thus for stability w\ must not be negative. i. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. we obtain V"(<t>c) .e. Classical Field Configurations fluctuation. 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. V"((/)c(x)). we shall see later that the fluctuations n(x.1. In fact. In fact.18).52) tan 6 := e±m^xo). 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£^)*.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . i.50) has the general form of a timeindependent onedimensional Schrodinger equation. (22. The equation of small fluctuations (22.
and as a consequence a limiting form of the small fluctuation equation of all basic potentials.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. We can investigate the spectrum as follows. ip+. ^+(0) d 0. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.55) The differential operator is even in z.53) = m(x — (22.e.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .2 (22. (22. which satisfy the boundary conditions V>(o) = o. i.* For such a potential the spectrum can be b o t h discrete and continuous. — XQ)] cosh[±m(x — XQ)]: 1 .22. even and odd solutions.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. A  *(* + !) cosh 2 z tp(z) = 0. Consider the equation fd2 dz2 .ijj^ respectively. .56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation. We can therefore construct solutions of definite parity. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable. Those of the other case are complex and periodic at infinity.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 . (22.
Classical Field Configurations We now set (with ip either ip+ or ip) tp(z) = (cosh z)~lx(z) and change to the variable £ = sinh2 z. A ^ An = .60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.Z)2.58) = (1 + sinh2 z)~l/2x(z). i.4 n ( n .59) Setting (22.57) Then the equation for x{z) 1S (22.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22.60) terminate after a finite number of terms.( 2 n . i.e. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . (22. and in the odd case n+\)U+\l)+\(\ i.I) . A * An = .e.Z ) 2 . Since for £ — oo: * the function ^>+ is normalizable provided Z > n.61) Discrete eigenvalues are obtained when the series (22.e.l2 = .e. 2n < Z. .( 2 n + l . when in the even case n ( n .Z ) + ^(A + Z2) = 0.560 CHAPTER 22. (22. + l2) = 0. i.
We can now derive the spectra of the stability equation for the potentials (22. (22. N = 0.1.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz. We can see the continuous spectrum of Eq.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.e.54) by setting X = k2 Then 1 .. 3m 2 .2Z = . From (22.22.j < I .6 The Small Fluctuation Equation 561 and the function ip. (22. w2 = Am2.62) we see that the equation possesses discrete eigenvalues w —2 . (22.u2)—^V 2 dio and ip(z) = elkzV{to). m N<2. the continuum is w > Am2. i. (22. for N = 0.51) into (22.2.( 2 .11).e.65a) _dz2 cosh 2 z l=2. (a) ^theory: the equation Inserting (22.. d w = tanhz. (22.62) where even (odd) iV are associated with even (odd) eigenfunctions. °2 i.63) we see that the continuum starts at A= 2 2^ = 0. 2 ..1: w2 = 0.is normalizable provided 2(n + .N<1.\=Kl2 From (22. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .Ny.50) we obtain with z = m(x — xo) tp(z) = 0. m i..e.10) and (22.
It is now particularly interesting to look at their wave functions. The zero mode is depicted in Fig.. . Classical Field Configurations (b) SineGordon theory: Inserting (22. .8 The zero mode as typical ground state.8 and we see it has the shape of a typical ground state wave function. We know that 4>c is a monotonic function. (22. i.27)) . Thus either spectrum contains the expected eigenvalue zero. In the $ 4 theory we saw that (cf. the zero modes.53) into (22. 22. . also be deduced from d(j)c/dx.562 CHAPTER 22. for N = 0: to2 = 0 and the continuum starts at w2 = m2. so that its derivative is nowhere zero (except at ±oo). 22. m d . d .2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. N<1.50) we obtain with z = m(x — XQ) the equation (22. This property can. and — as expected — this eigenfunction is even.66) We see that this is a nonvanishing function for any finite value of x. %(x) Fig. . of course. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue.65b) dz2 cosh^J. Eq. .=1A=^_. Thus 4>'c(x) has no node.. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22.e.
.t a n h x .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ). Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1. 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 . . Finally we consider the normalization of the zero modes. (22. without quantum corrections). ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. — XQ) OC Hence the conclusion is similar to that in the previous case.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 . i. (22.70) .6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . . d .22. (22. .e.e.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i.oc (22.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a.z 0 )]oo = 8 . 1 o rrr tanh x — .= E(<f>c).
71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . J As a first example we consider the expression (22. P(</>) = f ddxV{(f)) > 0. i. H. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. Derrick [68].g.2d. (2. P. to underline the fundamentally different nature of topological and nontopological finite energy configurations.e. however. f G .564 CHAPTER 22. *See e. (22. ddx {V4>f + v{4>) (22. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. Olive [113]. They may help. .* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. Affleck [5] who uses in his Eq.72) with T{$) = I' ddx]{V(j))2 > 0. we consider E(<P) which we also write as (22. One can have different types of scale transformations — see e. for if such configurations exist in a realistic theory with three spatial dimensions. ^For basic aspects of scale transformations see e.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . A.41) for the energy of the classical configuration but now with respect to a ddimensional position space.g. Classical Field Configurations 22.g.> 4>a{x) = 4>(ax). *The considerations of this section extend in part beyond quantum mechanics and into field theory. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). I. it is important to understand their physical implications. H. Kastrup [145]. I. This question has important consequences.4) <j>(x) —• a<f>(ax). Goddard and D.T(4>).
Consider first the Lagrangian density £[&M = \{d^)(d^) .73) we see that this implies (2 . Next we consider more complicated cases. to be finite the field <P ~^ 4>oo £ Mo for x — oo. (22.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}..79) In order to emphasize that (f> — (p^ in any direction.xd). we set 0 dE(<pa) da J a = i (22. i. the second derivative is positive) only for d = 1.80) .e. d = 1 minimizes the energy.\ eMo(22. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension.78) For the energy of a static configuration.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.74) From (22..e. 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.. a=l (22.x2. since d2E(<pa) da2 2(2d)T(<f>).. (22.d)T(<f>) = dP{4>).76) we see that E(<f>) can be minimized (i. i.75) Since both T and P are > 0 and d is integral. we write • = lim J r .V(<P) (22.22. we see that the stationarization of E with respect to a is possible only for d = 1 as in the examples we have been considering so far. (22. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>). In fact. x = (x1. (22.71). It is convenient to discuss this limit in terms * of the directions of unit vectors from the origin to the (d — l)dimensional surface in ddimensional Euclidean space defined as the unit sphere S*1 = {xx 2 = 1}.
e. —00.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . Table 22. i. in fact. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. i. x = 00. —00 or equivalently S = + 1 . (ptx = <f>oo. Then each point at infinity is characterized by a different value of the polar angle 9. and.1. — 1). S1.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. i. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . —1. that we have a set of disconnected classical vacua (</>oo><^oo. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. 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 . as is the case for the constant solutions. is a connected region.81) and (since d = 1) S = ± 1 .and sineGordon theories we assumed $ to be real. We see. These observations are summarized in Table 22. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively). If we go to two spatial dimensions (d = 2).38). one set can be mapped into the other. Thus in this case S^1 is a disconnected set consisting of two points. ( with — = lim <pc(x) 9J *>±oo TYl \ (22.e. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. of • course. .566 CHAPTER 22. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. therefore. are the same.e.
This can be achieved e.e. the one point S = + 1 . map into.82) Now Mo := \ 4>(oc) ^24>2 = a2\. i. that if the problem is to be topologically nontrivial. We see. 9) would depend only on r = x and hence the problem would be effectively onedimensional. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}.g.e. VMx)) with (0(oo) = linvxx) (p(r. if <p is real. phase transformations 6Q ^> 9Q + 59. Similarly. if cj> were real and position space twodimensional. * n reach any other <^(oo) by simple phase shifts.7 Existence of Finite Energy Solutions Suppose. If position space were onedimensional. i. for instance. we have w e ca {Mx)}> and 1 3 1.• M0. the field must map position space into field space at infinity.e.e. Thus. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}.83) . (22. cj)(r.e. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$.2.3 (^real) v{4>) = / (22. i. On the other hand. assuming invariance of C under the transformations of the group SO(3) in internal space. i. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. This will be "nontrivial" in the above sense only if 0oo is independent of 9. Thus. . $(r) : Sl . 00.1 . by allowing <fi to have three components in some "internal" (isospin) space. to have topologically distinct classical configurations. i. i. —00 or S = 1 . Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.22.e. therefore. we would have only two points at infinity.
a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. A) + T{A) + P{<j>).e. A) = T(4>. it is effectively a surface (and so a continuous set) S2. L. Glashow [108]. $ — $exp(ia) in a U(l) symmetric theory. 11 H.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". $ — $ + 27 > • T in sineGordon theory. $ — > > $ exp{i Xlo=i Pa. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). sphere. i. + i ( D " f t W ) a . Olesen [220].Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. circle. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. the subsequent discussion requires only the general form of the Lagrangians.„ where G r = 9 M £ .</>] = ~G^Gai.e. The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du.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) £(</>..Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. through each point at infinity). the appropriate rotation.e. invariance under replacements $ — —$ in $ 4 theory. Example 22.V{4>a]. a = 1. The symmetry of the potential (and Lagrangian) — i. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. going once around S%~1 (i.2. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. If.e.568 CHAPTER 22. . Georgi and S.. Nielsen and P. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. i. the winding number is n. .a M J . Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. so that details of these models are irrelevant. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. we cover <S^_1 n times. B. % .3.
89) P{4>) are positive semidefinite and not all zero. some contribution must be .87a) C?M„(x) — \2G^{\x). 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . (i)^ 2d T Considering the contributions of (22. A A ) = A2dT(^. = 0. and so B ( ^ A . (22. A).84) in this way one finds that the quantities involved transform T ( ^ .Ax) dX Since T((f>.84) is positive semidefinite and so must be finite by itself. l j ddxGijGai] > 0. Thus we set dE(<fix.22. negative. (2 .d T ( ^ .dP(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). >• A„ x ( x ) = AAM(Ax). Solution: Each contribution in (22. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also. A) + (4 . ^ A ) = A 2 . i.7 Existence of Finite Energy Solutions where T{<j>.T{A).A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0.86) (22.d)T(A) .d T ( A ) + AdP(</>). <0 >• <£A(X) = <KAx). Mx') = E with a scale transformation x — x' = Ax.<^[x) transforms like a vector. (22. J 4 ) + A 4 . P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. i.84) and investigate the existence of finite energy classical field configurations. we have {^/»)A^ (22. (22.e. A.i T(AX) = X4~dT(A). > Since T>IJ.d)T(4>. (22.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). (22.e.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. We now consider scale transformations with A £ R.. T(A) = f ddxV(4>) > 0.
Landau [111]. . 22. W. (22. J.89) the contribution (4 — d)T(4>).89). H. (e) For d > 5: In this case it is not possible to stabilize Eq. D. ft V . H.90) and we would have to add to Eq. (22. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. L. Ginzburg and L. Skyrme [253]. In the following we consider a related theory. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). (c) For d = 3: All three terms must be present as in the GeorgiGlashow model.89). a > 0. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. T^{<t>x) = \4dT(</>).** e. is temperature dependent and changes sign at the critical temperature Tc.92) The expression ^ ( x ) ! is known as the "order parameter. A) or P{4>) for d = 3. Tchrakian and H. (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). of course. This case is. R. (22. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example.570 CHAPTER 22. One can thus define a scalar field or wave function ty(x.t) such that  ^  2 describes the density of Cooper pairs. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. 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.e. a = a(T) = a T —T c . exemplified by soliton and sine—Gordon theories. We then have a situation of the potential like that described in the context **T. 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. MiillerKirsten [268]. Classical Field Configurations (a) Thus for d = 1: P(4>) must be present but T[A] and so the vector field would not be required to satisfy Eq. (22.g. Thus such a contribution could counterbalance T{4>.
assuming there is no superconducting material in this domain). 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 < 0.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. but for T < Tc it has a double well shape. Sec. Example 22. from x < 0) means — as is familiar from electrodynamics — replacing V by V .2) £(^. Then the static energy density becomes (cf.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. i.^ A n (this is called "minimal coupling of the electromagnetic field^ with vector potential A to the Cooper pair field ip in a nonrelativistic treatment) where g — 2e~ is the charge of the Cooper pair.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . Clearly the length (see Eq. ^(00) = a (22.22. 22.94) to check the dimension) 1 = I—a. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. i. (22. (22.2): For T > Tc the overall potential has a single well (minimum). setting S£.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf.e.(tp) = 0. 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.e. The parameter 7 serves to put the minima of the double well at V = 0.93) J' (22. Variation as before. is a measure of the distance from the surface of the superconductor to where .
e. yJ—a/[3. 2mi Setting < / > \P\e vp B = ViA. h [^DV'V'DV].572 no superconductor CHAPTER 22. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ). (22. p. i. as indicated in Fig. i.e.g.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B . 22. The magnetic field on the other hand. V x M = j s . 450. (22.9.98) where M is the magnetization resulting from atomic currents. J. In the Maxwell equation See e. The density of the Cooper or super current is given by the typical expression of a current. (22.e. H. Classical Field Configurations superconductor (x>0) Fig. . exhibits a completely different behaviour — in fact.9 Behaviour of ip and B in the superconductor. the order parameter \ip\ attains (approximately) its asymptotic value.M. 22.96) ipH. MiillerKirsten [215]. W. i. We can see this as follows.
100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter.e.22.L B = O. V x B = [i0V x M = /lois. Also dD/dt = 0 in a static situation like the one here.M i. this becomes A B = no V x j 5 .e. In the onedimensional case we have B{x) = B(0)ex/XL. m xi A B . K i ™*.. i. Hence 0 = V x H = V x ( — B .99) Inserting (22.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. Using the relation "curl curl = grad div . those of the socalled type I and those of type II. Approximating p by its equilibrium value. These vortices therefore carry magnetic flux which is necessarily quantized . and V x (V x B) = / J 0 V x j s . 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. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 .8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. 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). (22. (22. There are two types of superconductors.div grad" and V • B = 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 is zero in the present case.
Abrikosov [2]. Finiteness of E requires \<t>\ = \<fioo\ on this circle.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22.574 CHAPTER 22.e.9). Thus the asymptotic field </>(r. in other words. 'See the remark in H. i. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. de Vega and F. (22.102) In the onedimensional soliton case of Eqs. Schaposnik [69] after their equation (3. of two points on a line. A. i. De Gennes [67]. In the two models of Example 22. See also P. 22.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. In the NielsenOlesen model defined in Example 22. where exp(ix(0)) is an arbitrary phase factor. we must have that lim V ( H ) = 0.9) ^e^ '>\<poo\. The existence of these vortices with quantized magnetic flux was predicted by Abrikosov* and are therefore frequently referred to as Abrikosov vortices (the idea was originally rejected by Landau). *A. 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. Classical Field Configurations (see below).e. the space consists of two diametrically opposite points on the circle. A.37).36). In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). G.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. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. in fact a periodic function of the polar angle with period 27r. (22.101) (p(r. . J. The existence of these vortices has been confirmed experimentally.. We can think of (e.g) twodimensional space as being bounded by a circle at r = oo. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. (22.
States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig.10 States of winding number 0.e. Then a continuous function of (j> on [0. The map A : S1 > S1 defined on S1 C R2 by (22.9 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". Sl := {zeC\\z\ = 1}. The 1sphere is the unit circle in the space of all complex numbers. The two theories we considered above in Example 22. Misner [96]. commutative).1). 0 < 0 < 2TT. Finkelstein and C. W.102) is called the exponential map of S1 onto S1. 2n] remains continuous if the function assumes the same value at the two endpoints.1. Fig. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism.e. Naturally an integer cannot change continuously. Suppose <j>. (22. .10. Thus smooth deformations *See D. defines the orientation of this line element. X is called the domain of / and y its range. We may define this as « = ^[X(0 = 2 T T ) .* 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. 22. 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. abelian (i. 22. (22.22. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure.X ( 0 = O)].118)).103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. topological group with the usual multiplication as group operation. In mathematical language S1 is therefore a compact.
one hole. i. The gauge field required in the asymptotic limit (22.105) is described as "pure gauge" since it is determined entirely by the phase x(#). the winding number must be a constant of the motion. elements in the same class have the same topological characteristic.104) to be finite we must demand that g . In other .) remain the same under homeomorphic transformations and are therefore called topological invariants. The different kinds of connectivity of these topological spaces (e. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.576 CHAPTER 22. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. we can write 1 d h J ee rde w At h r^oo . since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. We observe that since time evolution is continuous..104) so that d2x hJ Since A has direction e#.e. no hole..i'ldx(0) 9A Thus for (22. The winding numbers n are such topological invariants. r^oo 1 dx(9) n r o0 (22. For that reason n is called a "topological invariant?.e. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). i. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e.e.g.. connectivity. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). if it has one twist as above or a hole. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. (22. i.g.105) the corrections falling off faster than 1/r. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. finite. Thus two homeomorphic spaces have equivalent topological structures. we required that lim r—>oo w o.
109) A W(0) = e^ W. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).22. A field configuration with one loop (or twist). 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". the vector potential A. if A = V x . the higher order contributions in (22. Ffiu. (22. into S i . However.107) The fields <f>(r. it is determined by a gradient (l/r)d/dd. cannot be deformed continuously into one with two loops (n = 2). 9){^ and hence the functions (22.105) insure that B ^ O . (22. cannot be pure gauge everywhere.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22.106) J g j r d d g Thus the magnetic flux is quantized. we can consider transformations which describe continuous transformations of a configuration with fixed n. It follows that the field Fij.l]. we have B = V x A = curlgradx = 0. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.108) map the spatial circle S^. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . (22. going once around S^. the gauge field. Therefore. i. Of course. with n = 1. for B / 0 .e. An example is the mapping H(t.e. the group space of (the phase transformations of) U(l) continuously. However.<b rd9—4^ = n.9 Introduction to Homotopy Classes words. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. is also zero at r = oo.9):=t(f>£\9) + (lt)(f)W(9) with te[0.e. i. i.110) . we can go n times around S^. the number of flux quanta 2irh/g being the winding number n.
We can consider (22. 8) = cfinl\9).T. i.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. these two maps (fin from S^.lQSl^S^ with H(0.110) could be written down for configurations with different winding numbers. Dittrich and M. The map H(t. H. S%. i. Thomas [270]. i. i. (22. Two curves which can be related to each other in this way. Hu [134].g. it is clear that this is also a homotopy. A superposition of classical configurations with different winding numbers does not minimize the energy.e.e. n' ^ n. e.9) = <fiW(9) and H(l. Solution: As stated. Classical Field Configurations which is continuous in the parameter t and is such that H(0. Example 22. H(1. — S\ > (22. (fin (9) in the above example.9) = ^°\0). the phase factors of (f)(9).e.e. the unit interval I = [0.111) as a map from the space described by 9. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. The relation (22. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).8) itself must be a configuration with definite winding number. 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. Reuter [77].e. Steenrod [261] and S. like <pn (0). time cannot make a field jump from one homotopy class (see below) to another. (b) symmetry and (c) transitivity.9) is called a homotopy (emphasis on the second syllable).0) = $\6). f T h e standard texts on homotopy theory are N. H. A brief introduction can be found in the very readable book of W.578 CHAPTER 22.1]. i. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). and the space of t.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).111) In principle an expression like (22. to the internal group space of £7(1). 9) is an equivalence relation. (22. Since the passage of time is continuous. But this type of combination is excluded sind H(t. we have to show that the map satisfies the three properties of (a) identity. In fact. (b) symmetry and (c) transitivity. n.113) . We verify these in the following example.
i<t<l. since and tf (1.e) tf2(2il. negative or zero). (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. 9) = <f>W (0). i? is a homotopy connecting </>(°) with <^>(2'. a class being a set of equivalent maps. 0) = <£{2). 0) = 4>W (0). is such that H(0.9) H(0. i. <^(2). 9) = H(0. these groups are then called homotopy groups (cf. \ M)\ H!(2t. ' . Moreover. On the other hand H(0. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. which therefore demonstrates that the homotopy relationship is transitive.* 22.e.e. so that H(t. Hence the set of all such maps breaks up into equivalence classes or homotopy classes.9) = .t. Returning to the maps \(x) : Si ^ S1^ (i. as they are called. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. we see that of the set of these maps some are equivalent and others are not.0) for for 0<t<. 0). Then H(t. 9) = H(l. is isomorphic to) the set of integers Z (positive. Hence <f>(0^ ~ <p 2 ). then </>W ~ <t>{0)• We define an H(t. U(l)). 0). 9) := H(l .22. (c) We can demonstrate the property of transitivity as follows.ffi(O. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. below). 0) = ^ (0) and 6(1. Homotopy classes are sometimes also called Chern—Pontryagin classes.e. 9) connects <pn (0) with itself continuously. In some cases the set of homotopy classes possesses group structure (satisfying the group properties).9) = <f>(0\6) and H(l.10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the onedimensional sphere S1. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. . . 0) = tf2(l. being the continuous map connecting </>(°) with <j>(1'>. at t = 1/2 both i?i and H 2 give c/^1'.0).10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. <f>W . We now define the function H ( f 0 ff .0) = </>(0) is continuous there. 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. (b) Next we check that if 0<°) ~ <p^.
Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. 52.1] 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. Rajaraman [235].11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. T < 9 < 2TT. How can we see this? We first sketch the idea. In the diagrams^ of Fig. S 2 the surface of a sphere in a threedimensional Euclidean space. n=0: t=1: t=1/2: i. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig.115) Varying t € [0. Fig. In order to really appreciate how topology comes into the picture here one should understand (e. but TTI[S2] = 0: S 1 is a circle. which is to be mapped into another circle Si indicated by dashed lines. .g.) why 7i"i[S2] = 0. p.11 The trivial n = 0 map allowing shrinkage to a point.580 CHAPTER 22. 22. Appropriate maps.11 the circle drawn with the continuous line is the circle S%. Sketch of idea demonstrating that nilS1] = Z. T (22.e. 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. to make the result plausible. and then introduce some of the related mathematics. 22. 22.
. is invariant under infinitesimal (continuous) deformations. 22.s. Coleman [55]. of (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. In the cases n = 1.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. 22.s. 0 < X(P) < 2vr. One can prove that the expression (22.117) in ±.e. (22.116) In fact from this we obtain (see below) 2  n and this means r.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig.12. For winding number n we have the mapping (i.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). (22.h.22. to the case of winding number zero. We make another observation at this point. i.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. i. the transformations^ (the result may be looked at as 8n of expression (22.e. 2 the maps are wound around the dashed circle. The maps for these cases are illustrated in Fig. d (22.e.117).h.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.117) 2^ d6X(x)\1(x). and hence the winding number.
Classical Field Configurations Under this transformation n changes by 5n Now. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0. . Sec. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations.e. 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.e. Steenrod [261]. Thus there is only the trivial map into a point.e. the cyclic group with two elements (since 50(3) is doubly connected). il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). 1 1 See N. whereas 7ri(50(3)) = Z2 (i.8. In fact. an integer modulo 2) i. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. there is a theorem which says thatH TTg(Sn) = 0 for q < n. only one homotopy class. i. 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)=0.582 CHAPTER 22. the identity class (the group space S2 is simply connected. 15.
Chapter 23 P a t h Integrals and Instantons 23. 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. different from literature.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. 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 such a way that the factors appearing in the result are exhibited in a transparent way.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. and in particular the exponentially small level splittings of the double well potential. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. We consider such a theory now with action S((p) depending on the real scalar field 583 . our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration. In the following we employ the path integral in order to derive the same result for the ground state splitting (only that in this chapter!). finite Euclidean time) classical configurations (which are nontopological). Since we have not yet introduced periodic (i. 23.e. An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential.
in particular E. i.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. "For comparison with literature.2.1. ! " ^ j = ^ 2 .1 The double well'potential.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . 2 2 4 ™ = ~. Chapter 6) 1 2 .e. and ^2(0) : m* . note that here the mass of the equivalent classical particle is taken to be mo = 1. n = 0. (23. Fig.v{4>) <?VV m4 __ 1 .584 <)>{t). Gildener and A.2) Fig. Path Integrals and Instantons S(<f>) / dt \i>2 .1)) and hence E0 = mh/y/2.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf.* /•tf CHAPTER 23. 23. The Hamiltonian of the simple harmonic oscillator (cf. 23. (23. Patrascioiu [110] and Chapter 18. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.
</. 23. as we did in Chapter 18 — but as we saw. must have definite parity as in a physical > situation.x) = x2 V(x). The states of the system here must be even or odd under x — ±x.g.2 The wave functions of the lowest states. this alone does not yield the level splitting. e. we have the Lagrangian L(x. We can develop a perturbation theory around either of the minima. This tunneling affects the eigenvalues._(0) = 0. and in the above 4> corresponds to x. W(0) = 0.2 — and we can deduce from this the boundary conditions.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). i.23. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. 23.e. but there is tunneling between the two minima if the central hump is not infinitely high. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. Fig.In mechanics of a particle of mass 1 with ~ position coordinate x(t). These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator .
e. Euclidean time T = it. that given by Eq. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23. dr We observe that Eq. i.7) so that (23.2. is finite if we go to imaginary. classically forbidden domains then become "classically" accessible. (23. Equation (23.586 CHAPTER 23.4) J <bi =—(An rTf (ITLE = I JT. dr m+™ (23.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).U) = J'D{<f>}exp[iS(<l>)/h] (23. i.0o. (23. Then (0o. SE. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.tf\ .<h. (23. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.e. i.7.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.9) .6) d*l dr 2 .e.2 V'(<l>) = . The transition amplitude for this.8) V(4>) + const.  . In the semiclassical path integral procedure that we want to use here. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55.e. For small h we can attempt a stationary phase method.£'(0) = 0. r . dLE = 0.5) and iS = dLE d dr[d(d(f)/dT)} i. But quantum mechanically it can tunnel through the hump from one well to the other. (23.3).2 m 2 ( l m z (23. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o.3) for tf — oo. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.
from fa — —rn/g. i. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".e. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). is depicted in Fig. we see that the equation is the same as the classical equation for a static soliton (i. Fig.e. i. and hence kinetic energy gained is equal to potential energy lost. 23. this means the total energy of the particle in Euclidean motion is zero. Thus we know the solution from there which is the kink solution (22. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. 23. in the present case (T — To) = m tanh m(r — TO). — (23.3.3 The instanton path.23.2 Instantons and AntiInstantons 587 In fact. the classical path. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest.18). . Its trajectory.10) where TO is an integration constant.e. 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.
TJ T fro.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.r 0 ) ' (23.13) into Eq. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. (23. Inserting the result (23.' . Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .10) 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. The configuration (23.]. 23.Tj) oc exp 4 m3 m^2 (23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. thus indicating the nonperturbative nature of the expression.5).* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.4.14) But this is not yet enough. Path Integrals and Instantons The Euclidean action of the instanton solution (23. Using Eq.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. (23.7) admits another configuration which is the negative of (23. Tf = oo.11) For this quantity to be finite for TJ = —oo.588 CHAPTER 23.8) we obtain T f SE dr[2V((j)c) + const. (23.4). The classical equation (23.13).^ . the constant must be zero.13). This antiinstanton configuration rises monotonically from right to left as shown in Fig.7"/! ). but with the same Euclidean action (23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.10) is given by (23.
g.4 The antiinstanton.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig. #(r) = </>C(T . 23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.5.5. Questions which arise at this point are: Is the nonmonotonic configuration (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 (23.15b) is not exactly a solution.TO far apart and T'Q 3> TQ.g. 23.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . 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. The sum (23. as we can see e. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.23. (23. b u t if TQ.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). and an antiinstanton localized at (say) TQ.15b) since we require configurations from vacuum to vacuum.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity. / > Looking at the instanton (23.15b) a solution of the classical equation? Is it a topological configuration — if so. 23. 0c(r) = &(T).15a) we have a simple superposition. Clearly we are here interested in configurations of type (23.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. from the sketch or from the variation of SE. e.15b) In the case of (23. . (23.r0).TO) + ^ c ( r . in the case of (23.15b) has the shape shown in Fig.
2 m ( T o . Ignoring exponentially small contributions the Euclidean action of the configuration (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. at oo the variation is <<> = 0). it also does not exactly stationarize the Euclidean action. 5^ In order to answer the second question we observe that (23.5 Superposition of an instanton and a widely separated antiinstanton.e x p { . since (23. Path Integrals and Instantons '. 23.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. 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 . In fact.15a) is not exactly a classical solution. vanishes as T\ —> ±oo (of course.16) . Finally.15a) can be argued to be zero.CHAPTER 23. g cosh m(r — TQ) (23. of course. T\] with LECIT = 0.15a) is a configuration which starts from 4> = —i^/g and ends there.T 0 ) } ] . we see that in the domain [—oo. #(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 .
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).
23.64) Inserting the explicit expression for the kernel given in Eq.63b)..lnJV(T)}] N(T) N(T) (23. i.64).64) D (23. so that (with T *T) 1/2 •Dz N(T) (23. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . T. whereas the determinant arises only once in the original determinant Dn. Thus D„ = l + 5 ^ K ( r i . (23. = 1 52 K(n. n )d? (23. p. where a f 1 See R.T2) I exp I K(n.n) K(Tj.7 Y. In the limit S —* 0 . Thus the factor a inserted above has to be taken as 1/2. Thus in our case one of the kernels is zero and we have to decide what we do when. we obtain exactly the nonvanishing terms on the right of Eq. the kernel K(r.TI) 0 K(T2.4 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.e. . Felsager [91]. Martin [43]. r i ) + . X(TI.Ti) i=l : exp \ £ K(T. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. 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.n) z=l x (23.n) i=l N lim exp £*(T. Cameron and W. the value of the determinant is half (or arithmetic mean value) along the diagonal.exp • /_:T JV(T')" dr' = e x p [ .N{T) Finally we add the following remarks concerning the discretization (23.n) o K(TJ. 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.j. (23. H.63a). [44]. We observe that if we discretize Eq.68) D above.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i.a { l n i V ( r ) . r ' ) is to be evaluated at r = T ' . 2! . 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.TI) K(T2.T1)dT1 + J dn J dr2.67) i=l v ) = lim exp V In ( 1 + ' L K(T. 192..T')dT' (23. .64) as above and consider the determinant of the expression.. ^This is the method explained by B. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.. (23. are summed over all values from 1 to n. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). as in Eq. (23.66) •Dr) .65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.
71) Hence (with h = 1) l = j .69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. Path Integrals and Instantons Example 23.Z ( T ).Z i . (23. In every integration / dzi the contribution €7. (23. n J=. N(T') . (21.( .608 CHAPTER 23. Eq.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. (23.J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ .l + £7t) «i (23.2o) 2 rc^O.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. Evaluating Ii.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this. Solution: From Eq. 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). so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. With partial integration (in the second step) we obtain (since (cf.59)) + f(T)] = — f dz(T) f°° dae^W+f™. Eq.59) the endpoint conditions »)(±T) = 0 and evaluate the integral. T + N(T) / T' = T JT dr' N(T') . I2 and using induction one obtains (cf.73) . represents a constant translation. (23. and hence may be ignored. f(T) = N{T) J^ dr'^^z(r') (23. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T).e—*0 In. Briefly. .0.(T) + /_ dr — . I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23.N(T) .70) and this now requires also integration over the endpoint coordinate z(T). (23.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l .^ . In = dzi dz2 • • • dzn e x p .
Faddeev and V.12).£Q) — aTo]dTod£o. (21. (3.4 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. i. . N. 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. j / dTT]n(T)r]m(l r . / — / / We started off with V(T) = ^CnVniT).11).4. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. Eq. Popov [89]. 1/2 T W 2 (r)J (23..e. J co § dr Recall that / f ^ doce^ = yfVfp. ^E. £n = ~2 dTT](T)7]n(l (23. This result verifies Eq. (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. (23. a = const.10) and (21.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE.59) and agrees with that in the literature.74) where we reinserted h. Ignoring the remaining metric factor with the reasoning explained > between Eqs. / d£o <9r0 5[F(TQ.^ 23. Patrascioiu [110].23. Gildener and A. 11 L. D." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function.
76) we set 4>c{r) = ^CnVnir). . Thus. at r = ro. / dTm{r)n{T). (23.83) **The collective coordinate in the present context is also called "instanton time".e. In analogy to (23.610 this means that CHAPTER 23. Then cn = ~2 / dr(pc(T)rjn(T). i. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).81) the rdependence is integrated out and hence the integral does not provide the function F(TO. e. (23.77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration.78) <t>ij) = Y.J drm(r)Mr + T0) (23. We now impose (i.81) 2 / dr»7o(r)0c(r) = .82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. Path Integrals and Instantons _^ ^ w . (23. in the direction of the zero mode. The delta function 8[F(TQ. where L = in + cn.e. The number £o is the component of the fluctuation rj(r). We consider the following integral / ( r 0 ) := . i. i /*oo (23.^O) with dependence on ro which the replacement (23.75) enforces the collective coordinate ro to be given by F(ro.g. (23.75) requires. this constraint implies that we go to the subspace which is orthogonal to the zero mode.£Q) — aro] contained in Eq.^n{r).79) (23.80) i r or •i /*oo dr7?o(r)0(r) = 0. We establish the function ^(To>£o) with the following reasoning. (23. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . In the integral of Eq. demand validity of) the constraint £o = 0.e.£o)/a. (23.
91) " F o r more discussion of this point and related aspects see also B.84) (23. Felsager [91]. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.(23. .87) l/M) (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).89) Here the factor App is called the FaddeevPopov determinant.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 ) .90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.y/SET0 = 0. We have AFP = /F(T0.85) PJ W i t h Eq.. (23.86) = Jdrrjoir^Mr F(TO.^) .88) ro0 Then in Eq. We evaluate this quantity as follows in the leading approximation.75) 1 = AFP f dT06[{F{rQ./ drri0(T)<pc(T) + PJ (23.£O) OTQ = = £ " . 158.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). (23.<t>c{r)} . .23.y/S^ro] (23. p. (23.Zo)y/SETQ We now define A FP '• =0 • (23. so t h a t with F(TQ. Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) .£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.
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. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). tangential to (J)C(T)) leaves this and hence the square of this quantity. i.0AFP8 4o H P . iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ).77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. (23.e. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i.612 CHAPTER 23. i. ~ There is a quick way to obtain the result (23. the "static energy" invariant. P the relative sign being of no significance.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. Inserting (23. i.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . if ^ + d£o = 0.93) where AT = TT — 00.e.e.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 (23.93). 7 Thus there is no fluctuation along the direction of the zero mode in the sum. Consider. =~ (23.91) into Eq. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.e.90). 770(7"). 0 d r (23. P J AT (23. 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. /(TO) ^/SE~T0} V~SET0 c d£./ dT0AFP. . if the coefficient of 770(70) vanishes. in d(j)c(To) + dr/fa). using the relation (23.
4. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.97) is a constant still to be evaluated. (23. (23.98) .60a) in not involving the zero mode contributions.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly. 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 (23.96) lim .52a) and replace in this J d£o by the expression (23. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.60a) for I0 and Eq. 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 determine I'Q by observing t h a t this differs from IQ of Eq. Since w\ is for finite T given by WQ of Eq.55b).93). (23.23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23.43)).96) does not yet allow a comparison with (23. (23. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.A T A F P i " o e x p T—>oo p where I'o n^O.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.4 Field 23. (23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo. We observe t h a t the result (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.
r).100) in such a way that it exhibits the crucial factors of which it is composed. i.TiMiy+'Mr «>.a n t i . ^See E. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23.100) We observe that the dependence on the normalization constant p cancels out.4. This means we are concerned with configurations like* 2n+l d(r . Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 . ~ rKo')e{Tx . 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. their Eq. 0c M. Patrascioiu [110].102) *For comparison with the parameters of E.32). they have to introduce the inverse of that factor on the far right of their expression (3. (3.e.55b)) w0 = v0e VQ 23. we drag the factor p2 along for ease of comparison with literature. Although we choose p2 = 1. (23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{. 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 .i n s t a n t o n contributions As already mentioned.101) 2T > oo (cf.614 CHAPTER 23.20). (23.5 I n s t a n t o n . *For the validity of the calculation the instantons or kinks have to be far apart.4>C exp h (23. Gildener and A. Gildener and A. In view of their normalization of the normal modes. .99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App.* Finally we rewrite the expression (23. the approximation is then known as the "dilute gas approximation".2 m T e x p 2 irgh ' T»oo (23.
<t>2)eSE{<t>2. T2 = T/3.<j>i) .fo) = Id2 St SE(<Pf. here (f>2. The factor ( .<Pi)2 + • • • . the total Feynman amplitude is given by (^/. 23.l ) i + 1 in the sum of Eq.T2\(f)l.106b) SE{<t>l.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .2 T / 3 .102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).r / 3 .Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23. are varied. (23.Tf\(p2.Tl}{4>l. r0(3) = 2T/3. (n = 0 implies the single instanton solution). r 0 ( 2 ) = 0.4 Field Fluctuations 615 for n = 1. (23.6.T2}((t)2.104) Considering the case n — 1 of a kink with one additional kinkantikink pair. Since these endpoint configurations. Consider (with T 0 = T.4>i.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.r/l&.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.r 0 ( 1 ) = . . ( 2 3 . Taking these configurations into account. .4>l)eSE{4>l.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. (23. (3) (23. this means we have to consider the amplitude (23.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . Proceeding as before we now consider first in analogy with Eq. 7 i = .Tl\(j)i. . T3 = T) d(f>l((pf. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f.<t>ff + h .H)2 + .4>I) ld2SE SE(c/>i.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.36) (/.23.(l>i) (23. 2 . 1 0 6 a ) and SE{(J>I.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.
Path Integrals and Instantons where the first expression in Eq.3 9 L 3m 2 and 9SE((/)) ± . dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .2J.106b) is to be substituted into the second relation of (23. ±m/g (23. each of the three intermediate kernels contained in Eq.108) where the last expression is obtained as follows.e. in the case of l2n+i therefore ry2n.e. (23. (23. that each applies only to a fraction of the overall Euclidean time interval of r = 2T. rn ^2 J2H 9 1m' ±2gcj). i. i.e. The allowed positions of the individual kinks have to be such that their order is maintained..4>f) d4>\ 4>s72 (23. (23. the dilute gas approximation.106a). On the assumption that the kinks of the instantons are widely separated. the .e. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.616 CHAPTER 23.12) we have JM~°°) i. i.105) can in other respects be treated in analogy with the single instanton amplitude. d2SE{4>) d(j)2 2m.(01 .107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.4>i) / d(f>2 e x p {<h .4>i) { 2 " 2d SE . It has to be remembered. With our remark after Eq.2d2SE ~c 4>i d(f)i e x p SE^f. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. each of which implies a multiplicative factor 7. of course.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations.
Am3 2m J rW ( r ( 0 and this followed by setting aj. we now have to integrate over three collective coordinates.m A .6 the collective (i) coordinate TQ has to be to the right of TQ . a 2 0. 23.r.112) We now compare this result with expression (23.35a) which contains AE in the argument of sinh and obtain for the level splitting A J resulting from consideration of the instanton plus any number of instantonantiinstanton pairs A*£ = 2n7AFP/0^=exp 2TT 1/2 4 m3 m V2 / m \ 1/2 4 v / g ^ exp ( 2TT irh _±rrv^ 2h ^With shifts r. Patrascioiu JT/3 ° ir< >2T/3 1 dr.110) Finally we have to recall from Eq. then sign reversals TQ i ) .(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23. and in the case depicted in Fig. In the case of (<^f.2 T / 3 this product becomes that given by E.r/^.93) has been applied. Gildener and A.V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.23.Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23. (23.^ .T/0j.101): (^/.e. and TQ to the right of TQ This implies t h a t in this example we have the product of integrals^ rp (3) (2) .4 Field Fluctuations 617 replacement (23.7 3 ^ — .(*) o ax = 2 T / 3 .a 3 [110].111) Finally we sum over any number of instantonantiinstanton pair configurations and obtain for the Feynman amplitude oo 1 oo 1 ( {JATAFPIO 2n+l I = lim V n=0 J 2 n + i = lim . T ] with ] T 3 = 1 A » T = A T = 2T. i. (3) ° JrW: 2T/3 (2T)3 .A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23.105) t h a t each of the three amplitudes contributes a factor e x p [ .)(3) 1 o A T 3 Ao lim .
. MiillerKirsten [30]. by either replacing A T by 2AT or doubling A*.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs.113) is very complicated. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state.With this correspondence A\E of Eq. W. Liang and H. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — . 23. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals. J. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton). To help the comparison we note that — with indices indicating the authors.182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. Bose and H.^r . even without explicit specification of the prefactor. J.618 and hence AE A.113) for qo = 1: A1E „ . has been obtained without explicit use of canonical commutation relations (i. MiillerKirsten [163]. explicit quantization).e. Wiedemann [4]. . We observe one more aspect of our calculation. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S. Finally we realize that the computation of (23. The difference between levels.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. H. AE. K. (18. ^See P. (23. we obtain AE = 2A* and agreement with the result (18. here as 1) is also taken into account.Q. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ .W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). Gildener and A. l CHAPTER 23. Achuthan. GP for those of E. J. (23.' We see therefore that the classical Euclidean configurations 4>c can be considered to be responsible for the quantum mechanical tunneling which gives rise to the level splitting AE. MiillerKirsten and A.113). — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92. W. W. In fact the equivalent to writing down canonical commutation relations is the evaluation of the Gaussian integrals which enter the prefactor of the expression in Eq. J. We see that the method — above we included a lot of highly nontrivial details which are passed over in other literature — is considerably more complicated than the method of the Schrodinger equation.5 Concluding Remarks In this chapter we used the path integral to obtain the level splitting of the ground state of a particle in the double well potential. Patrascioiu [110].
Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance. We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues. i. Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 . and their interpretation and uses. however. In order to really appreciate the difference between topological and nontopological configurations (the solitons and instantons we considered previously being of the first kind) we begin with a recapitulation of some crucial aspects of the former. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp. in a somewhat different formulation. 619 (24.a 2 ) 2 + c^T2.1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling).Chapter 24 P a t h Integrals and Bounces on a Line 24.e. ™=£('^° < 2 «> Here the two symmetrical minima of the potential are called vacua (classical or perturbation theory vacua as they are sometimes called because ordinary RayleighSchrodinger perturbation theory would be performed around these points).2) .
A a 2 ) . 24.1 The asymmetric double well. J 4. = 4(ACT2 cT 2 ).620 CHAPTER 24. We can destroy the reflection symmetry of V((f)) (i. but implies instability for T < Tc.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 .e. > / V[<H 'true'vacuum 'false'vacuum Fig.a2)4> + 2cTz 2 0. its symmetry under .2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. Path Integrals and Bounces on R1 Here dVT{4>) vanishes for .e. a spontaneous symmetry breaking leads to the double well (i. T = 2\(<f>2 . at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) . Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool. We see that <fi = 0 minimizes VT(</>) for temperatures T > TC. A2 _ a — \„2\ u ) ~r u/\y> 2 (24.
Guth and S.g. Cornalba.5) We see that one minimum of V now lies higher than the other. which (b) are saddle point configurations of E (or SE). the vacuum being understood as the classical.1 with (24.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e. Since the true vacuum minimizes the potential. a linear term as in w^^'A1^'*0V(<l> = m/g) = 0. and (c) determine the lowering of the energy to the true quantum mechanical ground state (i. V(<j> = m/g) = 2e. the quantum mechanical vacuum).) . .24.t In the following we consider first (after some recapitulations) a simple model of a false vacuum.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . the higher minimum is called a "false vacuum". False vacua play an important role in models of the inflationary universe* and elsewhere.g.7) *See e. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions. Pi [126]. T False vacua and bounces occur. M. (24. for instance. . Penedones [59] and references cited there. The Lagrangian density of the theory is given by C = dp*d»*U(*).g. (24.Y. We calculate the classical configuration — which in this case is called a "bounce" — which corresponds to the soliton or instanton in our previous chapters (but is not topological) and we use this in order to calculate the lifetime of the unstable state (or equivalently the imaginary part of the energy). Costa and J. A. also in contexts of string theory. <44 2) ' This is now an asymmetric double well potential as in Fig. S. and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE). tf(*) =^ 1 . On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions.^J >o. H. L. We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory. 24. see e.e. perturbationtheory vacuum).
14) .7) yields the following equation called stability or small fluctuation equation. small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention). h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x). — XQ) (24.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . We observe that this kink solution is odd in (x — XQ). The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c.12) Since <fic(x) solves Eq.8) This equation possesses the kink solution <t>c{x) m tanhm(a.t) dt 0 are written (f>(x). (24. Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24. Path Integrals and Bounces on R1 Static fields given by d$(x. which is an important point as we shall see. (24. we see that (24. i. so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). t) = <f>c(x) + Y^ iPk{x)ekt (24. (24.13) For small w\ > 0.e. (24.9) 9 for the doublewell potential given above.e. dx2 + u"(<j>) 5(x y).10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24. We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional.622 CHAPTER 24. i.8).
16) V[<H x= oo 0 d2 .17) . (24. 24. i. 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24. i. the zero mode).e. We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue. the other a positive quantity) and a continuum which are given respectively by w\ — 0. .2 The inverted potential. o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = .e.15) cosh mx From our study of the solutions of this equation in Sec. underneath the particle's velocity.24.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24.oo \ x== + CX) Fig. kink. . configuration (for our </>4theory). d \m .8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero. 3m 2 and w\ > Am2.1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite. In the case of the double well potential (24. thus the zero mode here is given by d . i. 22.e.
the associated zero mode. In this case the usual Newton equation reappears with an opposite sign of the potential.18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U.e. Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton. Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function. (24.e. 4>"{x) = U'(4>). 24. Path Integrals and Bounces on R1 We observe that this zero mode is an even function of (x — XQ) in agreement with our expectations for the ground state wave function. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function. F (24. i.624 CHAPTER 24. We see that this pseudoparticle starts from rest at one peak of — V at x = — oo and in losing potential energy gains just enough kinetic energy to reach the second peak at x = +00.20) The number 1 on the right is the socalled topological charge which is obtained as follows. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. where it is then again at rest. (24. i.3 (24. The classical configuration is then called a pseudoparticle configuration or instanton.8). and — clearly — we can resort to a classical interpretation of this function. so that its derivative representing the pseudoparticle's velocity is odd. by Q = \(°° dxf^eoid1^) = l[M°°) ~ M°°)] = 1(2422) . quantum mechanics. which has the solution 4>c(x).2. In the pseudoparticle consideration the coordinate x plays the role of time.e. We define the current k» := e^dvfc. Then the energy satisfies f°° . Integrating the classical equation (24. with Euclidean time. Below we shall see that the configuration described as a bounce is given by an even function of x. 4 TTT. The associated