•
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
in Feynman diagrams in quantum electrodynamics). the aforementioned exactly solvable cases. even though the expansions are mostly asymptotic. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. numerous problems could. whereas for the calculation of discrete eigenvalues the Schrodinger equation. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. 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. Thus this first part . that is the Coulomb potential and the harmonic oscillator. With the growing importance of models in statistical mechanics and in field theory. its mathematical foundations. These basic cases will be dealt with in detail by both methods in this text. screened Coulomb potentials and maybe singular potentials. again point the way: For scattering problems the path integral seems particularly convenient.e. one problem being that there are few nontrivial models which permit a deeper insight into their connection. basic postulates and standard applications. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. with degenerate vacua) were first and more easily derived from the Schrodinger equation. like periodic potentials. 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.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. In the first part. To what extent the two methods are actually equivalent. be treated to a considerable degree of satisfaction perturbatively. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. With various techniques and deeper studies.g. has not always been understood well. However. but also about complex energies that he encounters in a parallel course on nuclear physics. in fact. but arise also in recently studied models of large spins. 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. Chapters 1 to 14. Excluding spin. we recapitulate the origin of quantum mechanics.
Using perturbation theory.e. Thereafter the concepts of instantons.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. and their eigenvalues. and the elliptic or Lame potential — here introduced earlier as a generaliza . 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 most prominent examples here are the double well potential and its inverted form. we deal mostly with applications depending on perturbation theory. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. i. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). The solution of this case — however in nonperiodic form — turns out to be a prerequisite for the complete solution of the Schrodinger equation for the singular potential 1/r 4 in Chapter 19. In the 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. Chapters 15 to 20. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). In the second part. periodic instantons. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. bounces and sphalerons are introduced and their relevance in quantum mechanical problems is discussed (admittedly in also trespassing the sharp dividing line between quantum mechanics and simple scalar field theory). In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. the chapter thereafter deals with Yukawa potentials. The potentials with degenerate minima will be seen to reappear throughout the text. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. which is presumably the only such singular case permitting complete solution and was achieved only recently. After a treatment of power potentials.
for instance. it is. This puts Schrodinger equations with e. and whenever available also with the results of WKB calculations. quartic. 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. we can make the following observations. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. the Mathieu equation. from our precise reference to unavoidable elliptic integrals taken from Tables). 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. such as limiting procedures. anharmonic oscillator potentials on a comparable level with. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. this comparison on a transparent level being one of the main aims of this text. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. except that these are no longer of hypergeometric type. With a fully systematic perturbation method and with applied boundary conditions. In fact.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. We then illustrate the relevance of this in the method of collective coordinates. however. 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.g. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. All results are compared with those obtained by perturbation theory. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. cubic). an approximation whose higher . for instance. 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. This method is therefore more complicated.
* H. H. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there).g. Park (Masan). Tchrakian (Dublin) and Jianzu Zhang (Shanghai). Whittaker and Watson [283]. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. 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. Watson [283].XIX order contributions are difficult to obtain. thereafter University of St. D. not the least for permitting a more detailed and hopefully comprehensible presentation here. K. W. Liang (Taiyuan). . Andrews). For ease of reading. Throughout the text some calculations which require special attention. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. N. Since this comparison was the guideline in writing the text. as well as applications and illustrations. Whittaker and G. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains.g. e. The line of thinking underlying this text grew out of the author's association with Professor R.g. Their deep involvement in the attempt described here is evident from the cited bibliography. Dingle (then University of Western Australia. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. shows him that each such formula here has been properly looked up). are relegated to separate subsections which — lacking a better name — we refer to as Examples.Q. we also consider at various points of the text comparisons with WKB approximations. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. J. the references referred to are never cited by mere numbers which have to be identified e. 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. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. T. also for the verification of results. As a rule. The author is deeply indebted to his onetime supervisor Professor R. Instead a glance at a nearby footnote provides the reader immediately the names of authors. like E. D. 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. Nonetheless. B. in particular Professors J. with the source given in the bibliography at the end. B. MiillerKirsten *In the running text references are cited like e. at the end of a chapter (after troublesome turning of pages).
Instead. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery.2 cm at moderate temperatures T) is the radiation emitted by a body (consisting of a large number of atoms) as a result of the temperature (as we know today as a result of transitions between a large number of very closely lying energy levels).1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). very few explain in this context what he really did in view of involvement with statistical mechanics. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . the amount of radiation absorbed by a body is equal to the amount the body 1 . Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. 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.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.Chapter 1 Introduction 1. Kirchhoff's law in thermodynamics says that in the case of equilibrium. We do not enter here into detailed considerations of Planck. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it.
Let us look at the final result of Planck. (1.38 x 1(T 23 J K'1. and then measuring the increase in temperature of the heat bath.e. the formula (to be explained) u(u. These expressions are: (1) Wien's law.626 x 10 .1) containing the constant h by treating the radiation in the cavity as something like a gas? By 1900 two theoreticallymotivated (but from today's point of view incorrectly derived) expressions for u(u. Introduction emits. Fig.2 CHAPTER 1.T) = dv3eC2U/T. (1. A being the wavelength of the radiation. Black bodies as good absorbers are therefore also good emitters. i..1 Absorption in a cavity.e. u(u.e. How did Planck arrive at the expression (1. energy per unit volume) of the radiation (i. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. i. h = 6. 1.3 4 J s.1) ' y Here u(v. and the other in the region of large A (or AT). It was found that one expression agreed well with observations in the region of small A (or AT). v + dv in equilibrium with the black body at temperature T.e. In Eq. T)du is the mean energy density (i.l ) where x = ^ = ^ . radiators. 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. T) were known and tested experimentally.2) and the .T) = 2*?£(?)kT. The parameters k and h are the constants of Boltzmann and Planck: k = 1. y J c 3 \ex . kT kXT (1.1) c is the velocity of light with c = u\.
Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. This is the point. C3 being constants. e xhv. Here W is a number which determines the distribution of the energy among a discrete number of objects. "small" (i. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously.e. that the formulas (1. which are indistinguishable) among the N indistinguishable oscillators at . and thus over a discrete number of admissible states.2 Distributing quanta (dots) among oscillators (boxes). where the Fig.2) and (1.T) 2^^kT. T) u(i/.1) as approximations. Considering Eq.1. exp(x) ~ 1+x) and "large" (exp(—x) < 1). C2. 1. We see.47TZ/ 2 {x small). When Planck had found this expression. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. (1.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>.3) are contained in Eq. we obtain: u(i/.T) = av e6"/Ti' where a and b are constants. To this end he considered Boltzmann's formula S — klnW for the entropy S. in the first place Planck had tried to find an expression linking both. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". he searched for a derivation. Indeed.1) in regions of a. (x large). 3 (1. . and he had succeeded in finding such an expression of the form u(v.T) = 2^C3T. (1. discretization begins to enter.3) Ci. but — here the discreteness appears properly — only in elements (quanta) e.
g. p. Then W is given by w = (N With the help of Stirling's formula* {N + piy. .2. Oberhettinger [181]. formula 8. p. with the quanta represented schematically by dots as indicated in Fig.2) requires that e ex is. i. Introduction temperature T. u(i>. We visualize the iV oscillators as boxes separated by N — 1 walls. Ryzhik [122]. 940.e. one obtains (cf. e. N * oo.6) Fig. S. i. e = his.343(2).1)!P! (1. there not called Stirling's formula.3 Comparing the polarization modes with those of a 2dimensional oscillator.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. We now obtain the energy density of the radiation. h = const. 1.4) IniV! ~ JVlniViV + O(0). Agreement with Eq.T)dv. (1. 1. (1. . U{T) being the average energy emitted by one oscillator. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. W. Magnus and F.4 CHAPTER 1. = 1/T). as in most other Tables.7) *See e. v + dv.g. The Stirling formula or approximation will appear frequently in later chapters. as for small values of e).e. M. and the second law of thermodynamics ((dS/dU)v Example 1. with riydu — 2 x —w—dv.3. Gradshteyn and I. I. (1.
(1. 2 2 2 r2 L K — 7T n .UJ = 2KV.e.. We obtain the expression (1. rii = 1. so that .T)dX = ^ehc/*kT_idX. Then L^j = nrii. is given by .i = 1. as indicated in Fig..3.965 and xmin = 0. '''From the equation I \ JW .V 2 ) E = 0.8) This is Planck's formula (1. where we have for the electric field E oc elwt \ J eK sin KI^I sin K2X2 sin K3X3 K with the boundary condition that at the walls E = 0 at Xi = 0.UJ2/C2 + K? = 0.. KT = I J ..7).3. The number with frequency v in the interval v. We obtain therefore u^T) = Unv = 2^fJ^—i. (1.1. .2.. In terms of A we have u(X.3 (as for ideal conductors).1.T) has a maximum which follows from du/dX = 0 (with c = vX).3. L for i = 1. . so that (lvL\A 0 I I = rr.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. so that the derivative of u implies (x as in Eq.7) for instance. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. i.2. nvdv per unit volume. where^ 2 [2KUY . (1. v + dv. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^..1). as in electrodynamics. We observe that u(v. dj\l dM .2.1)) The solutions of this equation are ^max = 4.
8) could be derived much more easily in the context of statistical mechanics. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. Thus we have a rather complicated system here.10) Thus here the socalled zero point energy appears.6 The first value yields AmaxT = CHAPTER 1. One expects. However. Introduction he 4. (1. A. the behaviour of the system at zero absolute temperature. i.2 (1. . Since temperature originates through contact with other oscillators. 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=—. then (with x = hv/kT) its mean energy is En=0e = nX dx ^0 — /ii/— In = huf r%e dx 1 — e_x (1 — e~x)z hv . which did not arise in Planck's consideration of 1900. and from which the constant h can be determined from the known value of k. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. ra = 0.—v <L9) We observe that for T — 0 (i. x — oo) the mean energy vanishes (0 < U < * > oo).1. that of an oscillation system at absolute temperature T ^ 0. One might suppose now. we then have at T = 0 independent oscillators. .965K = Const.l. such a procedure leads to contradictions. We are not dealing with the linear harmonic oscillator familiar from mechanics here. of course. We therefore examine such contradictions next. n = 0. which had also been known before Planck's discovery. 2 . . Lorentz and Planck that Eq. but one can expect an analogy. .e.10). Later it was realized by H. . that it is easier to consider first the case of T = 0. which can not be eliminated without a different approach.e. which can assume the discrete energies en — nhv. This is Wien's displacement law.
d . 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) . 1.In .ln(TV . for 2 degrees 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.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later. so that 1 f ds\ T — \dUjy . with Heisenberg's discovery of the uncertainty relation..k dU 1+U\ f ln(l T + U\  U U k ( e . In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").1.2 Contradicting Discretization: Uncertainties 7 Example 1.InP!] ~ kN The second law of thermodynamics says 1+ 7 ln 1 + ( 7)~7ln7 \au)v For a single oscillator the entropy is s = S/N. we obtain T V S =fc[ln(TV+ P .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. .= .i n . kT/2. around 1926.1 ' u~ e . i.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .1: Mean energy of an oscillator In Boltzmann's statistical mechanics the entropy S is given by the following expression (which we cite here with no further explanation) S = fcln W.1)! .e.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.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
1). (2.P.Pi) is contained implicitly in the canonical variables q^ and pi. The total time derivative of u can therefore be rewritten with the help of Eqs. du \ x^/dudH du dH\ . S »(«.Pi. the equation of motion of the observable u. which can be observed directly at time t. Hamiltonian Mechanics is wellknown. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.1). Compared with an arbitrary function f(qi.pi) of qi. . All functions u(qi.1) as d .t). (2. (2.24 CHAPTER 2. Goldstein [114]. (2. qi(t + 6t) . A system consisting of several mass points is therefore described by a number of such variables. in analogy with Eqs. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. It suggests itself therefore to consider more closely the properties of the symbols (2. this expression is simply a functional determinant. which all together describe the state of the system.2) as This equation is.4) contains as special cases the Hamilton Eqs.qi(t) qi = hm f . The following properties can be verified: *See e.2) If we have only one degree of freedom (i = 1). i. Pi has to be considered as an independent quantity. chapter VII. since . One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. A. one obtains the Hamilton equations* • OH .3). M.1). (2. 6t>0 5t Real quantities which are directly observable are called observables. x^fdu. (2. Rather. the entire timedependence of observables u(qi. Dirac [75]. H. We can therefore consider Eq.4) as the generalization of Eqs.e. whereas the velocity requires observations of space coordinates at different times. One can verify readily that Eq.p% are therefore again observables.g.^ ^ j . ^As H.) = £ [wm + WiK) = £ [WiWi . chapter VIII. as mass times velocity.n n.. . (2. the standard reference for the application of Poisson brackets is the book of P. Goldstein [114] remarks at the end of his chapter VIII. (2.
Pk} = 0. {qi.5a) (2. which combines the Hamilton equations. B} is completely evaluated.C}.6) give the values of these.5b) (3) complex conjugation (note: observables are real. 25 (2. in other words without any reference to the original definition (2. (2.B}. {A. the Poisson bracket {A.5d) = {A*.6). (2.B*}. Later we shall consider noncommuting quantities. (2. Property (2.5c) The first three properties are readily seen to hold. like here.5d) above.Qk} = 0. As long as we are concerned with commuting quantities. B}. {pi. C}} + {B.B} = {B. {B. Since Eqs. a i S i + a2B2} = ax{A.B}C + B{A. but could be multiplied by a complex number): {A. it is irrelevant whether we write {A.B}* (4) product formation: {A. (2.A}. that the very general Eq.3) of the Poisson bracket. (2) linearity: {A.4). then the ordering is taken as in (2. These are {Qi. i.e.Pi. we obtain the fundamental Poisson brackets.BC} (5) Jacobi identity: {A. where A and B are arbitrary observables.2. (2. Bx} + a2{A. B}} = 0. {C. A}} + {C.B}C or C{A.5d) is useful in calculations.Pk} = 5ik. we wish to evaluate {A. The original definition of the Poisson bracket will not .5e) = {A. that of the linear harmonic oscillator. (2.6) We can now show. If we evaluate the Poisson brackets for qi.2 The Hamilton Formalism (1) Antisymmetry: {A. B2}. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. for example. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. If. As an example we consider a case we shall encounter again and again.
Since constants are also irrelevant in this context. (2. Hamiltonian Mechanics be used at all.9) Similarly we obtain from Eq.p\) P.11a) 'q' = q.4) we have for u = q.p) = ±(p2 + q2).9) and (2.6).t).11b) (2.t).p.. Pi—> Pi = Pi(q.26 CHAPTER 2. or q(t) = qo+ Pot . In the evaluation one should also note that the fact that g. (2.7) into (2.12) .10) In classical mechanics one studies also canonical transformations. we consider as Hamiltonian the function H(q. and P={p. These are transformations qi—>Qi = Qi(q.\(p2 + q2)} = l({q.p.yPot3 + •••• (2.q2}) = = 2i{q. (2.p. q + q = o. (2. Then we obtain: (2.10) we deduce q = p = q.8b) We insert (2.qot2 . From Eqs. from which we infer that q(t) — qocost + posint. (2. "4' = q.8b) p=q..H}. q = {q. (2. (2.H}. According to Eq.P2} + {q.. (2.8a) and use the properties of the Poisson bracket and Eqs.8a) (2.p}p + p{q..7) q = [q. and Pi are ordinary real number variables and that H(q.p) is an ordinary function is also irrelevant. and so q = q.
qk} = 0. Eq.P. chapter VIII.15a). (2. Example 2. i. i.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. Example 2.e. as {A.P One can then show that {A.e. (2.pk} = 5ik. {qi. which means that a Hamilton function K(Q.P.* Hence in Example 2. (2. Of course. Pi^Pi=Pi(Q. and Example 2.e.13) We write the reversal of the transformation (2.Pk} = 0.2. that (dropping the subscripts therefore) {Pi. {Qi. _ax p__dK_ (2.Pi back to their constant initial values. {QhPk} = Sik. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.P is canonical in the sense defined above. (2. i.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). i.1 we verify only Eq.14) With the help of the definition (2.B}Q>P. . i.B}q. {qi. {PhPk} = 0. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.P provided the transformation q.B}q. *H.Qk} = 0.p <> Q. B} of two observables A and B in terms of either set of canonical variables.t).e. P) exists.12) Qi—>qi = qi{Q. which transforms qi.p = {A. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation. those at a time t — 0.15a).2 below contains a further illustration of the use of Poisson brackets.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. for which Hamilton's equations hold. (2. verify that the Poisson bracket of two observables A and B is invariant. Goldstein [114].3) we can now express the Poisson bracket {^4. .3 deals with the relativistic extension.e.15a) °raS {AiB}Q.t).
236. we have —Et. their product Et — qp being relativistically invariant. § E x a m p l e 2. Mittelstaedt [197].Qj}q.+ dQ. Thus whenever q and p are multiplied. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo.E(t). p. Solution: Relativistically we have to treat space and time on an equal footing.3: Relativistic Poisson brackets By extending qi. we obtain analogously dA dQk' Inserting both of these results into the first equation.( 9A = {Pk.15b) dA {Qk.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential). § See P. OQk Pk}q. (2.p = dB dA dB \ rA 1 {A'B^ E b r s r .q/c) and (E. {A.B}q.A}q. (2.28 CHAPTER 2. we obtain dA {Qk.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. 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 „ „•. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.s F 5 7 r ={A.p = J2 d~Fk Replacing in the above A by P and B by A.p^v dP. dA {Qk. V V °Qk dPk dPk dQk J E x a m p l e 2.P^pr.15a) r„ ™ v .p(0) = poSolution: The solution can be looked up in the literature.F}r ~di d~E ~ d~E~8t Consider F = H(q.pc). Replacing here A by Q and B by A.p^— + {A.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.B}QIP. .A}q.p) .p^ "Pk.Pj}q. Qk}q. we obtain as claimed by Eq. Thus we extend q and p to spacetime vectors (t.
Po. but partial derivatives of F do not vanish.' ±=j1V c \dtj' u^. (*. 2TTE UJ and hence g oc — . But now we consider a system whose phase space coordinates are not known precisely. also boundary points of one domain are mapped into boundary domains of the other.1 Liouville Equation. it is the equations of motion which lead from Go(q.17a).3.Po.14) and area A of the phase space ellipse given by Eq.E(t)}r = Hence at Relativistically we really should have clu/dr. in the case of the linear oscillator with energy E given by Eq. and E as a function of t). i. du . we have A = (p apaq — J .t0.to.2.p). Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.._. to . Instead we assume a case in which we know only that the system is located in some particular domain of phase space.p) to G\{q. Of course. p= p(qo.p) around some point qo. numerically zero.p). of course. since H is expressed as a function of q and p. g.. . (2. c dt 2 2 w^ du du dr .16) For example.t). I . Let us assume that at some initial time to the system may be found in a domain Go(q.dt y/l  qZ/c2' 2. We consider first the a priori weighting or a priori probability. dE .Pi are the coordinates of some mass point m.e. and the space spanned by the entire set of canonical coordinates is described as its phase space. one obtains Gi with q = q((lo.t). 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. i. and at time t > to in a domain G\(q.y.. (1. .p) . This probability is evidently proportional to AqAp. so that Go(qo.. Probabilities 29 (This is. Since Hamilton's equations give a continuous map of one domain onto another.e.3 Liouville Equation.3 2..po is a point on Go. (1.po) = Gi(q. Then .)•'=(*)**£. if qo.po. g oc AqAp. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . H(q. dudH s l u . We distinguish in the following between two kinds of probabilities. du q\ P = —. rrr .p).
)curve for constant E and $ is — as may be seen by comparison with Eqs. Pi 2 / \.14) t o (1. (1. as is demonstrated by Liouville's theorem in Example 2. Example 2. 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. dt dq dq dp dp d2H dqdp d2H = 0.. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle.— = — Aq. dt dq and with with Hamilton's equations (2.1): d . Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. = 2TrIEsm6. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . Show that the total volume of phase space covered for constant E is 8n2IE. this has the same value at a time to.4: Liouville's theorem Show that A q A p is independent of time i. the (pg.4. and hence g oc %ir2IdE. which means. Je=0 .p^. J0=0 Hence g oc 8n2IdE. dpdq and similarly dt = — Ap.eded<t> = 8TT2IE. in view of this independence it can be expressed in terms of the conserved energy E. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. Hence from the difference: — .5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. Thus g must be independent of t. — dp A A — In(AqAp)s = — + — = Example 2. Example 2.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.15) — an ellipse of area § dpgdp.30 CHAPTER 2./.
3 Liouville Equation. (2.p.t)dqdp = N.p.1 The system moving from domain Go to domain G\.e. i.18) With a suitable normalization we can write this / W(q.2. W=p(q. 2.3. 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.q + dq. dn — p(q. (2. d d 1 P = JJdqidpi. by jr = P(9»P. The total number of systems N is obtained by integrating over the whole of phase space.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. whose positions in phase space are characterized by points. i.p + dp.20) .p. Since W has the dimension of a reciprocal action.t)dqdp= 1. (2. however. it is suggestive to introduce a factor 2KK with every pair dpdq without. leaving the basis of classical mechanics! Hence we set / «w)^ =. (2.Po.17) F ^GT^^ 0 Fig.p.19) Thus W is the probability to find the system at time t at q.*). Thus dn is that number of systems which at time t are contained in the domain q. Probabilities 2.t).e.Thus we assume a large number of identical sytems.
e.p. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.e. 2.t)dp{ jt Q.e. t + dt)dqdp.a t dt + q+dq.t)dp( — I . * q Fig.P dt p(q + dq. p + dp G P O q qndq Fig. how the system moves about in phase space. The number of points at time t + dt in domain G. The equation of motion for n or W is the socalled Liouville equation. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. 2.2 The ensemble in phase space.p. — if vq(q. we take into account.32 CHAPTER 2. t + dt)dqdp — p(q.3 and establish an equation for the change of the number of points or systems in G in the time interval dt.p. that in our consideration no additional points are created or destroyed. i. p(q. i. 2.p / .3 The region G. In order to derive this equation.p. In doing this.p) denote the velocities in directions q and p — p{q.p) and vp(q.p. t)dqdp p(q. we consider the domain G in Fig. i. We are now interested in how n ov W changes in time.
21 ) This is the Liouville equation which describes the motion of the ensemble or.21) we can also write dW(q. (2.p(q.p(q + dq. With Eqs.p)=p = 3H —.P *' % = iH>P} ( 2 .p + dp)dtdq.p + dp.t)vp(q. (2. Comparison of Eq.19).p)dtdq . put differently.p. vp(q.p. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . = K(9.P)].t)vq{q.p) .P)]K(Q. p.p + dp) dp = or  However.P.t + dt) . t)vq(q + dq.p.p.t) dt p(q.p.t)vp(q. the probable motion of the system under consideration.p) = q = g^. (2.t)vq(q + dq. Probabilities and thus p(q. (2.p)dtdp +p(q.p) .p).p + dp.3 Liouville Equation.p.t)vq(q.21) with Eq.t + dt)dqdp — p(q.p. t)dqdp = p(q.p.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . .4) shows that p and u satisfy very similar equations.t)^ = 1.p)dtdp . 33 Dividing both sides by dqdpdt this becomes p(q. p. dH Mq. .p) dq p(q.W(q.p(q.t)vp(q.t) dt {H(q. (2.p(q + dq.p.p.t)vp(q. . .20) and (2.2.p(q.t)} with JW(q.
The first and most immediate possibility is — as indicated .4). t) depends explicitly on time t (if determined at a fixed .1 the area Go is equal to the area G\. Example 2.26) we described the time variation of the observable u(q. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. p+dp.e. We define as expectation value of u(q.p. We deduce from the Liouville equation the important consequence that ^ M = 0. Equation (2.24) since the total derivative is made up of precisely the partial derivatives contained in Eq. With Eq.24). 2. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular. (2.p)W(q. Thus in Fig. i.p) be an observable.4).po are the initial values of q. We shall see that we have two possibilities for this.p (cf. (. (2.p). (2.4 Expectation Values of Observables Let u = u(q. We now inquire about the time variation of the expectation value (it) of u.that the density or probability W(q.34 CHAPTER 2. p.p. (2. for i<«>!/«<*p>"w>(^)".24) implies that p is a constant in time.p) the following expression: (u)=Ju(q. and hence that equal phase space volumes contain the same number of systems. that dtj\y v if qo. q+dq.t)(^J. i. 2. since no systems are created or destroyed.e.
and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.p) = u(q.Po. (2.f(q0.Po.f(qo. at t = 0.24): W(q. ie.p).t) W(q0. (2.8.p) = uo{qo. that (u) = (u)0.^ Solving the equations of motion for q. We expect.Po.Po. Reversing Eq. We verify this claim as follows.t). of course.31) i.27) becomes !<»>  /«(*P>!"W>(^)" (2. (2.p.33) po = f(q. we have Qo = g(q.p0)(^^J. we can also employ a more complicated consideration. Thus we can write.W(q.i). (2.0) = W0(q0.t).28) = Ju(q.22).po.32) In this expression the time t is contained explicitly in the observable u(q.po.p.p.34) S e e also H.t)W0(q0.29).t).t).p. (2.t)}(^y. so that Q = g(qo.0) =u(g(qo.Po.t). (2.30) p = f(qo. since W oc p is constant in time according to Eq.po. 8. where we used Eq. Observables 35 point in phase space).p0.t) = = W(g(q0. and hence u(q.4 Expectation Values.29) The distribution of the canonical variables is given by W(q. Sec. However.0) = u0{qo.p. (2.p){H(q. we can express these in terms of their initial values qo.t).p) assumes certain values) that depends explicitly on time.t). 1 (2. Then Eq. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t).2. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. .e.p.p. (2.t).t).po.Po) at time t = 0. (2.po. Goldstein [114].
M 27Th i .t).t) d t _ (du\ \ d<i dq / P=f(qo.p. Eq.p. (2.t)W "(q0.\„.f(q.po.po.29) q = g(g(q.t)J(q.u(q.=9(.36) = (2. d dt {U)0 = dt.t).30) u0(g(q. (2. (240) . (2.t) dp M du\ ° q 'P=f(qo. we obtain an expression which is different from that in Eq.P.po.39) Substituting this into Eq.(dq0dp0\n We deal with the partial derivative with the help of Eq.p)W{q.30): duo(qQ.t).t).p)})q=g(QO.25) into Eq.) f / u (q0. Taking now the total time derivative of (it)o.t).p). (2.t).o. (2.32).t). (2.p.t). [ du0{q0.t) (2. (2.p).t).31)) W0(q0.f(g(q.p. p = f(<j(q.p.p.38) we obtain £<«)„ = .Po)(^^)n.f(q. (2.f({H(q.Po.0) u(q.PO.P.p).p.f(q. (2. / ^fdq0dpc .t) dB__ (du\ °P \°PS P=f(qo.P0.0)=u(q.t) 0H_ 0 q =  {{H(q.36 CHAPTER 2. (2.35) Inserting these expressions into no we obtain uo(qo.37) as had to be shown. i.) Wo(qo.Po. Inserting Eqs.f(q.t).t) = (2 30) (2.28).32).po)(~ ." V " " ™0' > • • » « » 0 .P0.35).p0) = W(q.Po.t) .u{q.t).e. (cf. Hamiltonian Mechanics so that on the other hand with Eq. (2.t).30) Moreover.t)\—^\ = (u).Po.t) v{g(<j(q.POit)i p=f(qo.p.t) M f 9u\ V °P / P=f(q0. .(2.37) and (2.P)}).t). we obtain («)o = / u{q.p.p.t).p.36) and (2. Eq.
The phasespace integral of a Poisson bracket like I (2.28) and (2. from the properties of the Poisson bracket.. u}W + u{H.t) of u assuming certain values q and p. and the time variation of (u(q. u}W instead of u{H. W}. (2. described as "Schrodinger picture".p). Alternatively we could deduce from Eqs. This timedependence is described by the Liouville equation dW(q.41) This expression contains {H. x r „. dt <«)o v f u(q.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions.45) in agreement with Eq. In the first case. (2. ^ „.p)}W(q. we obtain {H.t) dt {H(q.p.41) and (2.45) the Liouville equation.p.42) the relation d .p){H(q.u(q.34) and use (2.p).. (2.t)}.p).t) dqdp\n 2irhJ (2.p)) is attributed to the probability W(q. However.25) and (2. the observable u is treated as a function u(q.W(q.4 Expectation Values. Consider {H..31).28). lim ——uW — 0.2. Observables 37 Here we perform the transformation (2.43) ——uW = 0.p. (2. we obtain zero after partial integration of I and hence from Eqs. so that ~d~t <«>0 = J{H(q.p. / d q d p \ n 2irh J (2.uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2. .28).p.p).W(q. uW} = {H.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). .W} in Eq.t)} .(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.
5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. as we observed.38 CHAPTER 2. 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.po. but does satisfy the usual associative and distributive laws. for which the axioms of a commutative group apply.e..4)) ^ = {«(.t). This formulation deals with observables u representing physical quantities. (2. (b) As usual a muliplication by a complex number is defined. (2. the reason being that — since qo.rt.p) It = {u(q. as the equation of motion of an ensemble or of a probability distribution on the one hand.po are constant initial values — we have du0(q0.p0.** 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. " T h e author learned this approach from lectures of H. the probability of the initial values Wo(qo. (cf.t) dt __ ~ (2J30) ( 4) du0(qo. and probabilities W.H(q.p). and the equation of motion (2.F(«. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations.po) is assumed." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.46) We thus also recognize the connection between the Liouville equation. which describe the state of a system.4) of an observable on the other.p)}.p)}. which.po. (c) A multiplication of the quantities among themselves is defined. Hamiltonian Mechanics In the other case. 2. . does not have to be commutative. The equation of motion is then that of an observable.t) dt du(q. Eq. " This means: Only in the Heisenberg picture dW/dt = 0. called u Heisenberg picture". Koppe [152] at the university of Munich around 1964. The time dependence of the expectation values can be dealt with in two different ways. i.
q} = 0 {p. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. (b) TV(cm + j3u) = oTV u + /3Tr v.2. In Chapter 9 we attempt a corresponding approach for classical systems with .p] = l.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. px > in—. Eq.) = TV (vu). (2. it is helpful to introduce already at this stage some additional terminology. (2.48) In matrix theory the symbol "trace" has a welldefined meaning.6) a socalled "commutator''' [A. ox In view of our later considerations.q}=0. also written "TV". so that we consider this next.48): (a) TV u* = (Tr «)* = TV^. can be deduced from the following characteristic properties of a trace. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. y:\p. that its use here implies the essential properties it has in matrix theory. (c) TV (u*u) > 0. Introducing it here assumes.p} = l {q. if u ^ 0.26)) (u) = Tr(wW). (d) Tr(ut.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. also called "density matrix". That this is the case. B] := AB — BA in the following way: 'i {q. for a phasespace integral we define the word or symbol "trace". which all apply to (2. Thus we can write the expectation value of an observable u (cf. and to be able to correlate these with the above classical considerations.p)(^y.p} = 0 • — ^ ~[q. Moreover. ~[q. by Traceu:=Ju(q. therefore. (2.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.
e. i.40 CHAPTER 2. electrodynamics. the Poisson brackets require modification to Dirac brackets. . and. a constraint) has to be taken into account. since gauge fixing (i. excepting the Poisson brackets. it will be shown that in electrodynamics.e. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. Hamiltonian Mechanics a wavelike nature. These aspects will be considered in Chapter 27. However. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation.
A set M. if the elements ipi of M satisfy the usual axioms of addition and K 41 .e. i. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. We found that the Poisson algebra permits extensions to noncnumber formulations.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. These somewhat abstract considerations — although later in many cases not referred back to — are a necessary prerequisite for the formulation of a mechanics which is not of the cnumber type as classical mechanics. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. 3. we can define the Hilbert space as the space of states of a physical system. Building upon this. i.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. 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. of measurable quantities. which turn out to be those of the theory today known as quantum mechanics. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. and selfadjoint operators in this space as representatives of observables.e.
ipn are said to be linearly independent. n exist (not all zero). (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). . • • • .ip2. .3) If all on = 0. fa. .2) Vectors ip\.^i) .M. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. so that n 5 > ^ = 0. (3.ip2). (ipi. A + tpj = tpj + i'i./?eK). i. (ipi + ipj) + ijjk = ipi + (ipj + ipk). if every vector ip E M can be associated with numbers Q .5b) .1) aipi + aipj. (V>. the vectors il)i. if any two elements ip\.^1) = (01. (3. .^) called inner product. ip2 of this space can be associated with a complex number (V'l. . . f > 0 if V T ^ O .tp2. ^ i = 0 if ^ = 0. n..ipn are said to be linearly dependent. such that n ^ = YJCi^i. with the properties (a* G IK): (^2. x M.~ip2) '• M.i = 1.M is said to be a metric vector space or a preHilbert space. E (3. .aiV>i + a 2 ^ 2 ) = ai(ip. (3 5c) + a2{ip.42 CHAPTER 3. In each case n linearly independent vectors are said to form a basis. (aptyi.4) The vector space . —»• IK. If n + 1 elements ^ € M. . i = 1. (a.e. 2 .5a) (3. ( n (3. i=l (3. . and n is the smallest such number. are linearly dependent. n is called the dimension of . . if numbers eti. 2 .  .02)* (hermiticity).
9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.3. A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i.^1) 2 h 11 0 <%«e + M . \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). tp2 £ M: 1(^1.tp2 £ M.7) In addition the following relations hold in a metric space M.9c) HV'i + tp2\\2 + \\1p1 .V>i + A</>2) > 0 .2 Hilbert Spaces 43 where the asterix * means complex conjugation.^ 2 ) =   ^   2 + 2 ^ i .*h)=\\AH\(38) (3. ^ ) + A 2 (V2.2. ^ i ) + A ( ^ i . WA+Ml2 = ll^il 2 + ll^2 2 . for ip\. »2 2 . In order to verify Eq. antilinearity in the first component (also described as sesquilinearity. which we can write 0 < ( ^ i .rh) (3.V') +"2*(V.^) = ai*(V'i. meaning oneandahalffold linearity).9a) (3.9a) we start from if) = ipi + \if>2 6 M. The distance between two vectors if>i. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).9b) = 0 (Pythagoras theorem). (3. ^ 2 )  .)> (36) i. (3. (3.^2)1 < H^ill • 11^211 (Schwarz inequality). linearity in the second component.9d)   V i   = sup  ( ^ i . for arbitrary A and ip2 ^ 0: (V>i + AV>2. (3.is defined by d(ipi.. ^ i ) + A * ( ^ . (310) (^2. The first two properties imply ( a i ^ i + a2ip2.V. if Wi..e.
i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. beginning with A = 1 in the second line of Eq. (3.44 or CHAPTER 3.ii) thus verifying Eq.11).9c).* Two vectors i f i . a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. (3.9d). for which CO 2 := V^l^l 2 < 00. (3.V2) IIV'l2 + V'22 + 21(^2.12) We omit here the verification of the remaining relations (3.^)1 <IIV>il. so that   ^ l + ^ 2   <   ^ l   + ^2. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . (3.9e). HIHI (3. .9a). (3. ll^ill + llVdl. In verifying the triangle inequality we use this result (3.\\ih\\. Mathematical Foundations of Quantum Mechanics 1(^2. hence verification in each case is necessary. ^ e M are said to be orthogonal if (</>!. o ^ V i = o.13) the vector is said to be normalized. (3. Examples of metric vector spaces: (1) Let M.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. V 2 ) = 0 .^1)1 Vi 2 + ^2 2 + 2   ^   . be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi.  ^ l l = (HlMI +   « 2 .
5b). are combined to an equivalence class [/] (with space L2). */ x } = / /o for x = 0. (2) Let M.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). (3. which differ solely on a set of measure zero. In order to avoid this difficulty. •'^ \ 0 otherwise. 2 and so on. Then L2 is the space of all these equivalence classes.5c). etc. all squareintegrable functions / which are "almost everywhere equal".5a). is defined by ([/]. (3.3. = C? be the set of all complexvalued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 /(x) 2 d 3 x < oo. and one defines addition and multiplication by complex numbers with respect to these classes.w) :=J2v*Wi.e. which satisfies relations (3. Elements of the classes are then called representatives of these classes. With the scalar product (f.e. i.5c)) / ( x ) = 0 = > ( / . The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. / ) = 0. .g)= [ /(x)^(x)d3x. i. Js the space C? is not yet a metric vector space although for (cf. for which the scalar product. with inner product oo 45 (v. (3. But this applies also in the case of any function which is nonzero only on a set of measure zero. Eq.[$]):= / JM /(x)* 5 (x)d 3 *.
46 and
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
ll[/]=0=»[/] = [0],
where [0] is defined as the class of all functions which are almost everywhere zero. This means that functions that differ only on a pointset of Lebesgue measure zero are looked at as identical. Unless necessary, we ignore in the following mostly for simplicity the distinction between C2 and L2. Convergence of sequences in Hilbert space is then called convergence "almost everywhere". With the help of the concept of a norm we can introduce the concepts of convergence and point of accumulation. Definition: A sequence {tpn} € M. is said to converge (strongly) towards tp E M., if the distance \\tp — ipn\\ tends towards zero, i.e. lim   ^  V n   = 0.
n—•oo
(3.14)
The vector ip is then called point of accumulation. The point of accumulation does not have to be an element of A4. If M. contains all of its points of accumulation, the set M is said to be closed. A normalized vector space M. which with every convergent sequence contains a vector towards which the sequence converges, is said to be complete, i.e. if ipn e M. with lim \\ipn  i/jm\\ = 0
m,n—>oo
(called Cauchy sequence), there is a ip 6 M. with ip = lim ipn,
n—>oo
i.e.
lim \\ip — ipn\\ = 0.
n—>oo
(3.15)
Every finitedimensional vector space (on IK) is complete in the sense of the concept of convergence defined above (so that completeness does not have to be demanded separately). In order to see this, we consider the convergent sequence
n
<Pa = ^2CmiPi,
where i/ji,...,ipn€.A>i ras)
CaieK,
(3.16)
constitute a basis in M Then (according to Pythagon
nv«  M = i sec™  c0i)A\\ =YI i c  c^2'
2
2
n
(3i?)
3.2 Hilbert Spaces
47
a relation also known as Parseval equation. The convergence of the sequence i/ja implies the convergence of the sequence {Cai} towards a number Cj. Then for the vector
i=l
we have
ll^aV'II^ElC^Cil2,
i=i
(3.18)
i.e. that the sequence of the vectors tpa converges towards tp. We thus arrive at the definition of a Hilbert space. Definition: An infinitely dimensional, metric vector space, which is also complete with regard to (strong) convergence, is called a Hilbert space "K. The given definition of a Hilbert space is that usually given in mathematics.t In physics this is generally supplemented by the requirement that the space be separable, i.e. of a countably infinite dimensionality. Naturally Hilbert spaces with a countable basis are the simplest. We supplement the above by referring to the concept of a dense set or subset M of "K. A subset M of "K is said to be dense in "K, if to every / 6 "K there exists a sequence of vectors fn, fn < M, so that fn — f, S > i.e. fn converges strongly to / , implying that every vector / e Ji can be approximated arbitrarily precisely. We consider next some examples. Examples of Hilbert spaces: (1) The hyperspherical functions Yitm(6,(p) define a complete set of basis functions on the unit sphere. Any function f(9,(p) with
[\f(e,tp)\2dn<™
can be written as a convergent series
oo 1=0 /
f{9, ip) = Y, E
m=l
C
l,mYl,m(0, ip).
(3.19)
For completeness we recall here the definition
(o<p)twi r(2*+iwm)!i x v„ sin ™ 0 (±y +m (cos2e _ lV Yl
*i,mW<P)2in
[
47r(Z + m ) !
J
e
sm
\
d 9
)
^cos V
^'
See e.g. N. I. Achieser and L. M. Glasman [3].
48 so that
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Y0fl = —=,
J Air
1
Yito = \ —cosO,
V 47T
/ 3
Yit±i =
I o
^\—e±tvswt
•
A Hilbert space contains a complete set of orthonormal basis vectors or a corresponding sequence precisely then if it is separable. (2) On the space L2(0,2n), i.e. on the space of squareintegrable functions on the interval [0, 2TT], an orthonormal system, called the trigonometric system, is defined by the functions
1
einx,
±n = 0 , 1 , 2 , 3 , . . . .
(3.20)
/2TT
(3) On the space L2(a,b), where (a, b) is an arbitrary but finite interval, a complete but not orthonormal system is given by
In order to obtain the orthonormalized system, one employs the orthogonalization procedure of E. Schmidt.* The sequence of polynomials thus obtained consists of the Legendre polynomials which are defined on the interval — 1 < x < 1. These polynomials are defined as follows: Po(*) = l, *>„(*) = _ 1 dn —(x2ir, n = l,2,.... (3.21)
These polynomials satisfy the following normalization conditions, i.e. are orthogonal but are not normalized to 1:
l
/_
Pm(x)Pn{x)dx
l
= 0, (m ^ n),
(3.22a)
and
J' [Pn{x)\2dx = 1 ^
l
.
(3.22b)
(4) By orthogonalization of the following functions
x2/2
xe~x 2
/2
x2px
2
/2
one obtains the following functions defined on the space L2{—oo,oo): ^[x) = (  l ) n e * 2 / 2 f ^ e  * 2 = Hn{x)ex2'2 dxn (3.23a)
''This procedure is wellknown in analysis, and hence will not be elaborated on here.
3.3 Operators in Hilbert Space with
/*C5tJ
49
j —(
bn(x)<t>m(x) = 2nn\yft6nm,
(3.23b)
where Hn (x) is the Hermite polynomial of the nth degree (which we do not go into in more detail at this stage).§
3.3
Operators in Hilbert Space
Having defined the admissible states of a physical system as the vectors which span a Hilbert space, the next step is to introduce quantities representing operations in this space. This is our objective in this section. We begin with a number of definitions. Definition: Let T>A, the domain of definition, be a (dense) subspace of the Hilbert space "K. Then one defines as a linear operator A on "K the mapping A : VA > H with (a,/3 G C, ^ , ^ e ^ c M ) A(cnl)i + 0rfo) = a(A^i) + /3(At/>2). (3.24b) (3.24a)
Definition: One defines as norm (i.e. operator norm) of the operator A the quantity sup l i M . (3.25) ipevA\{o} WW Definition: An operator A is said to be bounded, if its norm is finite, i.e. \\A\\ < oo. Definition: Two operators A : T>A — "K and B : T>B —>"Kare said to be > equal if and only if Atp = Btp, for every ip G VA and Z>A = T>BExample: An example of a linear operator is given by the differential operator D:= —
d
A:=
_.:.LU ^ f./. _ T1 ^ rwith VD = iil>eL*,jeL*>.
T2
(3.26)
Definition: We define the operations of addition and multiplication of operators by the relations (A + B)tp:=Aip
s
+ Bip, \/i/j<EVA+B = VAnVB,
(3.27a)
Hermite polynomials are dealt with in detail in Chapter 6.
50
CHAPTER 3. Mathematical Foundations of Quantum Mechanics (AB)ip := A(Bip),
V^GPB,
B^eVA.
(3.27b)
Definition: We define operators called commutators as follows: Let A : T>A —> "K, and B : T>B — "K be linear operators in the Hilbert space "K. Then we > define as commutator the expression [A, B] := AB  BA with V[A,B\ = D A n £>B » ft. (3.28)
Definition: If .A : X^ — IK for i/> 6 T>A\{0}, then ^ is called eigenvector with • respect to the eigenvalue A € C if and only if AV> = \i/>. (3.29)
Very important for our purposes are the concepts of adjoint and selfadjoint operators. Definition: Let A : T>A — IK and ip € P A C IK. Then A^ is called adjoint » operator of A if for A^4> := <jj, <f> € T>Aj, the following relation holds: {A* 4,il>) = (<!>, Ax/,). (3.30)
Definition: The operator A : X>A — ft, is said to be symmetric if and only if >
0 M ^ ) = (MV), V ^ i e P
A
c%
(3.31)
Definition: The operator A, A : X^ — IK, is said to be hermitian or selfad> joint if and only if A = Af, One can verify that: (A*)* = A, (A + S j ^ A t + (AA)+ = A*A , (AB)t = S U * , (A" 1 )* = (A*)" 1 .
f
i.e. D A = D A t
and A ^ = A</>.
(3.32)
(3.33a) fit, (3.33b) (3.33c) (3.33d) (3.33e)
We are now in a position to construct relations between operators A and B in K, which correspond to the relations (2.5a) to (2.5e) of Poisson brackets, however, we omit their verification here: [A,B] = [B,A], (3.34a)
3.3 Operators in Hilbert Space [A, aiBi + a2B2] = ax[A, B{\ + a2[A, B2], [A,B]* = [A\B\ [A,BC] = [A,B}C + B[A,C], [A, [5, C]] + [B, [C, A}} + [C, [A, B}} = 0.
51 (3.34b) (3.34c) (3.34d) (3.34e)
The last relation is again called a Jacobi identity. Comparison of Eq. (3.34c) with Eq. (2.5c) requires here in the definition of the corresponding quantity the introduction of a factor "i" (see Eq. (2.47)). As important examples we consider the following operators. (1) q3 : Vqj  L 2 (R 3 ). We write sometimes the application of the operator qj to <f> G L2(IR3): (qj4>){x), and we define M)(x) :=arj0(x). (3.35) We can read this equation as an eigenvalue equation: Operator qj applied to the vector <f> yields the eigenvalue Xj multiplied by <> in the present case on /, 3 R . Since Vqj = {<£ G L 2 (R 3 ) : qj* G L 2 (R 3 )}, we have
^ = ^ t 
Furthermore, for instance for ip, <p G L2(IR1), we have (ip,q<p) = / ip*(x)(q(f))(x)dx = / ip* (x)x<p(x)dx
=
(#,«•
(3.36)
Since for the adjoint operator A^ of A:
it follows (with Pg = V t from above) that qj — q3^. (2) P j : VPj > L 2 (R 3 ) defined by (p 3 »(x) := ih^<t>{y). In this case we have VPj = <<t>€ L 2 (R 3 ) : c continuous, ~ G L 2 (R 3 ) I = / > Vpj]. (3.37)
52
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Here for ip,(f> e L2(Ul): (V,W>) = /V(aO(p0)dx
(3
=7)ifi
f^*(x)^(f>(x)dx
= inr(x)<i>(x)\f00J^r(x)<i>(x)dx
— — I [ ih—i/}(x) \ (f)(x)dx = J (pip)*(x)<f>(x)dx = (p^,0) = (pty,0), (3.38)
so that p* = p. Something similar applies in the case of the following operator which represents classically the kinetic energy T: (3) T:VT^ L 2 (R 3 ) and
™ ( x ) := "£ A<Kx) = ( ^ & 2 V ( x ) '
As a further example we consider the commutator. (4) Let the commutator be the mapping
(339)
Then for <p E L 2 (R 3 ): \Pj,Qk]<l>(x) = i.e. formally The following commutators which define the Heisenberg algebra \Pj, Qk] = ihSjk, \pj ,Pk] = 0, [qj ,qk] = 0 = (pjqk ~ 9fePj)^(x) = (pjqk(t>)(x)  (qkPj(t>)(x)  i ^ f ^ X f c 0 ( x )  xfe—</>(x) J = ih6jk(j)(x.),
are called canonical quantization conditions with respect to a theory whose classical version possesses the fundamental Poisson brackets {Pi, ?*} = Sjk, {Pj,Pk} = 0, {qj,qk} = 0.
The simplest example to consider is the harmonic oscillator. We postpone this till later (Chapter 6). We add, that the quantization must always be
3.4 Linear Functionals and Distributions
53
carried out on Cartesian coordinates. Moreover, the above relations assume that the three degrees of freedom are independent, i.e. there are no constraints linking them. Systems with constraints can be handled, but require a separate treatment.^
3.4
Linear Functionals and Distributions
We now introduce the concept of a continuous linear functional on a socalled test function spaced Our aim is here, to provide a frame in which the formal Dirac bra and ketformalism to be developed later finds its mathematical justification. We require in particular the delta distribution and Fourier transformations. A subset of a Hilbert space "K is called a linear manifold D, if along with any two elements (f>i,4>2 G ^ C 'K this also contains the linear combination of these, i.e. « i ^ i + a2<^2 £ ^ , ai,a2€C. (341) A linear functional [/] in the Hilbert space "K is a mapping of the manifold T> into the set of complex numbers, i.e. [/] : V  C and is written [/]<<£> = / < 0 ) : = / with the property of linearity, i.e. f(<l>i + <h) = f(<l>i) + f{<h), (i.e. the expression (3.42) is antilinear in the first component). If / ( x ) is for instance a locally integrable function (i.e. for a compact set e R n ) like exp(ik • x), then it is clear that the function </>(x) has to decrease sufficiently fast at infinity, so that the expression in Eq. (3.42) exists. Functions which provide this are called test functions (see also later). Instead of the Hilbert space we therefore consider now a space of test functions (vector space of test functions), and on this space linear functionals. Definition: The compact support of a continuous function <fi : Rn — C is > defined to be the compact (i.e. closed and bounded) set of points outside
f
/(x)0(r)dx,
(V<P€V)
(3.42)
S e e Chapter 27. "We follow here to some extent W. Giittinger [127].
54
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
that of (ft = 0. Test functions cf> with compact support are exactly zero outside their support; they define the space D(Rn). A different class of test functions <j> consists of those which together with all of their derivatives \Dn(f>\ fall off at infinity faster than any inverse power of x. These test functions are called "rapidly decreasing test functions" and constitute the space S(Rn): D(Rn) S(Rn) := {(f> G C°°(DRn > C) : support of </> compact}, := {4> £ C°°(Rn > C) : \x\m\Dn(j>\ bounded, m,n,... e I > 0}. N (3.43) Definition: Distributions f{<f>) are defined to be the linear functionals on D(Rn) and tempered distributions the linear functionals on S'(IRn). A subset of distributions can obviously be identified with ordinary functions, which is the reason why distributions are also called "generalized functions".
3.4.1
Interpretation of distributions in physics
It is possible to attribute a physical meaning to a functional of the form
f (</>):= J dxf(x)<j>(x).
(3.44)
In order to perform a measurement at some object, one observes and hence measures the reaction of this object to some tests. If we describe the object by its density distribution f(x), like e.g. mass, and that of its testing object by <t>(x), then the product f(x)(p(x) describes the result of the testing procedure at a point x, since f(x)4>(x) = 0, provided f(x) ^ 0 and <p(x) ^ 0, i.e. if object and testing object do not meet. The expression then describes the result of the testing procedure in the entire space. If we perform the testing procedure with different testing objects and hence with different test functions <pi(x),i = 1,2,..., we obtain as a result for the entire space a set of different numbers which correspond to the individual 4>i{x). These ^dependent numbers are written as in Eq. (3.44):
m = j f{x)<t>{x)dx.
If f(<j>) = 0 for every continuously differentiable function <f>{x), then f{x) — 0. In general one expects that a knowledge of the numbers f{<j>) and the test
3.4 Linear Functionals and Distributions
55
functions <p(x) permits one to characterize the function f(x) itself, provided the set of test functions is complete. In this way one arrives at a new concept of functions: Instead of its values y = f{x) the function / is now determined by its action on all the test functions <f>(x). One refers to this as a functional: The functional / associates a number /(</>) with every test function (f>. Thus the functional is the mapping of a space of functions into the space of numbers. f{4>) is the value of the functional at the "point" <f>. With this concept of a functional we can define quantities which are not functions in the sense of classical analysis. As an example we consider in the following the socalled "delta distribution".
3.4.2
Properties of functionals and the delta distribution
The delta distribution is defined as the functional 5(<j)) which associates with every test function <fi(x) a number, in this case the value of the test function at x = 0, i.e. 6(4>) = </>(0), (3.45a) where according to Eq. (3.44) 6((j)) = f 8(x)(/)(x)dx. (3.45b)
The result of the action of the "delta function" 5(x) on the test function 4>(x) is the number 0(0). The notation J 5{x)(j>(x)dx is to be understood only symbolically. The example of the delta function shows that a function does not have to be given in order to allow the definition of a functional. In order to insure that in the transition from a function f(x) defined in the classical sense to its corresponding functional /(</>) no information about / is lost, i.e. to insure that f(4>) is equivalent to f(x), the class of test functions must be sufficiently large. Thus, if the integral J f(x)4>(x)dx is to exist also for a function f(x) which grows with x beyond all bounds, the test functions must decrease to zero sufficiently fast for large x, exactly how fast depending on the given physical situation. In any case the space of test functions must contain those functions <f>(x) which vanish outside a closed and bounded domain, since these correspond to the possibility to measure mass distributions which are necessarily restricted to a finite domain. Furthermore, for the integral (3.44) to exist, the test functions must also possess a sufficiently regular behaviour. For these reasons one demands continuous differentiability of any arbitrary order as in the case of 5(0?") above. Certain continuity properties of the function f(x) should also be reflected in the associated functional. The reaction of a mass distribution f(x) on a test
56
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
object <p(x) is the weaker, the weaker cp(x) is. Thus it makes sense to demand that if a sequence {(pi(x)} of test functions and the sequences resulting from the latter's derivatives of arbitrary order, (p^(x), converge uniformly towards zero, that then also the sequence of numbers ((pi) converges towards zero. We refrain, however, from entering here into a deeper discussion of convergence properties in the space of test functions. The derivative of a distribution f((p), indicated by a prime, is defined by the following equation f((P):=f(cP>). (3.46) This definition suggests itself for the following reason. If we associate with the function f(x) the distribution
oo
dxf(x)</>(x),
/
oo
then the derivative f'(x) becomes the functional
oo
/
dxf(x)cP(x)
oo
= /($'),
(3.47)
as in Eq. (3.44). Partial integration of this integral then yields, in view of the conditions <p(±oo) = 0,
oo
dxf(x)<P'(x) = /(</>'),
/
oo
as in Eq. (3.47). Equation (3.46) defines the derivative of the functional f(<p) even if there is no function f(x) which defines the functional. For instance in the case of the delta distribution we have 6'(cP) = S((P') = <j>'(0) according to Eq. (3.45a). Formally one writes, of course,
oo
(3.48)
/•oo
/
oo
dx5'(x)<p(x) =
[<J(a;)0(a;)]?foo  /
J—oo
dx5(x)(P'(x) (3.49) (350) (3.51)
= 5(<P>) = <P'(Q). For an infinitely often differentiable function g(x) one has apparently (5 •/)<</>> = / ( # ) , so that (x6)((P) = S(x(P) = [x(P}x=0 = 0 x (P(0) = 0,
3.4 Linear Functional and Distributions or ' x6(x)</>(x)dx = 0,
/ '
57
x5{x) = 0.
(3.52)
Thus formally one can operate with the delta distribution or delta function in much the same way as with a function of classical analysis. As a further example we consider the relation f(x)S{x) = f(0)S(x). According to Eq. (3.50) we have f{x)5{x)4>{x)dx = «J(/0) = [f(x)4>(x)]x=0 = f(0)<f>(0) (3.53)
/
=
f(0)5(<f>) = J
f(0)6(x)<j>{x)dx,
as claimed in Eq. (3.53). Formal differentiation of the relation (3.53) yields f(x)6\x) = f(0)S'(x)  f'(x)S(x). (3.54)
One can convince oneself that this formal relation follows also from the defining equation (3.46). In particular for f(x) = x we obtain the useful relation x5'{x) = S(x). (3.55)
A very important relation for applications is the Heaviside or step function 9{x) which is defined as follows:
From Eq. (3.56) we can deduce the relation 6'{x) = 6{x). (3.57)
For the verification we associate with the step function the following functional
oo
roo
/
oo
0(x)<f>(x)dx = / <f>(x)dx.
JO
/•oo
For the derivative we have according to Eq. (3.46) e'(x)(<p) = 8(x){4f) = dxc/)'(x) = (f)(0)  0(oo) = 0(0)
= 8{x)(<i>),
58
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
or symbolically 0'(x) = 5(x), i.e. Eq. (3.57). After the introduction of the delta function it is customary to consider briefly Fourier transforms, i.e. in the case of one dimension the integral relations f{x) g(k) = =  =
V Alt
I
1
\ /
f°°
dkg(k)eikx dxf(x)eikx
=
F~lg{k), (3.58)
Joo
roo
 =
= Ff(x).
\Z2TT JOO
We assume here some familiarity with these integral relations. It is clear that the existence of these integrals assumes significant restrictions on the functions f(x),g(k). As a formal relation in the sense of the theory of distributions we deduce from Eq. (3.58) the important formal integral representation of the delta function, i.e. the relation
I
/•OO
S(x) = — /
dkeikx = S(x).
(3.59)
One can see this as follows. According to Eq. (3.58) /(0) =   L fdkg(k) yzir J and g(k) =  L y lis J fdxf(x)eikx.
Inserting the second relation into the first, we obtain
/(0) =
Jdxf(x)^jdke^,
and comparison with Eqs. (3.45a) and (3.45b) yields (3.59). We close this topic with a comment. The singular delta distribution was introduced by Dirac in 1930. The rigorous mathematical theory which justifies the formal use of Dirac's delta function was only later developed by mathematicians, in particular L. Schwartz. Thus today the singular delta distribution is written as a regular distribution, i.e. as an integral operator, by writing, for instance, 5a= f <5(x  a)^(x)dx = ^(a) V 0 e S ( [ R n ) , (3.60)
and one derives from this that the delta function <J(x) has the (impossible) properties of being (1) zero everywhere except at x = 0 where it increases so enormously that (2) the integral over <5(x) is unity:
f S(x)dx = 1.
Chapter 4
Dirac's Ket and BraFormalism
4.1 Introductory Remarks
In this chapter we introduce the (initially position or momentum space representationindependent) notation of Dirac* which is of considerable practicability. We also introduce those properties of the notation which make the calculations with states and operators amenable to simple manipulations. The notation will be used extensively in later chapters. The integral representation of the delta function discussed in Chapter 3 can be considered as a formal orthogonality relation for harmonic waves exp(ikx) which finds its rigorous justification in the context of distribution theory. Thus we have — again for simplicity here for the onedimensional case —
1
/"OO
S(x x') or 6(k k')
= — /
27T
i
dkeikxe~ikx'
(4.1a)
J_00
/oo
= — /
27T J_00
dxeikxeik'x.
(4.1b)
Since k and similarly x here assume a continuum of values, Eqs.(4.1a) and (4.1b) are described as normalization conditions of the continuum functions exp(ikx) as distinct from orthogonality conditions for functions depending on integers m,n as, for instance, the trigonometric functions um(x) = cos(mx), vm{x) = sin(mx),
'See P. A. M. Dirac [75].
59
60 for which
CHAPTER 4. Dirac's Ket and BraFormalism
1 fn  / dxum(x)un(x)
^
= 6mn,
(4.2a)
JK
i r
 /
^ J7T
dxum(a;)i;ri(x) = Smn,
(4.2b) (4.2c)
i
 \
^
r
dxum(x)vn(x) = 0.
JK
In the following we shall therefore "orthogonalize" continuum states represented by vectors of a Hilbert space which is no longer separable (i.e. which has at least a subset whose vectors are characterized by a continuous parameter) in the sense of the relation (4.1a) to a delta function instead to a Kronecker delta as in Eq. (4.2a). With this formal use of the delta function we can manipulate continuum states easily in much the same way as discrete states which implies an enormous simplification of numerous calculations, t The Fourier transforms introduced in Chapter 3 permit an additional important observation: We have several possibilities to represent vectors in Hilbert space, since the Fourier transform describes the transformation from one representation to another (" configuration" or "position space representation" <> "momentum space representation"). A position space Schrodinger wave function ?/>(x), which we shall consider in detail in numerous examples later, corresponds to the representation of the vector ip of a vector space (as representative of a state of the system under consideration) in the position space representation, i.e. as a function of coordinates x. In the momentum space represenation the vector tp is the Fouriertransformed V>(k) of ip(x); this representation will be used in particular in Chapter 13.
4.2
Ket and Bra States
In the following we introduce the notation of Dirac.* We define ketvectors as elements of a linear vector space and bravectors as those of an associated dual vector space. The syllables "bra" and "kef are those of the word "brackef. The spaces are linear vector spaces, but not necessarily separable Hilbert spaces, unless we are dealing with an entirely discrete system. More as an excercise than as a matter of necessity we recall in the following some considerations of Sec. 3.1, expressed, however, in the notation of Dirac. Hopefully this partial overlap is instructive.
'Formally this means that we have to go to an extended Hilbert space, which contains also states with infinite norm, see e.g. A. Messiah [195], Vol. I, Sec. 7.1.3. *P. A. M. Dirac [75].
4.2 Ket and Bra States
61
In order to achieve a representationindependent formulation we assign to every state of the system under consideration a vector, called ketvector and designated ), in a vector space V. In the symbol \u), for example, u is an index of the spectrum, i.e. a discrete index in the case of the discrete spectrum, or a continuous index in the case of a continuous spectrum. The linearity of the vector space implies, that if £ is a continuous index, and A(£) an arbitrary complex function, then with £),
\w) = J\(0\Odt
(43)
is also an element of V. As mentioned, the space can be finitely or infinitely dimensional. In the vector space V of ketvectors a set of basis vectors can be defined, so that every vector can be expressed as a linear combination of these basis vectors. With the vector space V of ketvectors we can associate a dual space V of bravectors, which are written (x. The bravector (% is defined by the linear function <X{«» of ketvectors \u). For a certain ketvector \u) the quantity x has the value (xit), which is, in general, a complex number. For a better understanding of the concept of the dual space, we recall first the difference between a linear operator and a linear functional. According to definition, the linear operator which acts on a ketvector \u) G V yields again a ketvector \u') G V. On the other hand, a linear functional x i s a n operation, which assigns every ketvector \u) G V linearly a complex number {x\u}, i.e. X= with x ( A i K ) + A 2 « 2 )) = A i x O i ) ) + A 2 x ( M ) The functionals defined on the ketvectors \u) G V define the vector space V called dual space of V. An element of this vector space, i.e. a functional x is symbolized by the bravector (x.§ Then (x\u) is the number that results by allowing the linear functional (x G V to act on the ketvector \u), i.e. X(\u}) = <xl«>One should compare this with our earlier definition of the functional / on the space of test functions <> We wrote this functional /. «)eV>x(l«»
m = w».
By construction every ketvector is assigned an appropriate bravector. The reverse is not true in general. See e.g. C. CohenTannoudji, B. Diu and F. Laloa [53], p. 112.
(2) (u\u) > 0.e. i.3 Linear Operators.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).9a): \{u\v)\2 < (u\u)(v\v). Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . The dimension of V is the same as the dimension of V. the norm squared. We consider next the important case that a unique antilinear relation called conjugation exists between the kets \u) E V and the bras (x E V. if x\u) = 0 V kets \u) E V.4b) (4. if (u\v) — 0. (4.\u) = (u\0*. which we write (u\.and BraFormalism Furthermore we have: ( x  = 0. This expression is linear with respect to \u) and antilinear with respect to \v). (3) (u\u) = 0. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. Dirac's Ket. V \u) (in the case of £) with infinite norm!). or (4.4a) \v) = J\(m<% and (4.(t. It follows that we can now write the Schwarzian inequality (3.5) 4.5c)). \v) E V are said to be orthogonal. that the vector \u) E V is associated with a vector in V. if \u) = 0 (see (3. Two ketvectors \u).62 CHAPTER 4. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V.
e.or direct product of vector spaces. This does not always exist. ltt) = \u). Let the following bravector be given: (x\ £ V. denned as a linear operator on V. also acts on the dual space V.e. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear).3 Linear Operators.. i.. on bravectors are similar to those on ketvectors.. Then the following rules apply: A + B = B + A. as we can see as follows. It follows that (r. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . Furthermore linearity applies. B. The operator A.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). This product is commutative. let (771 6 V be the bravector defined by this functional. The product of such vectors is written: \u^u^) = \u)^\u)^. If Vi has the dimension A/i. if AB = 1. The inverse of a product is (AB)'1 =B1A~1. Then one says: (771 results from the action of A on (x.or Kronecker. i. The product vectors l?^ 1 )^ 2 )) span a new vector space. AB^BA. The inverse of A is written A . 1 and \u) = B\v). Two operators A and B are said to be inverse to each other.4. defines the unit or identity operator. and one writes: (V\ = (X\A. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. BA = 1. the space Vi <S> V2 has the dimension . Multiplication by unity. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. if \v) — A\u) i. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). Finally we define the tensor.e. Furthermore. the space Vi <S> V2. In this way the operations of operators A.
e.3).7b) (4. Every operator A^> commutes with every operator A(2\ i. \t) G V. we obtain the conjugate relation <tAt«> = (u\A\t)* (4.e. Dirac's Ket.7d) .A^] = 0.7C) («)(u)t = \v){u\. here designated with the same symbol. i. [A^. then \v) is a linear function of \u). 3. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. Since we demanded the scalar products to have the property (t\v) = (v\ty. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. As a consequence we arrive at the following properties (e. i. AW\U)W = \V)W. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). also Sec.6) for all \u). Let A^> be an operator in the space Vi and A^ space V2. AW\uW)\uW) = \vWUW). (v\t) = (u\A\t). we write \v) = A*\u). i. Every such operator is then also associated with an operator. and (v\ = (u\A. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.7a) (4. in the product space.and BraFormalism an operator in the Mi A/2.g.64 CHAPTER 4. If (v\ = (u\A. Next we construct the following scalar products: (t\v) = (t\Ai\u).e. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4.e.
Verification: Since A — A^ and A\u) = a\u). it follows that (u\A\u) = a(u\u). and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. It follows that a is real. [H. K] = 0. i.K]^ = = [HKKH]* KHHK = = K^H^ [H. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. The operator H is said to be positive definite. so that (n^4n) is real. Obviously the quantity \a)(a\ is an hermitian operator. . for we had (\u){v\)i = \v){u\. An hermitian operator A has the properties (a) that all its eigenvalues are real. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. An important theorem is the following.e. so that (a)(a)t = a)(a. or (u\A\u)* — {u\A^\u) = (u\A\u). i. Analogous considerations apply in the case of bravectors (u. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H.4.3 Linear Operators. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . if H is hermitian and (u\H\u) > 0 for all ketvectors \u). if H — H' and antihermitian. as is also (u\u). but only if these commutators commute. hermitian The product of two hermitian operators is not necessarily hermitian. The commutator of two hermitian operators is antihermitian: [H.e. 65 The linear operator H is called hermitian.K\.K}.
The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). Ps\ua. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. Dirac's Ket.e. since a ^ b.e.e. and let S* be its complement. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. so that (u\Ps = (us\. (u\Ps = {us\. we have 0 = (uArt) — (u. {us\us*) = 0. (v\A = b(v\. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). since a is real. i. Next we introduce the very useful concept of projection operators. which includes vectors of infinite norm. If these form a complete system. \u8*) E S*. as we can see as follows. (4. Ps2 = Ps(4. i.9b) (4.66 CHAPTER 4. i. Ps is idempotent.and BraFormalism Moreover. ( n  z / ) = 0 . a ^ b. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.Au) = a(v\u) — b(v\u) — (a — b)(v\u).u'). A continuous spectrum can immediately be included by passing to the extended Hilbert space. so that (v\u) = 0. (u\u') = 5{v . but the continuum vectors are normalized to a delta function. The above theorem remains unchanged.9a) . i. Let S be a subspace of the Hilbert space !K. then the hermitian operator defined on this space is an observable. The entire set of ketvectors spans the extended Hilbert space.) = 0. Assuming A\u)=a\u). (n\n')=5nnl.8) The projection operator has the following important properties: Ps is hermitian.e.
i. 67 (4. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. = (p2p)\p). P\p) =p\p). Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). 2). Set \u) = \ua) + \ua*) with (a\ua*) = 0. so that \ua) = (a\u)\a) = \a)(a\u). where £ is a continuous index. iV) is a set of orthonormalized states. The quantity \a)(a\ is called elementary projector. Ps is projector onto S. \u) arbitrary. .9d) Thus the vectors P\u). Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. (1 — Ps) is projector onto S*. i. (4. Very similarly we can construct operators which project onto the subspace spanned by the continuum states..e. (1 — P)\u) are orthogonal. Let a continuum state be written £). p 2 . i.e. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. if the set of vectors 1).. \ua) = c\a).1. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). c = (a\u).4. as can be seen as follows.. i. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c.p = 0.e. Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. Let p be an eigenvalue of the projection operator P. .3 Linear Operators. Then 0 = (P2P)\p) and hence p = 0. so that (Ps2 ~ Ps)\u) = 0.e.9c) This property can be seen as follows.
. in this case the spectrum also has a continuous part. See e.6. Coulomb potential. i. However. Observables are representatives of measurable quantities. Let \u\). see Eq. one says the degree of degeneracy is r . In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. which are orthogonal to each other (i. Degeneracy will be discussed at various points in this text.1 and 11.e... are linearly independent).6. be a system of basis vectors in this space. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. u2). that one and the same eigenvalue Aj is associated with several eigenfunctions.r\ "An example is provided by the case of the hydrogen atom.r)(ui. which we introduced earlier.114b). (11.g. Dime's Ket. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. P = £n)(n. 8. Examples 8. the operator P is correspondingly P2\u)= f2\0m\u). In general it is possible.114c).4 Observables Operators which play a particular role in quantum mechanics are those called observables.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.e. and Eq. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator)..68 CHAPTER 4. Sec. (11. 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. If there are r such vectors. 4.
which we do not enter into here. The uniqueness of the expression (4. we obtain i i i i i. (4.e.11) i. in the case of degeneracy with (ui.r\u) i.\i)Pi = 0.10) This expression is known as completeness relation or closure relation.r =^ i.r) (the convergence.4.r \(ui. ^ A i P i = A = ^Aiui. the operator A is completely determined by specification of the eigenvalues A.e. is always implied). and the eigenvectors \ui. If Aj ^ Aj'. or also as subdivision of unity or of the unit operator.e.r'} = Sii>6rr>. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui. (4. PA = y£pi i = Yt\ui. Applied to an arbitrary vector \u) G "K Eq. i i (4. i.r)(ui.r\u)\2. and APi = XiPi.r (4. i.r)(ui.10) follows from the fact that i Applying the operator A to the projector (4.10) expresses the completeness of the orthonormal system.10) gives the linear combination it) = ^ i. 69 i If A is an observable and if the spectrum is purely discrete.r\ = l. i.12) .r \ui. the relation (4.e.r\ui/.r\u). the projection of this operator is the entire space. then PiPi' = 0.r>(ui. Together with the orthogonality condition.r.10). The operators Pi are linearly independent. Let us set {A .4 Observables The dimension of this subspace is that of the degree of degeneracy.r)(ui.
(4. 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. written in Dirac's notation. (u„zv) = 8{v . for instance exponential functions.13) For an arbitrary vector \u) of the appropriately extended Hilbert space we then have v—<• fV2 \u) = PA\U) = 22 \ui)(ui\u) + / and hence i) = (u\PA\u) = J2 \(ui\u)\2 and A = APA = ) J dv\uv){uv\u). The ketvectors \uv) are orthonormalized to a delta function.and BraFormalism This is the expression called Parseval equation which we encountered with Eq. (3. i.e.17) previously. (4. i. In a similar way we can handle functions f(A) of an observable A. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv.70 CHAPTER 4.v'). Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). now however. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. the eigenvector \ui): f(A)\ui) = f{\i)\ui).14) . for instance. Dime's Ket. we have the corresponding generalizations. Let v be the continuous parameter which characterizes the continuum. If the spectrum of an observable A consists of discrete as well as continuous parts. One defines as action of f(A) on. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1.e. as explained earlier.
Therefore we want to reexpress the Fourier transform (3.15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx. x) € Fx.. [A. Since both expressions (4.4.15) and (4.5 Representation Spaces and Basis Vectors 71 This relation expresses an arbitrary function f(A) of an observable A in terms of the eigenfunctions of this operator. again later. For many practical purposes the use of the Fourier transform is unavoidable. as a formal orthonormality condition. for which the completeness relation is f dk\k)(k\=l. if and only if they possess at least one common system of basis vectors. / dx\x){x\ = 1. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies).. \k)€Fk.16) represent subdi . In the onedimensional case we write {x\x') = S(xx'). Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. if (a) they all commute pairwise.e. here presented without proof.and bravectors. i. Eq.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. (4.58) in terms of ket.17) The Fourier transform provides the transition from one representation and basis to the other. We shall return to this theorem. form a complete set of commuting observables. where the symbol ip is already indicative of the Schrodinger wave function ij.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^.C. their commutator vanishes. In the following we consider more generally ketvectors \ip) as representatives of the physical states. with energy E.B.. Extending this we can say: A sequence of observables A. i. First of all we can rewrite the integral representation of the delta function.. and (b) if they possess a uniquely determined system of basis vectors. (4. (4.59). which all commute with one another.e. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. B]. 4. (3.
Rather \x) and \k) serve as basis vectors in the representation spaces Fx.72 CHAPTER 4. which are representatives of the states of our physical system. (x\k) = ^=eikx. il>{x) := (x\i/}) : "K * C. we can rewrite Eq. Dime's Ket.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').and BraFormalism visions of the unit operator. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). i.1a).2a) etc. (4. this expression has to be identified with — / dkeikxeikx\ i. 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 (4.18) According to Eq.Fk. <a#) = [ dk{x\k)(k\ip).e.space representation. called wave function. Obviously we obtain this by inserting a complete system of basis vectors of Fk.21) provides the ketvector \ij)) in the A. (4. (4. (3. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. . V 27T (k\x') = .e.19) Comparison with the orthonormalized system of trigonometric functions (4.59) or Eq. (4. ${k) := (k\if>).e. 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). (4.15): 5(x . i.
Chapter 5 Schrodinger Equation and Liouville Equation 5. 5.p. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation.t). This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. with the complementary system or *E. 1. i. Schrodinger [244]. Thus we distinguish between socalled pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator. 73 . 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). We mentioned earlier (in Sec.2 The Density Matrix We shall establish a matrix p.e. which is the quantum mechanical analogue of the classical probability density p(q. the calculation of which for specific cases is the subject of Chapter 7. if we do not take into account the complementary part of the universe.5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe.1 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.
The actual and real state of a system is then a superposition of two types of states.i. i. \i). of an operator that acts only on the states of the system we are interested in.3b) The quantum mechanical expectation value of an observable A.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a.. Schrodinger Equation and Liouville Equation the measuring instruments.i..j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. The states \i) £ "K. we have to deal with a socalled "mixed state".e.e. (a\i)=0 = {j\b)... we have (a\b) = 5ab. p" ( C C * ) t = CTC* = p. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system.3a) and N J  a ) ( o  = l in the complementary subspace. where T means transpose: p = CTC*. where i represents collectively all quantum numbers of the state of the system under consideration. are called "pure states". (52) (5. but also of the states of the complementary set of the universe. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe.. \a).^ Since p is hermitian.b.. a (5. \a). T . which in Dirac's notation we can write as = J]C a »i) a. These are the quantum mechanical states which are also referred to as microstates.74 CHAPTER 5. i.. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. 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. that is.b. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij.
.5. In the following we write simply \i) instead of \i').) = YJc t .j(x'i)(ia.2 The Density Matrix 75 \i') 6 "K.* We can specify the state of our system.\p\i")(i"\ = Y. This transition to a new set of basis vectors in which the matrix becomes diagonal. In this sense we have p(x'. which are all subjected to the same macroscopic boundary and subsidiary conditions. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). P ^ ) = i)(#) = (#)i) (5.10) and is therefore..e. i i (511) In Eq.7) p = ^2\ii)(i.3a). in a specific representation. where x represents collectively the entire set of position coordinates of the (particle) system. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i). a projection operator.e.9) is diagonal. If the ensemble corresponding to the system under consideration consists of Z elements (i. i.2) and (5. as we discussed previously. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.ji(x / )i*(x).. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. which today is a fundamental term in statistical mechanics.^'\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.e. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.x) := (x'\p\x) = yju. seen macroscopically of Z copies of the system). .3 with Pi^YtCaiCaj*. i. but occupy in general different microscopic. This ensemble is interpreted as a large set of identical systems. (5. in the position state representation or xrepresentation. a complete orthonormal system with properties (5. Thus the real system is in the pure state \i) with probability uii/ J2i Ui. or (5. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. e. (5.6) (i'\p\i") = Wi'Si'vi = real.g. quantum states. (see below) eM. (5.e. as we saw earlier. Gibbs introduced the concept of ensemble.4) we had the expression (A) = Y/Pij(j\A\i) i.
In particular for A = 1. (A. Then the expectation value of the observable A is (cf. Eq.5) above) (i\p\j)=Pij. (5.14) which we can write (A) = Tr(pA). (5.18) Substituting this into Eq. we have Tr(p) = l.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). 3 (5.3 = pvv = uv. (5.16) (5. i t (5. Eq.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. according to Eq.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.17) Hence the sum of the classical probabilities is 1. (5.4)) (5. (5. We can also show that Ui > 0. as claimed by Eq. To this end we set A —>• Ai' := \i')(i'\ (no summation over i').4). Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. (5.18). Thus Tr(j>ii)<i)=l.76 CHAPTER 5.) := (mm = (#') W> = l(#')2 > 0. Hence Wj/ > 0. (5. On the other hand. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .
the system is in a pure state.22) In the following (x\i) is always to be understood as (x\i(t)}. and in the other states with probability zero.24) s (x\x') = 5{x . We see this more clearly by going to the position space representation.t) = \^{x. one says. (5. The effect of the interaction of the system under consideration with the surroundings.23) Instead of i(x.e. Thus we have a system in a pure state.e.§ {\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. Equation (5. i. We now consider the latter expression. (5.3 The Probability Density p(x. (5. We observed above. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. not in a mixture of states. If all ui but one are nought. defines the statistical operator.t)\2. since they do not take into account the (interaction with the) rest of the universe.t) we now write tp(x. (Question: Is the state of the universe a pure state?) 5.3 The Probability Density 77 We see therefore that p defines the mixed states.t). the only remaining one is 1. i. but we see here. . t) p= i The expression (5. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. In the following we are mostly concerned with considerations of systems in pure states. which is contained in the coefficients Cai. that these are actually not sufficiently general. in all other cases the state of the system is a mixed state. that — different from u>i — P. represents a quantum mechanical probability.x'). by representing the vector \i) by the wave function i(x.e. ^2ui\i){i\. so that p{x.t) = {x\i). For the case n we have TJ=J (520) (5. thus enters these states.20) expresses that the system can be found with probability 1 in the state \i). i.21) p=\i){i\=Pi.5.8).
(5. can be found with probability 1 in the space R1. which is chosen in such a way that the desired analogy is obtained. or the wave function tp(x.27) is known as the Schrodinger equation. We now want to obtain the quantum mechanical analogue. . Equation (5.e. t).22) can also be written (xi) 2 = (i\x)(x\i). This differentiation requires the time derivative of a state vector \i) E "K. 2.3. (5. we obtain ih % = ^ E ^ w o i = ^E^ij)oi+^Ewi(^iJ))oi dt +ihYJ^\J){jt{J\) (529) "More precisely this equation should be called Schrodinger equation of motion (since H is fixed. The generalization to a threedimensional position space is selfevident: JK 1 / p(x. (5. and hence the integral over all space dxp(x. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i.27) where H is the Hamilton operator of the system. We shall convince ourselves later that the postulated equation is sensible (in fact.8) and multiplying by ih.t)\2.28) Differentiating Eq. (5. the Schrodinger equation is always postulated in some way.t) = / dx{i\x)(x\i) = (i\i) = 1.78 CHAPTER 5. To this end we have to differentiate the density matrix p. (5. the expression (5. We therefore postulate first an equation for the time dependence of a state vector.e.t) = \rp(x. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).8). Schrodinger Equation and Liouville Equation The relation (5.4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec. (5. it is not "derived").^ The equation which is the conjugate of Eq. i.26) 5. with respect to time. but the linear operator states are moving — even if the system is observably stationary). Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).25) J Thus the particle whose state is described by \i).27) is iHJ\ jt= m .
not the partial derivative. (5.24)). (3.21).37)) (5. (5.p}. (5. We can now consider the Schrodinger equation (5.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.Xi)\j). (5. p also contains "quantum mechanical probabilities" (i.28) we obtain (see also Eq. (5.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H. i. (2. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. We thus obtain the operator form of the Liouville equation. With Eqs.32) with its classical counterpart (2. (5. both uji and i)(i). Eq.e.e. such a formulation is described as Schrodinger picture.27) and (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.32) Comparing Eq. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (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). the relation ih^ = [H. The reason for this is that in addition to the "classical probabilities". we observe that here on the left hand side we have the total time derivative. i.5.34) .p} + ihJ2^\J)(J\.27).e.28) the states are considered as timedependent. dp/dt ^ 0. (cf. Eq. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf.
with respect to time t. t ) . we shall describe as "time development ator given by U(t. operator" the exponentiated oper .0): where H(ih—. the state \i) is still timedependent. 10. a (5.36) in in The initial wave function or timeindependent wave function ^(x.80 CHAPTER 5. appearing in the Boltzmann distribution..t) = Hld . „ / ith—. which we shall need later. With the help of Eq. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. In expression (5. however. i. in p = ^2i^i\i}(i\. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. or H\En) = En\En).36) however. P. we can replace these states by corresponding timeindependent states. it is plausible to refer to this equation at this stage.e.. i. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H.e. .t): > ih—</>(x.1 Evaluation of t h e d e n s i t y m a t r i x As a sideremark with regard to statistical mechanics we can make an interesting observation at this point. **R. (5. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. but with respect to the parameter /3 = 1/kT. n (537) This equation is described as the timeindependent Schrodinger equation.x 1 ^ ( x . since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.35) can be written'! ip(x.4. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not.. 0) or \il))t = \il>)t=o exp (5. (5.t) = exp ip(x.4. T meaning temperature. since the solution of Eq.** In view of the close connection between time. Feynman [94].Xi){x\En) = En(x\En). in Sec.38) (x\j)t=o = Yt(x\E^(En\J)t=o.8) for p. (5.and temperaturedependent Green's functions.35) Here we can look at the Hamilton operator as a "time generator3''. 5. we write V>(x. to) = exp[—iH(t — to)/H\.
5.1). (5. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.4 Schrodinger Equation and Liouville Equation 81 Example 5.45) since on the one hand •PEi i i i and on the other hand . i ( 5 . we obtain p(x. i. Eq.e.x')^p(x.^ n tne position space representation Eq.44) r (TrePH)' (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. or ^ = ^ e 0Ei _m.42) Inserting here Eq. (5. (5. (5.x'(3) = Si e ~^W*V). x') = ^ i LUiMxWix').40).40) 1. (543) Since H(f>i(x) = Ei<f>i(x).9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. we can rewrite p. (1.41 ) or with 4>i(x) := (x\Ei) this is p(x. 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'). i (5.
Example 5.4> = ^ g ^ . (7. we obtain = HxPN{x. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5.35).52) Hence this expectation value can now be evaluated with a knowledge of pN.45). Schrodinger Equation and Liouville Equation so that p is given by Eq.49) In the position or configuration space representation Eq.tt tt For a more extensive discussion see R.L [B.x'.50) Differentiating this equation with respect to j3.(3)./3).46) is pN(x. Differentiating Eq.47) with respect to (3. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).46) we can write the expectation value of an observable <^Tr<„.x.x'.55)).1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.45) without normalization as PN(J3) := e~pH. Equation (5. (5.(3) := (x\pN(P)\x') = (x\e~0H\x'). (5.45) and (5. We now rewrite the factor exp(—(3H) in Eq. M.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.47) (5. (5. (5. (3). whose calculation for specific cases is a topic of Chapter 7 (see Eq. . Wilcox [284]. Tr(pN) = / dxpN(x. A}} + • Solution: The relation can be verified by expansion of the left hand side. [B. (5. With Eqs. (5. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A./?).82 CHAPTER 5. This solution is the Green's function or kernel K(x. (5. (5. [A.51) where the subscript x indicates that H acts on x of PN(X. (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.51). B] + L [A.x'.51) is seen to be very similar to the Schrodinger equation (5. B}} .X'. (5.48) = 1.
83 . we shall see on the way. like the Coulomb potential. * Comment of Y. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. which is often described as "second quantization" as distinct from the quantization of quantum mechanics which is correspondingly described as "first quantization" (thus one could visualize the free electromagnetic field as consisting of two mutually orthogonal oscillators at every point in space). can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory. However. (8. 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. S. can — in fact — be quantized in quite a few different ways..the present form of quantum mechanics is largely a physics of harmonic oscillators.53)). Kim and M. 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.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator.. Since the spectrum of the harmonic oscillator is entirely discrete.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". that the oscillator. E." Quite apart from this basic significance of the harmonic oscillator. This is a fundamental topic since the harmonic oscillator.51) to (8. Noz [149].
To this end we introduce first of all the dimensionless quantities rriQUi fl q. For instance we can go to the special configuration or position space representation given by q^x. h = l. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . in the present case + [EmooJ2x2)ip = 0.84 CHAPTER 6.1) for the hermitian operators q.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers.' ^mgLOh by setting and *~. However. Also observe that H does not contain terms like pq or qp. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. p^ih—.2) which for mo = 1/2. which is called Weyl ordering. This can be done in a number of ways. Quantum Mechanics of the Harmonic Oscillator 6. of which the only nontrivial one here is [q. This equation would then have to be solved as a second order differential equation. When such a term arises one has to resort to some definition. . hence we do not use a distinguishing notation. which is — in the first place — representationindependent.^ The first problem is to determine the eigenvalues and eigenfunctions of H. we can also proceed in a different way.p]=ih (6. 2mo dx \ 2 2 (6. so that there is no ambiguity due to commutation. like taking half and half.p.2 The OneDimensional Linear Oscillator In the case of the onedimensional.to = 1 reduces to the parabolic cylinder equation. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x).
e. (6. (6. C}B. C] = A[B.12) in order to obtain the following expressions: [tfA.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. Prom these relations we obtain ( ^ = ^ .1 ) . and [AB. A*} = A^A.13) ( ^ A ) i t = A\A^A From Eq.B}C + B[A.5) Reexpressing q and p in terms of A.A^] = 1. A] = [A\ A]A = A.14a) (6. if A*A\a) = a\a).8) The eigenstates of H are therefore essentially those of N := ^ f A (6.14b) [A* A.34d): [A. (6.C].7) (6. .10) then a = (a\A^A\a) = \\A\a)\\2>0. With the help of Eq. A^] = A\ (6. We now use the relation (3. (6. (6. (6. (6. i.2 The OneDimensional Linear Oscillator 85 (p. for ^ A  a ) = a\a). + 1).6) and p = y/motoh AAi =^.e.BC] = [A.6. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .11) Thus the eigenvalues are real and nonnegative. (6.1) we obtain immediately the commutation relation [A. i.14a) we deduce for an eigenvector \a) of A^A. C] + [A. A\ we obtain h A + A^ rriQLU y^ (6. q are hermitian).
(6. in view of Eq. Next we consider the vector A 2 a).11) this is not possible.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). However.17) (6. (6. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1.19) i.2)A a). Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .l)A\a) = A(A^AA . (6.A)\a) A{AAU .16) This means. unless A 2 a) = 0.20) .. we must have (a) Let \an):= Ana)^0. (6.l)A\a). since this equation implies that the eigenvalues cannot be negative. unless A^\a) = 0. A 2 a) is eigenvector of A^A with eigenvalue (a — 2).14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). unless A\a) = 0. (6. In this case we have (A*A)A2\a) (6 (6. If we continue like this and consider the vectors An\a) ^ 0 for all n. (6.e. Similarly we obtain from Eq. (6. or \\A\a)\\ = Va. but (b) An\n>\ llA An+1a)=0.l)\a) = A(a .2A)\a) 2 (a . The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. or Ata) = v ^ T T . we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).2A)\a) = A(Aa .  a) is eigenvector of A^A with eigenvalue (a + 1). Thus for a certain value of n > 0.l)\a) = (a . This would mean that for sufficiently large values of n the eigenvalue would be negative.21) .18) = '= a ) A(A^A .86 the relation CHAPTER 6.
25) (6.22) With relation (b) of Eq (6. (6.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).28) . i. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . \ I 2 2 x 87 Replacing in Eq.20) we have for a = n = 1: A2\1) = 0. that an = 0. For a — n = 0 we deduce from (b) of (6.17) a by (a — n). (6.24) This is a very important relation which we can use as definition of the state vector 0).27) According to (b) of (6. \ 9 \ I / II 4 T ) . we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6.n ) = " A»a>=» ' = A«a)F = 1 . (6.' " /.e.20) we obtain from this that the right hand side vanishes. Moreover. From relation (6. (6.I n U ... v. In the following manipulations we always use the commutator (6. (6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. The state 0) is called ground state or vacuum state.18) we obtain for a = n: p t  n )   2 = n + l. or a = n > 0. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.5) to shift operators A to the right which then acting on 0) give zero.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2. (6.6.20) that A0) = 0.n  a . I „ . But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0.
(6. Similarly we find 2)oc^Ut0). But using Eqs. as we can see as follows. (A^)3] = [A.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).^ ( O l A ^ n o ) .32) we have (nm) = .34) (6. A^} = 2A\ and [A. In general we have \n) = ^(A^n\0) (6. we have 2> = ^=dAi\0). and in general [A.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. i.31) (6.32) (6. (6. The states \n) thus defined are orthonormal.29). (0^ 2 (At) 2 0) = 2. A^A* + A*[A.5) and again (6.33) .e. According to Eq. (A*)2} = [A. (At)2]A+ + (A*)2[A. and in view of Eq. (6.11). 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. l>ocAt0).(A*)n] = n(Ai)n\ (6. (6. (6.88 and CHAPTER 6.26) the equality l) = A t 0).(6.11) we have [A.29) Hence we obtain and in view of Eq.
41) The contribution hu/2 is called zero point energy. ldnm = n\5nm. (6. in quantum mechanics. l . we obtain (n\m) = 5nm We also deduce from Eq. By repeated application of this relation on itself it follows that (since this is nonzero only for n = m) (0\An(A^)m\0) = n{n .6.1 0).1) (6. i. (6.8) we obtain therefore H\n) = hu (A^A + ].32): rf\n) = . (6.1 (A + ) m .38) (6. the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . . •(» (6. however.2 The OneDimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An1A^mA\0) + (0\An1m{A^)ml\0} 89 0 + m(0A n .36) + 1) (6. 2 . The terminology is. . we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle").37) n^)711}^) = v n  n . whose number is given by the integer eigenvalue of the number operator N = A^A. n = 0 .1) = n\n). . . Here. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.39) A^A\n) = y/nA^\n .33). .L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.34) and (6. In view of the properties (6. .35) the operators A^ and A are called respectively raising and lowering operators or also shift operators.34) A\n) = ±=A{Al)n\0) . (6.1 ) . With Eq. .e.40) (6.'n\ and = v n ![ ( A t ) M i= + (6.35) Inserting this result into Eq.J \n) = hw (n + i J \n). (6. chosen in close analogy to the postulated "second quantization relations" of field theory.
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 (6.1. (6. A^ from the relations (6. and — in fact — we have used this already previously (see e.g.'\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 .39): {n\A.. is defined by projection of operators on eigenfunctions of energy. i. n = 0. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. En = hu(n + ). (6.7) the energy representation of the operators q and p.44) yfn.43) = = y/n(n'\n — 1) = y/n5n^ni. (6. (5..38) and (6.3 The Energy Representation of the Oscillator The energy representation.39)). also called Heisenberg representation. (6.42) We can deduce the energy representation of the operators A. Quantum Mechanics of the Harmonic Oscillator 6.e.2. other elements zero.6) and (6.90 CHAPTER 6. Vn + 1.45) Correspondingly we obtain with Eqs.. other elements zero. Eq... ( At = Vi 0 0 0 ° \ f° 0 Vi 0 0 0 0 0 0 0 0 V2 0 0 0 0 0 V2 0 0 v^ 0 0 0 0 0 V3 0 V5 0 v / ( ° VI 0 V2 0 0 v / (6.46) v^ .
Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). (6. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h. (6. 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. 6. (6. (6. i.1) and (6.47) LP 0) = 0 . by (cf.e. The ground state wave function (/>o(x) is defined by Eq.3)) ^ (q 2n \ + (6. A\Q) = 0.6.32). 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. mow / (6.5) 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)}. Eq.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. 2 .
32) to apply the appropriate number of creation operators A^ to the vacuum state 0). (6.l ) n f f „ ( 0 . Vn! Now {XlA (6. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. so that <t>n(x) = ( . ) e"^2. they are called Hermite polynomials.0 = ( .56) < ^ = We have as an operator identity the relation .L ( x  ( A t ) »  0 ) .54) (6 55) i/4 rzri7r 1 / 4 vo. i. it suffices according to Eq.e.51) \l 2h {X q \ m0uP)\l 2h \X m0u.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 .92 CHAPTER 6. (6. (6.l ) > n ( x ) . These polynomials are real for real arguments and have the symmetry property # n ( . In order to obtain the wave functions of higher states.2 n n! The functions Hn(£) are obviously polynomials of the nth degree. c/>n(x) := (x\n) = .
It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. p.4 The Position Space Representation i. 81. 93 . (202"2.58b) by W. ( 2 0 4 .12(20 .en_ dt.12.2)m = (±t)m for some function VKO. Oberhettinger [181]. Magnus and F. . gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt. (6.g. d£" (6. (6.57b) • Generalizing this result and inserting it into Eq.6. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . 2£.e.58a) ' T h i s definition — apart from the usual one — is also cited e.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. (2036(20. e.
59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. 81.59b) = (l)^2/2^.(e^/2) or Hn(0 = 2n/2Hen(V2ti).He'n{£). 80. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. . £. (6. (6. (6.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. pp.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. = nHe^iO.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . (6.^ + n) He n(0 = 0 (6. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. = ffen+itO+ntfeniCO. £2l. Magnus and F. £ 4 . (6.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) .94 CHAPTER 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 £ 2 + 3. £33£.60b) J—< "See for instance W. Oberhettinger [181].
62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.^ = e^/42~n/2Hn(aV2).6. Thus the generating function can be written F(Z.. . Magnus and F.64) n\ See e. F(S. and so on.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. p.l ) ^ e^HeniO 2 / 4  .g.s) = J2 n=0 (6..63) The meaning is that if we expand F(£. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0. (6.s)=F(£.4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( . s) as a Taylor series.61) e?l*2nl2Hn{Z). 93.s) = eM?(Zs)2}(6. W. Their normalization is therefore given by OO /"OO / OO J —OO z (6. Oberhettinger [181].
Quantum Mechanics of the Harmonic Oscillator Fig.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator.96 CHAPTER 6.e. i. mow' Apx = ^m0Luh/2.e. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. 2 . has parity + 1 . In classical mechanics the energy of the oscillator can be zero. and we see t h a t even and odd states alternate.e. 6. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . We also see from the eigenvalue t h a t the energy of the lowest (i. ground) state is fko/2. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). In fact we can use the uncertainty relation AxApx ~ H/2. in order to estimate the lowest energy of the harmonic oscillator.1. 6. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. for x(t) = 0 and p(t) = mox(t) — 0. But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. i. We see t h a t the lowest state is symmetric.
2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. original path q . In the plane of complex q the integral has no poles for finite values of q. In Example 6. The kinetic energy is always positive or zero. 6. we obtain the integral . and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q)..: 27 T mov2a2. This observability can be taken as direct proof of the uncertainty relation. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. back to 0. to 0.plane Imz z . mass mo. n*h. 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)).4) to obtain by contour integration the eigenvalues of the harmonic oscillator.1/a O l/a Rez all at infinity Fig.1. in corrected version instead with n* = n + 1/2. like twoatomic molecules. 14. Thus. Example 6. classical orbit i.6.plane pole at z=0 ^ .2 The original path and the transformed path at infinity. which is the complete orbit.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. The analogous Example 6. and is directly observable. E . According to "old quantum theory" with integer n. to —a. pdq. setting q = 1/z.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. Sec. 27rmoi/ <f> V'a 2 — q2dq = n*h. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". But it has a pole at infinity.e. /classical orbit Here q goes from 0 to a.
since these two cases.1.65) f ^Va2z2 . 2 . In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. 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).5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials.e. 6.3. ..V62 a2). . for instance — without deriving all properties of the solutions.2b = Ze2.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . 1 . we start with a Laplace transform ansatz for the remaining part of the solution. harmonic oscillator and Coulomb interaction. Eq. in q at infinity).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. (11. . (m .2: Contour integration along a classical orbit Setting q = 1/z. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.5. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .) hv.2TTI a ^ = . With a2 = E.( 6 . £ = n*/w = ( n + . i. a J \d Solution: The solution proceeds as in Example 6. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). hence we consider the following differential equation in which n = 0 . / " ^ . 6.e. i.114a)). In the following we investigate in more detail the important differential equation of the harmonic oscillator.1)! (6. are the most basic and exactly solvable cases in quantum mechanics. After removal of a Gaussian or W K B exponential.1. « qdq + I2 = f dq . but here we leave this . Example 6.c2 = (I + 1/2) 2 (natural units. E = Z2e4/4(l + n+ l ) 2 (cf. z = 0 Vi 2Kmov(Ka2) = n*h = E/v. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z .5 .a) m = 2 . and obtain with this all properties of the solution.e.98 CHAPTER 6.
(this determines the limits). Eq.66) First we have to remove the £2term.t*ij& + ny{t) = O. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. (6.69) **See e. . (cf.g. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .J ±(ps(p))e*dp. (11. . and / dpf(p)pn the Mellin transform of / ( p ) . 80. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6.p t dp = \ps(p)ePt} .68) § = J p2s(p)e*dp. Oberhettinger [181]. so that the integrand of the remaining integral vanishes.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).** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. i. Then (with partial integration) *dt = ~*/Ps(p)e. (6. W. Magnus and F. (6.111)) the Mellin transform of e~p is n!. This has to be true for all values of t.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . Substituting these expressions into Eq.e. / dpf(p)e~px the Laplace transform of f(p). E. (6. p. fp{ps{p)) = {v2 + n)s{p).g. We demand that C(p) := — \ps(p)e~pt] = 0.
Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. For singlevaluedness. Hence a possible path of integration > is from — oo to oo. of course). when p = —t (which is possible. (1) This will be satisfied if \p\ — ±oo.100 and hence CHAPTER 6. 6. n must then be an integer.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. We can show that if \p\ is allowed to be large. Therefore we take ptp2/2 y(t) .e. (3) We can choose a contour once around p = 0 in the complex pplane. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer.71) . We now investigate possibilities to satisfy the boundary condition C(p) = 0. The value of the exponential at this point is ~ exp(£ 2 /2).70) = 0 P. so that p cannot go to infinity. This can be attained if p is allowed to be large (t can be of either sign).n+l between the two limits of integration. 6.3. = / • Im p Rep n not an integer Fig. (6. consider the exponential in the integrand. so that y(t) ~ exp(i 2 /2).3 The path when n is not an integer. as indicated in Fig. i. To see this. we can choose a path from zero to infinity (the condition will vanish at p = 0). The quantum mechanical wave function is then (cf. This exponential assumes a maximum value when pt \—p2 = minimal. (2) If n < 0.. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. then for large t we have y(t) ~ exp(i 2 /2). since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. Eq.
3. if n is even and v = 1. pt p2/2 (6.„y2 v l { ' ^ r (^) \ 2 ' where v = 0. . 4 . In y(t) above. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0. The exponential function here is described as the generating function of the Hermite polynomials Hen(t)..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. (676) ..5.P 2 / 2 00 C—tY ( .. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .6.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. this becomes jfj. .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. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). We can also obtain the derivative form of this Hermite polynomial. We have 27T! Setting q — p + t. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. . 2 .if n is odd.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( .
for the evaluation of expectation values. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. (6. (6. Solution: We multiply the generating function by a factor t and perform a partial integration on the factor exp(— pt) contained in the following integral (the differentiation of the other factor leading to two contributions). as in the normalization of asymptotic expansions of Mathieu functions).n + l ) f f e n ^ i ( £ ) + (n. 8. Pptp 2 / 00 /2 dp. Hei(t) Henj. Example 6.1 '2dt using Hen(t) = (l)n27 r .e. 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 (6.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials.n— 1) = n. we obtain tHe„{t) (l)"n! ~ i.3 and 6. n — l)Heni(t).4.102 CHAPTER 6. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. Hen(t). 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+<. (6.59a).75) as Hermite polynomials.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . Starting from Heo(t) = 1. n + l) = l. in radiation problems.g. (n. Solution: We have to evaluate °° 2 .3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.70). one can construct the tower of polynomials Example 6.7 that we use throughout this text) and their orthogonality and normalization to Examples 6.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). Example 6.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. and elsewhere. ( n . Reinterpreting the remaining integrals with Eq.77) = t.
80) f(p. g) is n! Next.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).q) = = e .78) \/27rn!.n)\ i s an+2rij.q).5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. (6.jQ) n n!I ^—' ^Q fi\(n .^ + 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. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(./(n __ 2r — /i)!.66) is \Zn\V2w Example 6.6. zero otherwise (6. and 0 otherwise \/2n(—l)n+mn! if n = TO.5 nm . 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. so that /2 Q2(n+rrf in / ( p .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.82) ^ . Hence the normalized wave function of Eq.
the coefficient of T 2s »{sr u)\ As we observed.78). n + 4)_ffe n +4(t) + [(ra.t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .^ ) ! ( n + 2 r .n + 2) + (n.l ) ( n .v)\ ^ Multiplying both sides by the value of the normalization integral.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . n — 1) are given by the first index. n + l)(n + 1. the relation (n.A o?(.. i.83) = (n + 2r)!e a / 2 V „ „.ra + 3)(n + 3.v)\(s . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. n. n + l ) ( n + 1.v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!." ) a2^) r coefficient of (2s)! in f^ (s . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . we obtain the result (6.r .104 CHAPTER 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 .e.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .A i ) ! (6.l .e.(t)e«2/2dt Now. n + l ) ( n + 1. n + 2)]ife„ + 2 (t) + ••• .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h .^ / 2 * . n + l ) ( n + 1.79). i. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). Hence t4 He„ (t) = = Hen+i Hen+4{t) (t) + [(n + 3) + (n + 2) + (n + 1) + n] H e n + 2 (t) + • • • + 2(2n + 3)Hen+2(t) + . n ) ( n . n + 2) + (n. n)(n. n + 2)(n + 2.r .i / ) ! ( 2 r + i/)! > . It follows that the coefficient of a 2 s / ( 2 s ) ! in this expression J is the quantity K given by f^ot2sge„(t)Jfen+2t. n + l ) ( n + 1. n + 2)(n + 2.e '' ^ o^M ! ( n .2s _oo ( a 2 / 2 ) i . by n!(7r/2) 1 / 2 .' . . i / ! ( n .3 we know that upgoing coefficients are 1 and downgoing coefficients (n. K is the coefficient of a / ( 2 s ) ! in the expression J . . n + 2) + ( n .( n + Jr). The result then follows with half the normalization and orthogonality integrals (6. /i = (n — v) _ / r o o e .77). This result can be verified by inserting in the integral for t Hen(t) the linearized expression obtained with the help of the recurrence relation (6. n . n + 2){n + 2. i. n + l)(n + l .u)\ 2s<—" v\(n .— — ^ .e. n + 3)(n + 3. Prom Example 6.n+ l)(n + 1.
Let H^ = E^> (7. more generally. We consider first timeindependent Green's functions. we consider the sojourn time of a quantum mechanical particle at a point. This example also offers a convenient context to introduce the inverted oscillator potential. which is a point of unstable equilibrium for a classical particle. and is thus an important point to be noted here.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. We 105 .Chapter 7 Green's Functions 7. which we write G{x. in Chapter 21 (with a different calculation). The first of these cases will reappear later.1) be the timeindependent Schrodinger equation which has to be solved.x'). These are the Green's functions of the timeindependent Schrodinger equation or. and those which are timeindependent. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. of a differential equation of second order.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent.t). 7. In particular we derive the timedependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. With reference to this case. in Feynman's path integral as the kernel or free particle propagator. which is normally not discussed.x'. which we write here K(x.
Eq. (7.13). a perturbation part like (3q4.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . (7. appropriately decreasing behaviour at infinity). Green's Functions H = H0 + Hi. i.x'). (7. of the kinetic part p2/2mo and some part of the potential.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).x') is defined by the equation* (Ho . e. aq2. We can readily see that G(x. the relation X') = 5{x . (7. The boundary conditions which G(x.x') or ^2^(x)^*(x') i = S(x . The timeindependent Green's function G(x. i. square integrability.EW)xG{x.x') by recalling the completeness relation. . where £ ( 0 ) = lim E.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. for instance. 'The case of £(°) equal to some Ei is considered later.g. since the generalization to higher dimensions is selfevident. We can obtain an expression for G(x.e. and Hi is some other contribution. e.e. x') has to satisfy correspond to those of * (to insure e.x') can be written t G (7.106 assume a Hamiltonian of the form CHAPTER 7.2) where Ho consists.g. although the explicit calculations can be much more involved.x'). cf.g.
E^ We can check this by considering (H0 .0) = S(xx'). We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.Hi{x'))^(x').E® (EE{0) Hi{x))m{x).EM)X*(X) ( .x'. The timedependent Green's function K(x. We see that the initial condition (7.x'. (JL x'\ t) = H0K(x.E^ of the eigenvalue.7) with the initial condition K(x.t).7. (7.(*'))*fr') Hi{x'))^(x') f dx'5{x ..t) (K for "kernel") is defined as solution of the equation ih—K(x. (7.x')(E .11) = = 3) = fdx'(H0 . x'){E .) = f dx'G(x.x'. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E . .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.9) = i H0YieE^i%(x)^(x') = H0K(x.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').t) since ih^K(x. .E<®)xG(x. 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). i (7.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x. (7.x'.x.10) The contribution on the right hand side is the socalled perturbation contribution. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.x'){E . x'\ t) (7.#.£(0) .
Green's Functions Can we add on the right hand side of Eq.6) is not defined. Sec. Merzbacher [194]. In general. integral equations are more difficult to solve than the corresponding differential equations.108 CHAPTER 7.6) we would have (7. see B. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. D.2 (not contained in the first edition!). J. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. that (E . Instead of Eq.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. if we assume that E^ = Ej.11) with a system of linear equations of the form Vi = Mijyj. W. MiillerKirsten and A. But if we assume that E^ ^ Ei for all i. . In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. see E. 2nd ed. (7.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. Wiedemann [183]. All such methods of solution are based on the analogy of Eq.15) ''This problem and its circumvention can be formulated as a theorem. In cases where an inhomogeneous contribution is given. The theorem is also known as Fredholm alternative. (7. i. Bleecker [33].12) Equations of the form of Eq. But then the Green's function (7. Examples in various contexts have been investigated by L. Booss and D. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). (7. (7.14) We can now see the significance of the timedependent Green's function K(x. 17. H.. Maharana.t) for the complete problem as follows. then at best such a contribution would be something like cipi(x).e. (7. This will be different. however. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x').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).
(6.e.20) is the timedependent Green's function. (7.42)) 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. Comparison of Eq.x'.7) with Eq. (7.0). this relation provides the probability density \ip(x.t) describes the evolution of the wave function from its initial value ^(a^O). We use this in Sec. 1 See e. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. in different formulations.t) = where if(x^'. *>0.t)il>(x'.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih.t) = ^ e £ " f / « V n W < ( ^ ) .2 for the computation of the sojourn time.158. t > 0.21) dx'K(x. t)\2. 2nd ed.15) which contains (7.17) is the initial condition of \4>)t.5.16) As usual.x'.20) the Green's function K(x.17) as initial condition is obviously W)t = Y.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x. p.x'. Merzbacher [194]. E.19) i/>(x. (7.g. t) with the Note that when K(x. (7.7. We can write this expression also as§ ip(x. e. (5.. \En)(En\ip)t=oeE^\ n (7. (7. 7.g.0)eEnt/in. . i. we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.' which obviously satisfies the initial condition K(x. (7. (7.t) is known. for the oscillator potential.The solution of Eq. H\En) = En\En). x'.Q) = 6(xx'). Eq.18) or. x'. 109 (7.22) According to Eq.t) = J2 fdx/^n(x)^*n(x')^(x'.
23) along the contour C in the plane of complex E as shown in Fig. t) = J2 e £ " f M i ( x ) < ( x ' ) . 7.E .x'. we obtain then also the corresponding density matrix. 7.1 The contour of integration. (7. Inserting for GE+ie the expression above. we obtain / W : = .x') and K(x.x') = GE(x.1. Green's Functions difference that (5 = 1/kT plays the role of it. x'. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x.t) K(x.ie (7 . In the following we shall derive the Green's function for the case of the harmonic oscillator.x'.25) in agreement with the timedependent Green's function . between (with E^0' = E) G(x. Jc 27T „ hn . x')6(t) (7.110 CHAPTER 7. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.t). i.i /f e «/. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.«^«2 W . As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. 24 )  ReE Fig.e. t > 0.
(7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.29) and (7. .7). we obtain *I£I = h2 —(2AB).x'. A and B being constants. i. In this case the Green's function is the solution of the equation d hK(x.30) into Eq.t) t h2 = .3 The Green's Function of a Free Particle 111 7.26) This is the case of a free particle with mass mo.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. K(x.x'. / dxK(x. (7.31) The constant A has to be chosen such that K(x.X>.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.e.— d2 —2K(x.29) tJJ 2 2B(x . (7. which is moving in one space dimension. and identifying coefficients of the same powers of t on both sides.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.t).7.30) Inserting Eqs. (7. 2mQK h h2 ih(AB) = —{4B2A).x'.28) B(xx')2 t2 DB(xx') 2 /t (7. v ' 2m 0 v h i. Thus we try the ansatz. It is clear that it is nontrivial to solve an equation like Eq.0) = 1.e.x>.0) = 5(xx').27). B m0 n _ 2ih 2iK (7.
x't) = Y.H We can insert the expression for K(x.t) into Eq. Merzbacher [194].t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. n For a free particle moving in the onedimensional domain \x\ < L — oo.112 CHAPTER 7.x'. from K{x.oo 27T (7.38) .20) and can then obtain ip(x.t) = — / dkeak2+Pk 2vr J. i.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. we > have to make the replacements V ^ fdk.ihMxWn{x').36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. (7."737) \27rii/i in agreement with Eq.21). 32) It follows that x'. (7.35a) K(xy. we have This is 1 provided A=JB= ^ (7 . Green's Functions For parameter values such that the following integral exists. (7.t) — for instance for a wave packet given at time t = 0 of the form ^(x.34) K(x. p.0)oce"Qx2+ifcox. 158. "See the excercise in E..e. (7.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS . (7. v27r En^^. j3 = i(xx'). *>0.33). t) = /jJo_ e mo(««') 2 /««. 1st ed. 2m0 (7.zEnt.x'. Mx)^M^) „ n so that J =^ .(7. (7.
25) — in the context of Feynman's path integral method.. z .x'. (7. (7. K(x. . We rewrite this initial condition in terms of £ and use for a = const.d r. . I (7. i.t).33).40) Then Eq.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. mQLu . §jK(x. HQ = ^—p2 ZrriQ + \m0uj2q2.e. . —ih—K(x.t). 1 2mo u>n ox n with the initial condition (7. . (21.x usn ot i.——K(x. in this domain *Recall that S(x) = ± f eikxdk.e. x'. x'. 7.42) .e.3 The Green's Function of the Harmonic Oscillator 113 The result (7.7) is the solution of d ih—K(x. / ) (7..37) will later be obtained by a different method — see Eq.x'.7. i. i.x'. 5{ax) = ± J eikaxdk = ArS(x).4 Green's Function of the Harmonic Oscillator The next most obvious case to consider is that of a particle subjected to the harmonic oscillator potential.39) In this case the timedependent Green's function K of Eq. / ) = ~^K(x.8).x'.0) = 6(xx') at / = 0.t) 1 + m0u2x2K(x. x'.x'. (7.40) becomes 2 . .t) = h2 2 d2 TW . the relation* 6(x) = o<y(os). (7. —KK(X. .t) We now set h2 d2 = .X .e.. We consider the onedimensional harmonic oscillator with Hamilton operator HQ./ ) + fK(x. (7.t) + ~x K(x.
Identifying coefficients on both sides. .45) in accordance with (7.^ .4 a 2 .c o t h 2 / . (7. f i. so that for / — 0: > a^—.. (7. (7.f) with a(0)»y.50b) (7.4a&£ + 2a .e.b2. that is. (7.Z'.50a) (7.50c) (7. Green's Functions to the particle behaves almost like a free particle. 50. we must have /o = 0.48).47) We insert this ansatz for K into Eq. (7. P.b2.x'. a = .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.47) becomes (7. 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.) a'£2 + b'i + c' = (1 .45) in this sense. Integrating Eq.46) If we interpret Eq. we obttain a' = l .114 CHAPTER 7. it • is suggestive to attempt for K the following ansatz^ m.t) become similar to expression (7.33). as the limiting case / — 0.4a 2 )£ 2 .e. c(0)^.42) and obtain the equation (with a1 = da/df etc. i. W ) . (/ = 2a. (7.51) See R. p.50a) we obtain (7. and we expect K(x. Feynman [94]. b' = Aab.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.
115 To ensure that in accordance with Eq.t) to obtain this element (x.50c) yields for c(/).In B. B independent of / . Comparing the Eqs. Felsager [91]. we can use K(x.c(f) K = 5 =exp Vsinh 2 / into Eq.b(f).48) 6(0) = — £'/2f.40) of the timedependent Green's function with Eq.5.x'. (7. (7. we must have A = ('. (7. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . with A. in particular for the derivation of the sojourn time in Sec.X'. K(x. p. .0). 7.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o.7. i.50b) Kf) = —r^77) smh 2 / A independent of / . and to ensure that we obtain the prefactor of Eq./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. (7. (5. t. (7.i) (7. as we shall see in the following.45). as one can verify.27).33) for the Green's > function of a free particle. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. In order to satisfy Eq.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. i.e.51) for the density matrix PN(X.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.48). c(0) = £' / 4 / . x .2. / —————rexp y 27rmsin(u.x'.t) = . x'.53) or. (7. if we return to x. 174.* With this result we have another important quantity at our disposal. Finally Eq.55) 'For an alternative derivation and further discussion see also B. we must have (besides A = —£') = /m 0 u. (7. (7. 2TT^ Inserting a(f).e.54) For t —• 0 this expression goes over into the expression (7.
. ^—• Irpiv J pN(x. Eq. = jdxp(x.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. we skip the algebra here. P. (5.116 CHAPTER 7.uij>0 >• 0.. .P) 2.p)dx W ) = —^r = —p .56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf. )x Inserting into Eq. we obtain § < Z> = ^ c o t h ^ ^ ° — .56a) the expression (7. For instance we have with Eq. Thus the system is in a mixed state and the expectation value "Cf. <2 (7.x. R. Feynman [94].» l. at temperature T): {A) = Tl{pA) = ^ > .52): Tr (PNq2) Jx2pN(x. p.56a) {q2) = = Y.55). (7. Eq. [ [ 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.x.e. (5. 52.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. (5.e. Green's Functions With this expression we can evaluate (cf.40)) 1 epEi 1 Ui = y. 11 For T ~+ 0: u>0 . eP E i = i + y. we verify the relation: (Q2) ~ dxdx'(x\p\x'){x'\q2\x) = / / dxdx'(x\p\x')(x'\x) x2 x){x\p\x')(x'\q2\x"}.x.52)) the expectation value of an observable A in the canonical distribution (i.
x'.3 The Green's Function of the Harmonic Oscillator 117 (7. 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. e > 0.e. We assume t > 0 and t —> oo.54).J (w — ie)< e (n+l/2)Et e i(n+l/2)u.56b) is that with respect to this mixed state (whose cause is the finite temperature T).58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states. setting a = mooj/h. 2 + x'2)le^2xx' to—>w—it t—>co 2 .7.50) we obtain. t) of Eq. which means in the oscillator state \i) with eigenenergy fku(i + 1/2). of K(x. If we consider the system in the pure state \i).QUJ (7. and we replace to by LO — ie. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . (6.59a) E 0 = ^fiw. We return to the Green's function (7.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . In this case the Green's function K(x.21).x'. 1 mo^ exp lmow / 2 2 ft X (7. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.t "With the normalized ground state wave function of the harmonic oscillator given by Eq.54) is K(x. i.59b) n+ l \ ftwt here 2hf exp _ i ( n + . This is the first and hence dominant term of the expression (7. (7. x'. t) = J2 eEnt/ihct>n{x)<t>n(x')For t > 0 and En = (n + l/2)frui the factor exp(Ent/ih) exp is (7.
Considered classically.1 W a v e packets The simplest type of wave is the socalled plane wave or monochromatic wave of frequency UJ represented by the expression exp[i(krwt)].(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. 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.e. (7. (7. Green's Functions For t large (i. as in Eq.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. i. to infinity) the contribution with n — 0 dominates. of dpN as pN(x.118 CHAPTER 7. the velocity of planes of equal phase.e. In Example 7. the particle will stay there only for a finite length of time T.e.1 we estimate T semiclassically. lie on surfaces. and with this we estimate the sojourn time T.5.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples.x'.59a).62) . In a very analogous manner we obtain the solution of the equation for the density matrix. (7. k = k. which are planes. However quantum mechanically in view of the uncertainties in position and momentum. i.. The wavelength A is given by The phase velocity v<p. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0. 7. is defined as vv = .
Q ( k . / oo J—oo .e. (7. a > 0 .k ') 2 . If we assume for / ( k ' ) a Gauss distribution. The wave packet describes a wave of limited extent. oo (7.)¥'(*:")) +ism(<p(k') .)ll/(fc")l[cos(¥'(fe.7. then ^(r. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.63) The relation u = w(k) is known as dispersion or dispersion law. to the function ip given by oo / fik'y^'x^dk'. essentially again a Gauss curve. i ) = /*/(k')e i(k '" r '' .65) Let f(k') = \f(k')\eia and <p := k'x .64) where / ( k ' ) differs substantially from zero only near k' = k.<p(k"))}dk'dk". e. We now ask: How and under what conditions does the time variation of the function ^(r.e. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. i. ' t) dk'. duo' da . (7.Jt + a. t) is the spatial Fourier transform of this Gauss distribution — as we know. i. for which \ip\ assumes its largest value: OO . ^ ( r .4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. e . or dco' da The centre of mass determines the particular phase. i. u = w(k).e. the fundamental postulate on matter waves in Chapter 2): E = hu. xt— + — = 0.
1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . dE v 9 = . In the present case. .^ g=r a rad. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. (7. Solution: The function to be calculated is the integral dxeax OO ezkx. For E = hio.— g(k). as claimed for =0. The threedimensional generalization is evidently du . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. .120 CHAPTER 7. however. we obtain the deBroglie relation p — hk. = — (768) It is instructive to consider at this point the following examples. p = hk the group velocity is equal to the particle velocity v. 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 = . i.69) In general one uses the theory of functions for the evaluation of this integral. we can use a simpler method. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. a > 0..67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)].fc w> vv = ~a^ = S r a d P E «i7 Sd u .e. Example 7. (7.
J.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. . MiillerKirsten [215].69) and (7.Afc = 81n2.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .r „• With the help of Eq.75) ' E .1 verify the uncertainty relation Aa.74) Example 7.73) 27T 7 . where c = 9(0) is a constant.71) (7. XZ (e > 0). (7. \a — b\ (7. (7. (7.70) we obtain 6(h) = lim — f°° dxeax2eikx = . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration.69) and (7.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example. Example 7.7. 2 f + . (7. (7. Appendix A. (7. with partial fraction decomposition. W...2: Representations of the delta distribution Use Eqs. or see H..4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a . g . We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 ..0 0 7T 7 0 7T e From Eq.&)] for a + b..a) + 6{x .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. (7.72) Solution: From Eqs.
Thus a slim maximum of the curve of /(a. (7. and hence implies large values of Ak.r)1/2a Re f(x) Ax Fig. where g(k) = max<?(fc)/2.77b) (7.70) the Fourier transform of f(x) is (7.g. Green's Functions 1/(27i)1/2a — 1/2(2. is Afe = 2 V 2 1 n 2  (7. 7.2 The Gaussian curve. In quantum mechanics the square of the modulus of the wave function ip(x. A simple calculation yields A i = 2v/2~In2a.) leads to a very broad maximum of the Fourier transform \g(k)\. i. 7.77a) small values of a. The wave function corresponds to the function f(x) in the above considerations (e. Thus a sharp maximum of the function f{x). According to Eq.78) It follows that the product of the uncertainties is AxAk = 8 In 2.77a) g(k) = e . at time t = 0).t) is a measure of the probability to find the particle at time t at the position x.« 2 ( * .e. Physically . The breadth A/c of the curve g(k) around k = kg. (7.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. when Ax is very small.* o ) 2 / 2 . i. where  / ( x )  = m a x  / ( x )  / 2 . requires according to Eq.e.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. the width Ax is a measure of the uncertainty of the probability. Solution: In Fig.CHAPTER 7.
76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . the less precise is the determination of its associated momentum.71) > and (7. (7. This means the particle is localized at x = 0. The function g(k) is therefore described as momentum space representation of the wave function f(x). Thus G.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. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. In the same limit the momentum uncertainty Afc grows beyond all bounds. In the limit a — 0 we obtain from Eqs. (7. Lamb at Oxford set the following problem in an examination in 1957: "A pencil is to be balanced so as to stand upright on its point on a horizontal surface.«> = / ^ C ' W . many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. 7.5. so that Ki0(x. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate.7.81) *G.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. (7.75) implies therefore: The more precise the coordinate of a microscopic particle is determined. Barton mentions at the beginning of his paper. <7*» We assume that initially. the wave packet has its centre of mass at the origin and has the following Gaussian shape as the pure initial state: iP{x^) = ^L=ex2l2b\ (7. For reasons > of simplicity in the following we set in addition t mo = l. ^In Eq.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . Barton [15]. (7. (7.54) with the substitution ui — iu.x'. Barton recalls that W. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.79) According to Eq.".. at time t = 0. The result (7..e. E.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. As G. .h = 1. which means all values of the momentum are equally probable.u> = l. O ) .2 ) and U)t. i.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. nr = S(x).
. (7. (7. Green's Functions Inserting (7.88) .82) i/f(x.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.85) = —r= o2 7 2 sinh 2i.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .83) ip(x.84) we obtain* ip(x.124 CHAPTER 7.79) into (7.81) and (7..86) Evaluating the Gaussian integral of Eq.t) = J oo P . 9 (7.cosh t sinh t' Wisll2b ^ (7.t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.(i/2) sinh 2t 2 2 2C 2 (i) + *c^. 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 . ' 2\Aswht % cosh t (7.2 (7. = . + cosh 2 1. A .t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = ..(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.
since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.93) where —{dQ/dt)5t is the probability that the particle leaves the vicinity of the origin in the interval St around t (partial integration leads from the second expression back to the first). 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.t)= where Co(t) / dx\^(x. (7. We have therefore .£ 2 (7.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 . t) the phase ip drops out and one has (x.t)\' 2 125 (7. •&(t)=//c(t) = / or.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.7. _2 1 exp C 2 (i) = l + 2sinh 2 i.4 The Inverted Harmonic Oscillator The real part of Eq. In taking the modulus of ip(x.91) C(t)' (7..t)\2 Ji == 71 / V " Jo dO ? 6' (7.89) = exp[x2/C2(t)} 7rV2C(t) (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1. C2(t) (7.95) .94) / dxf(x)=g'(y)f(g(y)).90) with the expected limiting behaviour \im \ilj(x.
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.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.98) We insert this into Eq.99) . we obtain dT _ 1 ~dJ~l 'See for instance H.96) °° l Jr. p. C(t) = ..97) sinh 2 1 = . V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.From this we obtain C(t) C'(t) Z2 ^(r?2_.(4 2 .= \ / l + 2sinh 2 t.=0 V^ C'(t) I Jt=0 V^ (7.126 We set I C(t)' so that dT ~d~l Now. B. + 0(e'2) (7. 136.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.1 ) 2 \n J ^"•""ff^^V^K?. (7. Dwight [81].
B. Explain the parameters entering the calculation.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. U) (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.100) Reintroducing the dimensional parameters we had set equal to 1. . is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. this is T ~ ]n(ly/mou>/h). 238. this equation is the equation of motion of this scalar field with negative masssquared. also motivated by this paper.g. Fig.102) 'As a matter of interest we add that in the theory of a free scalar field x. Zwiebach [294]. Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.4.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. The equation there describes the classical "rolling down" of this tachyon. It follows t h a t T~lnZ. (7. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon".7. p. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. See e. T h e following Example 7. 7.
.5)) Ax = <J(x2){x)2.102)) p(0) = ^/(p 2 >. Green's Functions x = Aco sinh cot + Bco cosh cot. (10. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). 2 10 x(0) = B u . Ap= v /(p 2 )<p)2. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. (x 2 ) h 2mou 2mou> and so l^J^ULl). the smallest macroscopic length. / 7 ^ 2 =2J(x }.128 with x2 = LO2X2 + const. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). (7. ^(x2)(p2). Quantum mechanically x(0). 2V ' Let A and Bio be the values of x and x at time t = 0. v A x A p = \ (x2)(p2) v = m0u(x2). but instead are subject to an uncertainty relation of the form AxAp > h. A p are defined by the mean square deviations given by (see Eq. (p2):=mW(x2). i. CHAPTER 7.103) 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. 2 x(0) ^ where A i . Eq.g. and for simplicity we take (cf.e. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. Thus quantum mechanically x(0).p(0) cannot be determined with arbitrary precision. For h —> 0 the time T —• oo in agreement with our classical expectation. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0).e. so that e. i. x(0) = A. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. 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 . (7. and solution x = A cosh Lot + B sinh cot.
For mathematical purists the question of convergence of the series is an even bigger challenge. 129 . i. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. The perturbation method generally described in textbooks — and frequently called RayleighSchrodinger perturbation theory — consists in assuming power series expansions for the wave function * and the eigenvalue E in terms of a parameter like (3 which is assumed to be small. Frequently even the calculation of the next to leading contributions ip^.E^ is already a bigg problem.e.(i)+/^(2) + ". but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.e. one would set tt E = = ^(°)+/ty. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E.1 Introductory Remarks The Schrodinger equation can be solved exactly (i.Chapter 8 TimeIndependent Perturbation Theory 8.2h2 cos 2 a # = 0. in closed form) only for very few potentials. In general perturbation series do not converge.
It will be seen later (e. N. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x.2) n—»oo n It is interesting to compare the behaviour of the series (8. J. In the following we explain the difference between convergent and asymptotic series. W. Whittaker and G.= 0 < 1. (8.g.g. at the end of Chapter 26) — and by comparing the results of WKB.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 .+ . see e. h hl are in the strict mathematical sense divergent.2 However..130 CHAPTER 8. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . This convergence can be shown with D'Alembert's ratio test since''' lim .. Meixner and F . .. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence.e. 581.* i. which is known as the Mathieu equation. T. E. . p. concerning also the uniform convergence of the exponential series.g.. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified.g. 0. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. TimeIndependent Perturbation Theory In the case of this equation. E = E^ + h2E^+hAE^ 71 72 7 + . the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. Watson [283]. 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.3) *See e.2 Asymptotic Series versus Convergent Series Before we actually define asymptotic series we illustrate some of their characteristic properties by considering specific examples which demonstrate also how they differ from convergent series. ^For further details. the expansions in descending powers of h2. Schafke [193]. (8. e.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). 8.
the series (8.8.5) Cf.g.22222 . E. p.. Comparing the asymptotic series (8.000 000 06 + • • • . From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ ....000 002 00 . N.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.. . for x = 3. and we see that the moduli of successive terms first decrease and then. We can see this as follows. Thus e.. However. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. It is evident that the larger the value of \x\. i.e. / ( I ) = 1 — 1! + 2! — 3! H • However.33333 . .1).2. + 0. in spite of this the series (8. .3) with the convergent series (8. begin to increase again.22222 . and in fact increase indefinitely. 235. +0..9876.0.. .0.29629.4) for every value of x. 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.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. for larger values of x.0. /(1000) = 1 . + 0. .4938. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20.001 + 0. Whittaker and G. after reaching 6/27 = 0. T. + •••. Watson [283]...g. ..222.3) diverges for every value of x.. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 .l n ( l + ^ ) n=l L ^ ' (8.0.376 . the partial sum of terms up to and including the least term yields a reasonably precise value of the function at that point with an error of the order of the first term of the remainder. We observe in the first place that since lim n—>oo co (8. The theory of asymptotic expansions claims that if the expansion is truncated at the least term.0. the better the approximation obtained. e.
E. 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. 152. meaning convergent. Vol. 16. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. It is interesting to observe the comments of various authors on this point. Whittaker and G. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). ip) can be represented by rapidly converging power series in e. the dominant approximation is." These "rapidly converging power series" in physical contexts are most likely very rare cases. Messiah [195]. ^E.41) and (16.1. p.5). This latter observation hints already at the importance of asymptotic series in applications. 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. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. the leading term of an asymptotic expansion. In many respects Messiah aims at more rigour in his arguments. 11 L. it is sensible to make the assumption. 371. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. . Watson [283]. p. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent.132 CHAPTER 8. In fact. Sec. (16.e. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. p. 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.5) are practically unknown. T.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. 236. I. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one.J In2. although this has not been investigated except for a few simple problem^. in fact. TimeIndependent Perturbation Theory where 7 = 0. Merzbacher [194].1 between Eqs. N." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. whereas applications of the convergent series (8. Schiff [243]. that E and \ijj) (i. This is not appreciated by mathematical purists. § However.
Replacing t h e exponential in Eq.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." Of these three authors t h e first. (8. II.2. solutions of second order differential equations. there is no need for such bias.11) A . 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. Dingle [70].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.^=e~x2 / e^^dt.8. In fact. We know from books on Special Functions t h a t all of these functions. Vol.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.10) Vn Changing t h e variable of integration to u = t2x2..2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. . which have been studied in great detail. B. we can rewrite Eq. Merzbacher. (8.7) as / e'* dte* dt = 1 ..7) This function has been considered in detail in t h e book of Dingle. 8. 2 (8. (8.* and we follow some of t h e considerations given there.. p. He says. *R. It is therefore suggestive to write iy poo <t>(x) = 1 . Messiah [195]. ft Jx du = 2tdt. (8. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . (8.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . 198 (German edition). possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.= / e~t2dt.
du. 7T (ni)!sin{7r(n+^)} ( „ . Using the reflection formula (*!)!(*)! = for z = n + i .12).' we have 2 nH 7T 7T (8. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . (8.16) . we obtain 1+ i.^e~x \Ar i.e. « \~2 =E n=0 l 1 V5F(ni)!(l)n nlTT (M x2) / « r oo 1+ For the domain \u/x2\ <f>(x) U \2 1 \/K n=0 2< n! V (8.^ X e U ~ ( 1 + * \ 2 .134 we obtain CHAPTER 8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8. TimeIndependent Perturbation Theory <P(x) = 1 .15) < 1 we can insert this expansion into Eq.e.14) sin7T2. since (—4)! = y/n. Then e (n —\)\( 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 (8.13) presupposes t h a t < 1. (8.i)[(_!)«' Then.
We now see how the asymptotic expansion (8. i. (8. (8.19) approximates <f){x) the better the larger x is.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. (8.18) or that we insert the binomial expansion (8. convergent) only in the restricted domain u < x2. the fact that we effectively use a binomial expansion beyond its circle of convergence implies that the resulting series is divergent so that even if the first few terms decrease in magnitude the later terms will increase eventually as a reflection of this procedure. and in fact so slowly.15) into (8.e.12) ignoring the fact that the latter is valid (i.19) originates. Whichever way we look at the result.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq.r<7r. the asymptotic expansion of the error function is.l)! = / Jo Then 2 cHnxdt.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 "~". argx = 0. i. much more useful than the convergent expansion.e. In an asymptotic expansion the first few terms successively decrease in magnitude. As in the case of the gamma function. We can actually understand the deeper reason for this in practice.17) 0(X) = i. It is fairly clear that the above considerations remain unaffected for argx < 7T. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. We now want to relax this condition and allow for phases argx ^ 0. whereas in a convergent expansion the terms first increase in magnitude. In the foregoing discussion of the error function we assumed that x was real and positive. /•oo T(n) = (n .8. The expansion (8. . ' arg. (8. in general.—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^.
TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. The situation becomes critical. (8.26) or else return . (8. (8. thus with \v/y2\ < 1.2 y V . (8.20) where we set t = is. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.21) = y2s2.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.136 CHAPTER 8. (8. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 .23) Throughout the range of integration 0 < v < y2 the integrand is real. we can try to proceed with Eq. We therefore consider this case in detail. when argx = 7r/2.24) We can read off the binomial expansion of the factor in the integrand from Eq. V7T Jo Changing the variable of integration to v (8. =dv. however. dv = 2sds = .22) we obtain 2 4>{iy) = . (8. and we can write (j>(iy): y^ Jo t^Q n] \y J If we want to proceed with the evaluation of the integral in order to arrive at an expansion of (f>{iy). we set x = iy in Eq. VK Jo V^ Jo (8. i.15).e.vds.^ = .^ f V e ^ .
and the larger y. If we proceed with Eq.e. Eq. ey is much larger than something of order 1/y..^ ^ p .30) yvn f Jo V y J Cf.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 .27) and (8. (8. (8. the better the approximation expressed by Eq.17): J£° e~vvndv = n\.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! .24). (8. — (8. (8.28) ignores the integral contribution of Eq.28) because then the correction term oc f°£ becomes smaller and smaller. y y 2 ( n _i)! y arg(* = iy) = i * . Returning to Eqs. e ^ ( i y ) ^ ^ . We have therefore <j>(iy) = ^e^TL fV e~v ( l .27) Then . We know that for v < y2 the integrand is real.27). However.19).8.2 yir (8. and so the integral is real.28) we observe that the expansion (8. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. i. (8. Suppose now we consider Eq. i (8. These integral contributions are of order 1/y since the integral behaves like e~y and is multiplied by a factor e+y in front. (8. .e.29) where 5t means "real parf. for large values of y.24). (8.^] * dv. which will be studied in more detail below.2 Comparison of Asymptotic Series with Convergent Series to Eq. i.28) n=0 We observe that this expansion differs significantly from (8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. (8.
33) The sudden disappearance of "1" at argx = 7r/2 hints at something like a discontinuity of (j>{x) at arg x = n/2 which was discovered by Stokes and is therefore known as a Stokes discontinuity. We have therefore found that XTT n=0 ^ ' and . Since (cf. B.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i.31). (8. With a similar type of reasoning one can show that the same expansion.26). (8. (8.25) and integrate from 0 to oo. 7r = . since (8.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. TimeIndependent Perturbation Theory If we insert here the expansion (8. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.28). This result is identical with that obtained previously. R. Chapter 1. i. Looking at Eqs.31) and (8. Dingle [70].e. ." — E ^ ^ forarg. and • higher order terms of the associated series all have the same sign. y. (8.3!) which is (8. . (8. e xZ ^ (n — £)! „ ^ ) . The corresponding phase lines are called Stokes rays. (8.. *Cf.138 CHAPTER 8. is obtained for <f>(x) with arg x = —ir/2. Eq.33)) would have • a maximally increasing exponential e+x .2 OO rz=0 V . For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2.e.
pp. 9 .34) the contributions 1. 5. (8. for arg x = 7r/2. R. Equation (8.e.2 Stokes discontinuities of oscillator functions Dingle' has formulated rules which permit one to continue an asymptotic series across a Stokes discontinuity without the necessity of a separate calculation of the asymptotic series in the new domain.8.g. though.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. Eq. i.32) and (8. apply predominantly to asymptotic series of the solutions of second order differential equations. '"Cf. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . B. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. 91. in the sum of the right hand sides of Eqs. Magnus and F.e.35) 8. (6. . See also Tables of Special Functions.2). Chapter 1.13. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a.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. —1 cancel out.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. Oberhettinger [181]. i. Dingle [70]. e. p. and so may not be universally applicable. Thus if we set x = XR + ixj. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x).rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. These rules.37) Cf. 1/1 ipv{x) v1 vl (±ix). W. We also observe that the expansion for  arg x\ = ir/2 is half the sum of the expansions on either side of the Stokes ray.2.
e. Chapter 17. 0 < argz <  (8. x~l/~1ipu(x). TimeIndependent Perturbation Theory i. 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.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. Dingle [70]. Thus D{J\x) (i. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a.37).7.x 2 oo «*)  — E^fi=0 v .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. B.40) § Cf.140 where^ CHAPTER 8. = 7r/2.39) (as in the example of the error function). We consider Dv (x). > > The equation (8. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. Eq. with a real proportionality factor. Chapter 1. as can most easily be seen in the case of the Mathieu potential. We also observe that the late (large i) terms of the series in (8. v — — f — 1. 9) and XXI. (8. i. i. R.36) is invariant under these replacements. R. The continuation of (8.39) onto the Stokes ray and from there into the neighbouring domain is determined by the following rules which will not be established here:" • Dingle's rule (1): On reaching arg a. 8. Hence Dil){x) = xv(j>v(x). Chapters I (in particular p. . Dingle [70]. ti asymptotic series (8. (8.e.ei<v+VxvHv{x\ argx = £ .e. (14). (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. 1 1 Cf.
we must have at arg x = 7 in the dominant factor on the right of Eq. = ir.41) xuct)v{x) + a{x)~u~lipu{x). In (8. 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.41) the Stokes multiplier a is.e.44) + ^iap) = 0 . beyond arg x — 7r/2. Since Dv (x) is real when x is real. 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). for xj = 0) and therefore possesses a Stokes discontinuity there. as yet an unknown real constant.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2.e. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. with \x\ = (—x) for arg a. i. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. (8. where in the first line on the right hand side the second term contains the real proportionality constant a/2. (8. i.8. The value of Du (x) on the other side of the Stokes ray.41) to arg a. (8. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. = ir. sm(iri/) = —a/3/2 or a/? = 2sm(irv). (8. 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. < 7T. On reaching ir the part containing ^ ( x ) .42) T ^(eiwu i.e.40). another half of the discontinuity appears on leaving the Stokes ray on the other side.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).e. We therefore proceed to continue (8.
(8.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.e. . a{y)P{y) = a(u . V i=0 \U0 v 2w . i.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .e. i. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.e.e. (8. Eq.51) .1) (8. i. (8.42)) for argx = ir (in the second line of Eq. Hence the left hand side must have the same property. (8.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.42) we insert from Eq.—^ ON • — ^ > + ^ ^ L .1) = 2sin7ri/. (8.l)0(v . TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. Dl \X) is (cf. argx = vr. i.1) = ±a(y). (3{v . (z)\(ziy.14).42) that a multiplies ipv(x).38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .45) that af3 = 2sin(7r^)) D^\x) = (8. 8.1). (8.45) is therefore given by a{v)a(v .142 CHAPTER 8. We can see that the equation is satisfied by (3{u) = ±a(u . a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .1)! TT.48) We compare this with the reflection formula (8.v .46) For a given function a(y) this equation determines /3(u).47) Equation (8. (8.1) (8.
(8.I / ... (8.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8..53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.7).2.A / ± sin^Cx)"M* 2 V ^ . Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir. solutions of Eq.50a) because this enables us to define normalizable parabolic cylinder functions. Chapter I.„ _ix2^(2ivl)l e 2 > —.1 ) ( .e^ .1. (8. 8. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .8. ' ^ (i/l)!z! ^ .35).52) (argx = 7r). It should be noted that in Eq.1 ) ! 4~L i\{2x2Y i=0 V2^(x). Eq.36)) in the domain 0 < arg x < IT.e. We observe that for v — n = 0.) (8. the second contribution vanishes and we obtain £>«(*) .53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0. —f.e.7r). i. the error .*"ei°> f ( ^ _  _ L _ = (!)««(. ' (X) (cf.e. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di. (8. For details we refer again to the book of Dingle.)(.. However. (2x2)1 = COSTTU(X) v . i.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. u 1 y> {2i + v)\ 1 n 0. i..36). (8. which vanish at x = ±oo..
60) 2 x x l Inserting (8. % = {. x—>±oo (8. + — + ^ + • • • . ^ + 2 . and demonstrate.36) in the form (coefficient of — x2 here chosen to be ~. Eq. • • • can be determined. we obtain equations from which the coefficients So.54) we consider again the important equation of the harmonic oscillator.. i.37).144 CHAPTER 8. x. Eq. (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a. i. how series of the type (8.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation.54) (8. 0.36)) 0 where + x2Q(x)y = 0.^ + 1 ) .38) can be obtained. 5„=4 Sl=o.60) into (8. previously we had ^.. S2.59) and equating to zero the coefficients of powers of x2. We write Eq. x°. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 .. (8.e. (8. S\. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. (8.e. cf. Instead of dealing with Eq (8. Solving these equations we obtain = = 0. \ + U (8.55) The large x asymptotic expansion can also be obtained from the differential equation. (8.36).
z) is a confluent hypergeometric T. are asymptotic series in a variable x. i.35) and (8.61) (the dominant terms of (8. Magnus and F. The derivation of the asymptotic series from integral representations (cf.8. (8.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. Nonetheless we shall need asymptotic expansions like (8. In the following we are mostly concerned with asymptotic series in some parameter.e. and it is clear t h a t higher order terms follow accordingly.+ X M W = Q. 2 X 2Wi.36) have correspondingly e±x / 2 ) . 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+ . p. do not enter here into their derivation and simply quote the result: W i t h \x\ .b. a z a a ( + 1) z2 which has the unit circle as its circle of convergence.19). 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. therefore. n 22e"T JziL (8. / i N .38). T h e expansions (8. 4~l~2 ' i i _i. . Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). function.37). In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains.62) 4 \ 2 XiFi 2'2 .91. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  .63) **W.38) in a variable for the solutions of approximated differential equations. etc.37) with (8. » 1. We. Oberhettinger [181]. (8.
146 CHAPTER 8. then the series is said to be an asymptotic expansion of f(x) in t h a t range if for a fixed value of n lim \x\—>oo ft xn\f(x)Sn{x)\=0 3. 8.4x2 (8. whereas e~x '4 increases exponentially. Whittaker and G. E.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.64) (c) for —7r/4 > a r g £ > —57r/4: Dv{x) e 4x !*£=£2 2x i/i + +• (I/1)! + xe 4 x (i/+ !)(„ + 2 ) 2x 2 " ' (8. N.+ — + ••• of a function f(x) for a given range of a r g x . Thus the series expansion (a) in this case is multiplied by an increasing exponential and the asymptotic equality " ~ " means an exponentially decreasing contribution has been ignored.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. Watson [283].1 a2 an a0 + — + ^ + . 4 (8. T.0. 349. TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . . Thus there is no contradiction^ between (a) and (b). top of p.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.67) Cf.
8.68) The definition(8. (8. lim xneW x—>oo = 0 (8. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. In fact. 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. in particular with regard to possible exponentially small contributions to the expansion for which e.73) Vn + q/#> + e2T/i2) + • • • .69) and the consequent nonunique definition of the expansion. 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. (8. A critical discussion of the definition can be found in the book by Dingle [70].5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .g. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. which is assumed to be small.Sn{x))  = oo n—>oo (8. first published in 1902. The definition is due to Poincare (1866). 8. possesses a whole chapter on asymptotic expansions. we set E^En * > ^ n = En°1+eEnV = + e2EnV + .68) is (effectively) simply a statement of the observations made at the beginning.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. .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). i.e. Chapter I.
(879) We multiply this equation from the left by ip^ and integrate over x. (8. We insert the ansatz (8.e.75) HoW + vW = E^+E^+E^. (8. i.148 CHAPTER 8.i4 0) )V4 1} = {EM .74) (8.4 0 ) M = / ^C(4 1} . We obtain then the following set of equations: Hrj.As a consequence the state vl/) is a vector in this space. i. Ho^+V^n = EW^+E^n.V)i. 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.78) into Eq. i. (8.n = 40)V>n. (8. The second equation can be rewritten as {Ho .e. With the help of the orthonormality condition (8.w » . so that we choose / • dxrpM^O. TimeIndependent Perturbation Theory We insert these expansions into Eq.e. .77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .76) The first equation is that of the unperturbed problem. 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. J f°° dxrn E(^ 0) . (8. ^ ) = 0.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.n.70) and equate on both sides the coefficients of the same powers of e.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n. °° i J or 0 = EWJdx{rnVil>n). ( ^ .
(8. (8. (883) Thus the procedure forbids the equality of any E\ '. (8. (8. i. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ .5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.79). multplying Eq.Vil>n) (880) In order to obtain ipn . ^ n ) + 0(e 2 ). the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. with En .V)^ Setting + 42tyn. (8.i ^ 0.m ^ n and integrate: oo /•oo / oo dxrmY. S ' ti^+Q^)• We observe that Eq.83) makes sense only if Ef] ^ EnV for all i + n.4 0 ) M 2 ) = ( 4 1 } .e. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi.e. we return to Eq.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above). i.76) which we rewrite as (H0 .^(EW ~ V ) ^ + E& / J—oo dx^miln . We multiply the equation by ipm.8.Viprt e (8. We therefore consider Eq.82) *> = ^ + E . the coefficients a\ . (il)i.
VV .Vlpn)(lpn.Vlpn] M _ P (0K 2 ^r .87) (i>k. (8.85) together with (8.90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .89) • • • where we guessed the term of 0(e 3 ).(1) (i&0) .e.88) (Q) Inserting Eq.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.V4>n){lpn.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.^j)(V . FVn) + e 2 ^ ( V > n . ~Ej E E.^w .79) •/—oo oo / Using (8.Vlpn) E^ .V^n) . n) £f) (8.£i°») 2 Hence +^£o (£«? . (8. j.81) we obtain (2) _ _ {lt>m.150 CHAPTER 8.Vlpn)E.B i 0 ) ) ( £ f .V^j)(^j.4 0) ^+ e 2 E (lpk. En0) (8. ipn ) = 0 i.V'>pTl (E^Eny V'fc (8. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.2 (V> m .86) into the expression for En we obtain En = 4 ° ) + e ( ^ n . (8. V ^ i ) X (Q) (Vi. Inserting the expression (8. ^ n ) j&iift + ^n .^ ) + $& w £ (^fc.^(8.
(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. i.90) is the term with only one summation J2 m the contribution of 0(e 2 ). (8. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.8.7 is based on this procedure.g.91) The coefficients Cj follow from the recurrence relations.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0). An ugly feature of the expansion (8.89).e.Vlpn) = y^Ci(lPk. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. We then obtain (lpk. One would prefer to lump it together with the other contribution. We can therefore rewrite tpn ': The perturbation theory described in Sec.lpn+i) = ^ Cj5k. 8.90) can be found in just about any book on quantum mechanics..5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. The functions ipn(x) are eigenfunctions of some secondorder differential equation. . Merzbacher[194]).n(x) = J2crtn+ii (8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w.
e.93a). Em can now be linear combinations of ipni'ipm.V)(c2l^n + ci 2 ^ TO ). (8. (8. (8. (8.ipn).97b) with ipn y£ ipm. i.70). however. (8.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• .94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. . (8.96) are effectively the known equations of ipnitpm and so yield no new information.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . + c22^m).96) (Ho .94a) into Eq.97a) and (8. (8. all n. m = I.e. (8. so that En = 4 ° ) + eE™ + e2EP + ••• . —I.e. are allowed to be equal since otherwise the expansion becomes undefined.. that of double degeneracy En = Em . Equations (8.93b) The lowest order eigenfunctions belonging to En. I — 1 . Equality of eigenvalues.97b) are .. .4 0 ) ) ^ } = (Em] .97a) (8. TimeIndependent Perturbation Theory 8. i. .94b) with (ipn. in a central force problem all have the same energy unless the system is placed in a magnetic field.ipm) = 0 = (^m. i.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.93a) (8. . (H0 + eV)^n and (HQ + eV)Vm = Emym.152 CHAPTER 8.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8.93b). is not uncommon. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.We now insert Eqs. For example the magnetic states enumerated by the magnetic quantum number m. degeneracy. In order to deal with the problem of degeneracy we consider the simplest case.94a).95a) (8. (8. . Em = 4 ° ) +eE$ + e2E$ + . Equations (8.
cn(ipm.V^n). Vi/>n)] . Vlf>m)] .cnVipn and tyi.C22(Vn.J#»a&> = 0j since ii4i = En (degeneracy). Wm. (8. We now multiply these equations from the left by ip* and integrate. H0lP$) = C22[E$ ~ (Vm. ^ ) ) = CU[EW .£ < « = ag)^) .(Vn.H0xl>$) .^ ( ^ .H0^) . (^n.^ ) ( ^ m . and ( ^ m . V^n).ipm ) the equations (iPn. Similarly Wn. Then (^. V^n)\ . Vl/.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .6 Degenerate Perturbation Theory 153 inhomogeneous equations. (^m. £ aW^°Vi) .ci 2 (V„.Vil>m)] V^m). H0lpV) .98b) + Culpm) + cuV^m).(lPn.EJ®(lPn./>£>) = C2l[E£> .m.m).C 21 (^ m .98a) Setting I = n. ^ ) = ( V .8. . t f ) = a k 1 } 4 0 ) " Eg>aU = 0.y>m. c2iV^n + c22V7pm).ipn n) = 0 = (ipm. we obtain with (ipn.99) and ( ^ ^ O ^ ) " ^ ^ .H0il>W) = cn[EW .(^04°))41)) = EM^CuTpn (ipl. Equations (8. But (8. (8. ^ o ^ ) ) = 0.
E)tp = 0. TimeIndependent Perturbation Theory (l/>m. the degenerate perturbation theory must be used.100a) determine the coefficients Cn. as we saw. It is then possible that a special perturbation. splits the problem into two independent onedimensional systems. At x = ±00 the constant is zero. Let ip and < be two eigenfunctions belonging to the same eigenvalue E.Vrl>n) ..e. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.6. i.E™ (Vn. If one wants to perform a perturbation calculation in such a case.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy.e.101) This secular equation determines the first order correction En • The equations (8. d>" E=^. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. 4. in particular in parabolic coordinates (see Sec. if)' (/>' — = —. . Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V .3). (8. A onedimensional system. 11. .Vlpn) CHAPTER 8. (8.100b) determine C2i. 9} ip.^ ' V = — (•>!>'<t><t>'i>).Similarly Eqs.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. \nip = ln(j> + c o n s t . like that providing the Stark effect (proportional to rcosO). However.ci2. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). Example 11.C22.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. (V>V</>V>) = const. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. It follows that ip and <j> are linearly dependent. dx i. Example 8. i> 4> Hence there is no degeneracy.154 and V (^m.1. i. Then / > i>" 2— = V ip Hence 0 = V ' V .e.^Vm) 0. In this particular case therefore nondegenerate perturbation theory can be used (cf.5). the secular determinant. ip = c<p.
W. also possible in areas not of immediate relevance here. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.t Since in general a solution is valid only in a limited region of the variable. B. We assume that V can be expressed as the sum of two terms.104) are chosen such that the equation Dcf) + (E0 . W.106) (8. The subdivisions (8. T For instance spheroidal functions can be studied with this method. V = V0 + v. It is immaterial here whether D does or does not contain a first order derivative — in fact. Thus the method is a systematic method of matched asymptotic expansions.103) such that if E is written correspondingly E = E0 + e. the method has to be applied in each domain separately. with some restrictions on the function V. J. Muller [73]. (8.105) (8.103). anharmonic oscillator potentials and screened Coulomb potentials. We consider an equation of the form Dcj) + (E . Muller [216].7 DingleMuller Perturbation Method 155 8. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. a potential. of course. Eq.102) where D is a second order differential operator and E a parameter. Applications of the method are.e)<t>. B. i. in particular in applications to periodic potentials. Muller [210]. J. cf. W. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. J. (8. (8.104) "This method was developed in R. Dingle [72] and H.102) can be written D<f> + (EQ . . and the solutions then have to be matched in their regions of overlap. the considerations given below would be valid even if D contained only the derivative of the first order.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. and such that one can even obtain a recurrence relation for the coefficients of the expansion. Dingle and H.V0)<t> = (v.8.V)(p = 0.V0)<f> = 0 (8. See also R.* In various versions this method is used throughout this book.e. (8. H.
m + .e ) ^ = 2 ( m .) This result is obtained as follows.m + s)4>m+s s (8. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . since D<f>m + (Em .VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . We write the coefficients "(m. Considering (frm+s we have. m + s)0m+s.e\ < \V0 . the equation Dcf>m+S + (Em+S .+s ' (8.V0)<f>m+a = 0. v = {D+EmVo)<t>m. (8.102) with E ~ Em. and so v (8. the function 4>(0) = <t>m (8. (8. s an integer: (ve)(pm = 'Y^{fn.109) D(j)m+s + (Em .111) s then (see below) 0(1 = E (m.108) with coefficients (m.107) represents a first approximation to the solution (f> of Eq. 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.Em+s)(/)m+s.E0\ (the latter in a restricted domain).V0)<f>^ .Vo)(j)m+s = (Em . Then if ]e < \E\ and \v . m + s) which are determined by the recursion formulae.113) .156 CHAPTER 8. where m is an additional integral or nearintegral parameter (depending on boundary conditions).m + s)" in order to relate them to "steps" from m to m + s. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.V0)4>m = 0.110) If we write the next contribution <f>W to <p(°> in the form m+st (8.
115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.n\ ^119) + .T? S The first expansion determines the solution 4> (apart from an overall normalization constant). (8.119) follow a definite pattern and can therefore be constructed in a systematic way. We observe that the coeffcients of expansions (8. Thus (m. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. . .m + s)(m + s. (8.110) and are therefore related to the inverse of a .m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .8. (8.m + s)(m + s.m + s)(m + s. .^ (Em . the coefficient of the term (m.m)4>m in (8. /01_. The factors (Em — Em+S) in the denominators result from the right hand side of Eq. (8.Em+S) = 0 (second approximation).m) — 0 (first approximation).„N This procedure can be repeated indefinitely..118). (8.Q (m.7 DingleMiiller Perturbation Method and so (m. (8. . £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.m) E J^ ( m ~ Em+S) ' • fa. (8.m) — —.e. .m) 1^1^ IW .m) + > ^ .102) to order (1) of the perturbation (v — e).119) is effectively the secular equation).TP \(T? .m + s)(m + s.m + r)(rn + r. V ^ V ^ {rn.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. Eq.117) „ . We then find altogether ^  l^m  Em+S) + y^y^ SJ.116) m ^ v (m.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.113) must vanish.
i.158 CHAPTER 8.2). S. (f>m in the above. and is typical of a large number of cases. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. W. Other applications can be found in the literature.[ 4 ( 8 5 g 4 + 2q2 . since it will be used extensively in later chapters. . . 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. as already mentioned after Eq. . (8.9) + 4096J4] + 0 ( l / g 3 ) .120) n4 . Kaushal [146]. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. 1 . The eigenvalue expansion obtained is an asymptotic expansion. TimeIndependent Perturbation Theory Green's function.2: Eigenenergies of the Gauss Potential^ Use the above perturbation theory for the calculation of the eigenvalues E = fc2 of the radial Schrodinger equation dr2 + k2 1(1 + 1) V(r) ip(r) = 0 with Gauss potential V(r) = g2e~a r for large values of g2. for the investigation of the behaviour of the late terms of the perturbation series.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. § H. i. J. The following Example 8. The result is the following expansion. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. (8. as will be seen later. An additional aspect of the method is the full exploitation of the symmetries of the differential equation.I2q + 64) + 256Z 3 (41 g . This means the solution is not normalized. has been treated by R. This can in fact be done and will be dealt with later.6q + 1)1 + 24(5? . for instance. this is — so to speak — the price paid for obtaining a clear systematics which is essential. in which q = in + 3.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential.[q(llq2 + 1) + 2(33g 2 . . does not occur in any of the later contributions. An explicit application of the method (not including matching) seems suitable at this point.1)( 2 + 64J3] — . ^The potential A r 2 / ( 1 + gr2).* Example 8. n = 0 . Miiller [211].38).e. as will be seen in later chapters.423) + Z(2720 9 3 . 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . 2 . for instance.
(. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2. a .2 '(l + l) .V ! 4 I z S *! V2 ipq(z). The solution i. S) is a confluent hypergeometric function. (fc2 + <?2) 4ga and b = I \—. 2 .b._ „!2 ^ ( " « 2 r r 2 > . Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . i/> with Dq = d2 . „2 J„2JL. b.2a2 A. Therefore in the general case we may set k2 + g2 = ga(2l + q) .122) where A is of order 1/g. 1 . 2 The solution XQ(S) = $ ( a . . . We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. 3\ 2 V ' 2. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar.l)ip(a . . (8.1). The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.6 and 16. 11. Setting q = in + 3 implies k2 + g2 ga(2l + q).123) the contribution fl<°> Ah + .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 .z2 is a normalizable function if a = —n for n = 0 . multiply the equation by onehalf and set h = —a/g. Sees. (8. \2 . a)ip(a) + (a.121). 1 1(1 + 1) 1 . (8. .2) where 1/ . .0(z) = zl+1ez2^fa.. We now insert this equation into Eq. .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.b.123) The first approximation (note the last expression as convenient notation) ^ .a + l)if>(a + 1) + (a.8.
a . a + r) satisfy the following recurrence relation: Smi(a. a) = 1. fa. a + r — l)(a + r — l . a + i) = 0 for i ^ 0. of course. o + r) + S m _ i ( a . a]i + h fa. So (a. This equation shows that a term fiip(a + j) on the right hand side of Eq.e.a—lli.b = (g + 3) . ( 5 m ( a . (a.2a = I + q. Now we make the following important observation that Dqip(a + j) = jip{a + j).a]2+L ^ J1[a + 2 .i ^(i)=£V+i i=0 J2 j = (i+2).a — 211 . a + i)(a + r. . TimeIndependent Perturbation Theory (g . one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context.a]i = . _ J [a .Z.120).a + 2]i P . 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. If we now evaluate the various coefficients in the last expansion we obtain the result (8.1) = a . . where [a.120).l . a + j)i/>(a + m). In fact. a ] i + L ' ^ J 1 [ a + l . a + 111 r [a.3). a + 1) = a = In general CHAPTER 8. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a .a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". 0 = h[a. when j = 0. fa.S m (a. (8. a + r) = 0 for r > m. a + r + l)(a + r + l . a ] i + 0(/i3) [a.160 where (a. one could now use the RayleighSchrodinger method and rederive the result (8. Now proceeding as explained in the text one obtains finally the expansion ip = ^(°) + v ( 1 ) +y>(2) H — along with the expansion from which A is determined. a + r) with the boundary conditions Sn(a. Hence the next order contribution to tp^ becomes oo i+2 r [a a .a + j}i+1^(a + j). the latter determining to that order the eigenvalue. The coefficients S m ( a .j^0 ' + jU+1 J ^(a + j) with h[a. As a check on the usefulness of the perturbation method employed here. a + r) + S m _ i ( a . a ] x + L _ 2 J 1 [ « .2.a]1=0. (a.^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously. i.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. „ . a) = b .
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. proceeds along similar lines. 9. 161 . except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. and we shall then see that these lead to analogous results as in Chapter 2. We therefore begin with the appropriate classical considerations with a view to generalization. as we shall see in Chapter 27. We know from classical electrodynamics that a planar light wave with propagation or wave vector k and frequency u> is represented by a vector *So far our considerations were restricted to oneparticle systems. light.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. although not exactly straightforward. i. and this means electrodynamics. the generalization to many particles. 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. Regrettably this has to be left out here in view of lack of space.Chapter 9 The Density Matrix and Polarization Phenomena 9.e.
A 0 = A0R + iA0I.1 The vector potential ellipse.4) (9. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.t) = 3ftA.1) Fig.i). When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized. (9. 9. t) given by A(r. i. 9.3) (9. In general the constant Ao is complex. This is the equation of an ellipse as indicated in Fig.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.2) (9.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0.1. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.e. (9. t) = A 0J R cos(cji) + A 0 / sin(u.6) . (9.162 CHAPTER 9. The Density Matrix and Polarization Phenomena potential A(r. It follows therefore that A(r.
i.2) the vector potential A is completely determined by its two (say (x. . p. Rae [234]. The intensity 7 of a light wave with amplitude coefficients A\' . for instance.y)) components A i .8). we recall. 17. represented by a filter "F". a calcite crystal or a film of nitrocellulose.t in which k and ui remain unchanged. We set therefore A 0 = Ai + A 2 . AW = £ fc=l. the direction of polarization.2 *i*4 0) (98) with coefficients F^.11) 'This means we consider the polarization. the intensity of a wave. the vector potential A. (9.k (910) We give the initial wave the label "0" and subject this to an experiment. Then their connection is given by Eq. not. (9.AW).A2 is the sum of the moduli. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. Observable quantities can only be those which are invariant under translations and rotations.A«). See for instance A. the fields E and B are observables. i. Here we are interested in observable properties of light waves — like. 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. From the vectors A(°).e. I=<l2x2>:=E40)*l*40)i.A 2 .9. (9.e. (AW. which leads to an outgoing wave which we give the label "1". determined with the help of a polarizer. the connection must be linear. with k . (A(°). A2. i.A« we can construct only the following scalar products with this invariance: (A(°\A(°)).7) We consider now measurements with polarized light. however.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.e. In classical electrodynamics.A i = 0 = k . ie.A(°)).g. (9. as e.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. (AW.
13) the quantity F must be an hermitian matrix and thus representative of an observable.e.A( 1 )) = ^ ^ F i.k l f c 4 ° \ (9.A*kF£A* = E and hence F = Fl A p i tkAk. pki = AkA*.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.13): j2^FikAk = Y^AiF*kAi = Y. i.164 CHAPTER 9. (A(°).A*FikAk. (9. Thus. Then according to Eq. (9.17) . Thus (F) = J2FikPki = ^(Fp)a. i.k (9. so that (F) can be looked at as the expectation value of an observable F. (9. (F)=Tr(Fp).16) i.13). But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. We use definition (9. the connection between the initial polarization and the final polarization. i. (9. To this end we observe that as a consequence of Eq.12) i.e.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. (9.e. (9. if (F) is an observable quantity. We write the observable quantity as (F):=Y. it must be real and hence (F) = <F>*.k i (9.15) in order to rewrite Eq.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%.
b = 0 and o = 0. In the present case of a polarized beam of light we have two possible. i.22) For a — 1.* For a lightwave travelling in the direction of z. This admixture is described by introducing matrices p.21a) follows from Eq. b = 1 we obtain the expressions (9. that of a pure state.18) The vector A describes a particular ray of light.21b) This state is still a pure state. A2 = b as </*) . 165 (9.e. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. For a 45° polarized wave we have 1 .e. i.( J o ) .9. . "incoherently. is in general unpolarized. (9. i. like that of a bulb. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. we have (9 19) ' In this case there is no preferential direction. (9. 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. is given by a single polarized wave. The density matrix for the pure state (9.15) with A\ = a. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .e. <»^> (9.d "»=(o ! ) " In the case of an admixture with equal portions of 50%. But light.(AAt) = ( £ £ ) .19).2 Reconsideration of Electrodynamics In particular we have / = Tr(p). The socalled pure case.
(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". Then according to Eq. 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 .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. Equation (9.24) The following two 50% admixtures yield the same effect: 1 1 (\ 0\ P = 2P*+2Py=[Q 1 P = 2^ 4 5 ° + 2P135° i J. in which the intensity I = Tr(p) is not conserved (e. b = —=. 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. (9.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. We set in analogy to Eq.8) A(1)=E^40)> k R = R(<*). Pi35° = \ \ 2 ? )• (9. l l\ " i °\ ) ' ^925^ 9. (9.23) 1 . as a consequence of absorption).Z^ 3. 1 = . We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.™ V 0 n kmAm / ^^jPim^mk' .166 and CHAPTER 9. (9.g.
The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.9.e. the observable is transformed. In the case of Eq. (9.pW). In the case of Eq.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9.q. (9. 9. (9.34) . in other words without creation or annihilation of systems. i. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. the dependence on a is contained in F' = R^(a)FR{a).30) Ti{FRp{®I$) = Tr(F.4 The Liouville Equation 167 pW = RpMRl (9. in the state functional) in analogy to the time dependence of W(p. that the initial intensity Jo is equal to the final state intensity I\. (9. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)). (9. i.e.32) (9.e.33) i.t) in Chapter 2 in what we described as the Schrodinger picture there. here this contains the dependence on a. The corresponding condition here is that no absorption of the wave takes place. and p(0' remains unchanged. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . (9.28) (9.31).31) (9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.29) where (i.e. the observable F remaining unchanged. on the other hand.
(9. (2. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. We set therefore M^:=iH. the matrix M is antiHermitian: Mf = .39) If the dependence on a is not contained in p^1' but in F' (cf.F}.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.2) has to be taken into account. j£ = i[H.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.46) suggests to define Poisson brackets of matrices by the correspondence {A. Eq.B}:=i[A. (9.40) can be considered as quasiequations of motion. In view of Eq. . i.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 2.40) Equations (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.39) and (9. (9. Comparison of these equations with Eqs.B].35) Thus the matrix R has to be unitary.p(°\ao)]da.37) (9. (9.21) and (2. (9.e. CHAPTER 9.168 i.M . = Tr(R^Rp^).p]. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1.36) (9.e. (9. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. (9.
(10. and consider timedependent perturbation theory. 10. the Schrodinger.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. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor.Chapter 10 Quantum Theory: The General Formalism 10. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u). we establish their Heisenberg equations of motion.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Thus in particular we establish the uncertainty relation for observables in general. Heisenberg and interaction pictures. To insure that the 169 .1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real.
Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.y.Pi are called fundamental observables.1 U n c e r t a i n t y relation for observables A. we must have (u\u) = 1. y. E.3 represent the three Cartesian coordinate directions of x.170 CHAPTER 10.4) *Even in those cases where this is possible the operator correspondence may not be unique. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space. q).g. Such a correspondence with classical mechanics is not always possible. \pi. Here. 10. (10. where i = 1.Qj] = 0.Pj] = 0. however.q). we assume this now. The phase space coordinates qi.Pj] = ihSlj.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. [qi. and assume that A = A(p.2) (F(A)) = {u\u) <"lf. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. such as for instance of the angular momentum operators Li.* a conspicuous exception being for instance a spin system. Pi • Pi. . which clearly does not hold in the case of operators.2.B] = ih. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. Naturally one can also write (10.2. We can describe a system as one with a classical analogue if the Hamilton operator of the quantum system is obtained from the Hamilton function of classical mechanics by the correspondence •^i • Qi. (10. classically qiPi = PiQi.
2A(A)}  {A)2)1'2 (10. (10. AB defined by the mean square deviation AC=((C2)(C)2)1/2. using Eq.2 States and Observables then their uncertainties AA.B.12) where \u) 6 "K is a ketvector representing the state of the system.12) the Schwarz inequality (cf. (10..7b). Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).B] = ABBA > ^h. Then [A. = \\A\u)\\2\\B\u)\\2 (10.7a) (B) = 0. Ai = = = ((A2) .{A)2)1'2 {{A2) . (10.2(A)2 .5)) (for A\u).7b) = = [A.(A)2 = (AA)2.9a). we have A2 = A2 . Eqs.8) and.10.(A)2)1'2 (1 °= a) ((A2 + (A)2 .9) ((A2) + (A)2 .1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). (3. so that (A) = (A) .11) (A2) = (A2) . (10..10) . (4. 111 (10. Applying to the expression (10.(A) = 0. Since A = A— (A). (A{A))(B{B)){B{B))(A(A)) (10..6) B:=B(B).13) AB = AB = (B2)1/2. (10.B\u) G IK as vectors) we obtain (for an application see Example 10.2A(A) + (A)2.0)1/2 = AA. Thus AA = AA = ( i 2 ) V 2 . C = A. (10.. (10. B}= ih.
(10.13) (AA)2{AB)2 i. (10. (10. (10. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. Zi Zi (10. From this we obtain (uiJBu) 2 = i22 + / 2 .18) >R2 + I2. Zi (10. (10.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). We can therefore rewrite Eq. The condition (a) is satisfied when. Zi + BA)\ + ^ih. since A and B are hermitian. and with Eq.14) Hence (u\AB\u) = /^(AB where. (AB + BA) This means (AB + BA) is real.e. .17) as had to be shown. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. and hence for condition (b) 0 = = tSee Sec. (10.15) as (u\AB\u) = R + iI. for a complex constant c.4).15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*.172 CHAPTER 10.3. 4. A\u)=cB\u). and (b) when (AB + BA) = 0. (10.16) I = h. R= (AB + BA). AAAB>I=h. The equaltosign applies in the case when (a) it applies in the Schwarz inequality.
= 5R2 + G 2 > 9 2 = hi2(Lk)2. the following uncertainty h2{Lk)2. in view of Eq. Separating the product into symmetric and antisymmetric parts. ALy = (u\L±\u) ^ 0. (Ly)=0. i.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li.20) or. For these we have in the onedimensional case the relation [p. i. i.3 OneDimensional Systems 173 Thus c = — c*. + LjLi\u). q) = ih. (Lz)2^0. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. The observables like angular momentum are functions of p and q. = 0. For states \u).Lj] = (LiLj + LjLi) + ih£ijkLk. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. What does the relation imply for states u) = \lm).y.e.e. (ALZ)2(ALX)2 > \h2 = (Ly)2. Ly are not "sharp". {u\LiLj\u) = R + iQ. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.) 2 > = x. . h2(Lz)2. i. (Ly)2=0.e. m\lm)).7a).13) to operators Li — Li. (A{A))\u)=i(Zc)(B(B))\u). 9ftc = 0. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0.10. (Lx±iLy) ^ 0.z. implying {Lx)=0.i relation holds: (ALi) 2 (AL.e. Thus in this case Lx. for which one component of L is "sharp"? Solution: Applying Eq. 9 = h{u\eiikLk\u). Example 10. such that Lz is "sharp". ALX 10. (10. (10. (ALy)2{ALz)2 > ~h2(Lx)2.
(10. This is why the operator p requires a representation in the space of eigenvectors of q.p] = ih we obtain therefore [q. Eq.21a) Equations (10.21b) now follow with the assumption that A(q. Similarly we obtain \p. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). they do not have a common system of eigenvectors.e.p] = ^ihp ~ Proceeding in this way we obtain [q. Q2] = [p.pn]=ih^(pn).A(q. We have (cf. Then (see below) we can derive for A the following commutator equations [q. i. q]q + # > v] = ~2ihq and \p.q).p) can be expanded as a power series in q and p. and (c) the spectrum is not degenerate.23) = ihg(q2) (10. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. .174 CHAPTER 10.B]C + B[A.34d)) [A. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators). (b) its spectrum is necessarily continuous. A — A(p.p}p + p[q. Next we have a closer look at the operator q.BC] = [A.p)] = .qn] = ih^(qn).A(q. (3. We obtain these equations as follows.p2] = [q.21a) and (10.p)]=ih^^and \p. Using [q. i. as remarked earlier.C\.i h ^ ^ .21b) (10.e.22) d ih—(p2). (10.
27) (10. = eipa/h = Uia)'1 = U{a).21a) A by U. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. 10.10.26) it Here p is an observable. > We begin with (a).e. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. and a a cnumber.f^ ^ J—oo dke ikx —ikx' (10. U] = ih^= aU. we obtain [q.1 The delta function for K — oo. (10.e. an operator. U is unitary.25) It is shown in Fig. (10. 10.3 OneDimensional Systems 6K(x) 175 Fig. i. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1. Replacing in Eq.28) i.1 how the delta function arises in the limit K —> oo. .24) lim SK(x — x') — S(x — x').
In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. cf. (10. the spectrum is not degenerate. (10. (10. we can see that also p possesses only a continuous spectrum (from —00 to 00). Evidently every eigenvalue has only one eigenvector.31) . which means that the spectrum of this operator is a continuum.176 i. or qU(a)\x') = (x' + a)U(a)\x').30) (10. i. 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. WU = 1.29) This means: U\x) is eigenvector of q with eigenvalue x + a. q any other observable F(p.29). 10.00). and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p. q) can be constructed representing a dynamical variable. CHAPTER 10. Thus all these values belong to the spectrum of the operator q.q)\i>) = (i>\U^UA(p. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').e. or qU\x) = U{q + a)\x) = (x + a)U\x).3. b a cnumber).e.1 T h e translation operator U(a) We saw above.00). Since this holds for any arbitrary value of a in (—00. ^Observe that {iji\A{p. that with U(a) = e~ipa/h = U. Eq. we obtain qU\x) = U(q + a)\x) = {x + a)U\x).q)U^U\ilj). Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).
(10. we have q\x' + a) = (x' + a)\x' + a). \c{a. U(a)\x') = c\x' + a). the operator U(a) acts to shift the value x' by the amount a. The operator is therefore called translation operator.e.10.x" .38) h 6(x' . (10.5(x> .e.37) With the help of the latter expression we can calculate. (10.x')5(x" = (x" + a\c*c\x' + a) x').e) ~ 5{x' . for instance.32) we see that \x' + a)ocU(a)\x'). It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a)./ (10. (10. (10. Comparing Eqs. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .x') = = i.e.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . (x'\p\x").e) (x \p\x ) = .e.x')\ = l. We choose the phase such that U(a)\0) = \a).a). offdiagonal elelemts of the {x}representation of the momentum operator.31) and (10. In the expression of Eq.33) (10.x" .x") .x") U{x'\p\x" % pz.39) .x" . i.3 OneDimensional Systems 177 Moreover. (10.36) i. it follows that 5{x" .hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). since q\x') — x'\x').e. i. i.x")c(a.35) (10.32) Since U is unitary. (10. \x') = U(x')\0).34) {x"\U\a)U{a)\x') c*(a.
(5.° ° d(p) (ip) e"« = S'(x).40) 10. In this case the classical Hamilton function does not depend explicitly on t. (10.to)m0)). to) with respect to t we obtain ihjtU(t. it follows that 5'(x) is an odd function of x.L fdpeipx.tQ) = eiH(tt<Mh.e. (io. .x") = (x'\p\x")*.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. Quantum Theory: The General Formalism We mention in passing that.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . This is in conformity with our earlier postulate in Sec.e. Differentiating U(t.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. i.4.27) and (5.t0). 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.4i) We assume first of all. H\uE(to))=E\uE(to)). i. .44) (10.45) (10.to) exists which is such that^ m)) = u(t.f dpipeipx. (10.35). 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)). so that U{t.178 CHAPTER 10. a conservative system.e. 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). Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.to) is called the time development operator. since 5(x) = . 5.t0) = HU(t. The operator U(t.43) (10. 6'{x) = !. See Eqs.
47) e.to) (differentiating we obtain immediately Eq.t)l]U(t.t).to) can be expressed as an integral as the solution of Eq.*l)^(*l.tiM*i.4 Equations of Motion 179 We now demonstrate that Eq.t) = l^eH{t).t) and therefore [U(t. = U(t0. The operator H is d efined by this relation.t) = l.to) = 1.t)U(t.45) has very general validity. (10. t0) (10. (10.t0)\^(t0)).t 0 )^(*0)> U(t2.t0) or ih—U{t. (10. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. Evidently U(t. i.to).41) we obtain ^(t 2 )> = = J7(t2.48): [/(Mo) = 1 .t) = 1. From Eq. But obviously (cf. Eq.48) with U(to. t0) = E/(t 2 . Hence U(t.t0) [U(t + e.t0)]1 Let us set for small values of e: U(t + e.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.44)) we have U(t.z / dt'H(t')U(t'. (10.*l)^(*l)> = I7(*2. t0) = H(t)U{t.48)).t0)U(t0. Then with U(t + e.10.e. Ufa. (10.49) . (10. (10.
t0)\ii>s{to)) (10.51) Since </. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.e. it follows that U(t + dt.41). Quantum Theory: The General Formalism We obtain the differential equation for the development of states of the system in the course of time by differentiation of Eq. (10.t) = l % ^Hdt) \il>(t)) Hdt (10. >(*)> = i. the instant at which a measurement is performed on the system.5o) ihjtm)) = Hm)). the Schrodinger equation ju(t. Eq.53) . i. 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. Then \Mt)) = u(t. The operator U(t. The eigenvectors of these observables are also constant in time. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. Since the only measurable quantities are moduli of scalar products.t0) m0)). Comparison with the equation postulated earlier.30). Another such description is the Heisenberg picture. One describes as Schrodinger picture the present description of a system in which the state of the system is represented in terms of a vector \ip(t)) which changes with time t. t)\t/>(t)) =U~ and since H is hermitian. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.180 CHAPTER 10.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). (5. (10.e.(£ + dt)) = U(t + dt. requires the identification H(t) = H= Hamilton operator. the predictions of different descriptions are identical. and these remain unchanged under unitary transformations. (io. Now let \tp(t)) be the state of the sytem at time t.
However the associated observables are timedependent. (10.HH]. (10.55) Thus AH is explicitly timedependent. {AH.58) HHAH (U*ASU)(U^HU) [AH.54) Observables transform correspondingly with a similarity transformation.HH].H]U + iW]^U. In the Heisenberg picture it is HH = U]HU. 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.e. but here.56) U^[As. (10. For As(q. Here H is the Hamilton operator in the Schrodinger picture. It does not always have to be the case that As does not contain an explicit time dependence.10.H]U = = = U]ASHU AHHH — U^HASU . i.i).p. we consider only cases without explicit time dependence of As.e. Differentiating Eq. Thus. subject to a continuous change.e. (10. t0)AsU(t.4 Equations of Motion and \^H) = U\t. for instance in cases of nonequilibrium or if part of a system disappears through absorption. 181 (10. (10. As(q. . if ^ = 0.p) in the Schrodinger picture dAs/dt = 0. = so that ihjtAH or ihjtAH = = [AH.(tf HU)(rfASU) (10.48). we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10. in order to obtain the Heisenberg picture (description). i.57) and rf[As.With explicit time dependence. a scalar product) remains unchanged: AH(t) = U\t. as a rule. we have to take into account the contribution dAs/dt.t0)\ips(t)) = \ips(to)) = constant in time. t0).55) and using Eq. even if As is not timedependent. so that a matrix element (i.59) This equation is called Heisenberg equation of motion.HH]+ihU^U.
In the treatment of the pictures we used unitary transformations of operators (and vectors). 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. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. 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. Finally we add a comment concerning conserved quantities.21a) and (10. since in the transition from one to the other the commutator remains unchanged.e.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. we used in essence unitary transformations of matrices. provided the Poisson brackets in the latter are replaced by commutators. In our treatment of representations. [CH. In the Heisenberg picture essential properties of a system can frequently be recognized more easily in relation to their counterparts in classical mechanics — in both cases the time development of a system is given by the time dependence of the dynamical variables. in the case of the fundamental dynamical variables qi.H] = 0. . We end with a word of caution. (10.61) Thus conserved quantities commute with the Hamilton operator. Orfanopoulos [224]. (10.182 CHAPTER 10.13) to derive the uncertainty relation for energy and time. i. Example 10.HH}=0.21b). In particular. This holds also in the case of the Schrodinger picture.Pi we have in the Heisenberg picture (replacing in the above AH by qi.Pi): and where the expressions on the right follow from Eqs. if ffCH = 0. Solution: We have for an observable A (A) = (i>H\AH\TpH). A. We see that formally one obtains Hamilton's equations of classical mechanics. An observable CH is called a conserved quantity. In many cases the Schrodinger picture is more amenable for explicit calculations. We also observe that the momentum pi is conserved when \pi.
\HAM\2) \(nA. (10. AW = {A(A»V>. (10. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t.64) = 0. then if a measurement is made at time t'.HH]) Hence we obtain by ih(A).15): (AAf(AH)2 Using Eq. of A\<t>) = a<£)) are then called "good quantum numbers".e. > (10.H]) = 0. H] = 0.Such a characteristic time interval can be defined for any dynamical variable. 2 d{A)/dt " and.7a). if we set AH = AE we obtain TAAE (10. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) . dt in.62) and dAs/dt > \WAH\i. (10. Let \ip) = \"4>H) represent the state of the system. We construct the following vectors using Eq. i. since \IPH) is timeindependent.68) To put it The eigenvalues a of A (i. .13) and (10. is displaced by A A in the interval A T = TA.62) The Hamiltonian H does not depend on time. 183 dV ' \ dt >"•"»» + ( £ ) • (10.67) 1 d(A)/dt is oo and AE = 0. we can replace ([AH. (10.4 Equations of Motion Then.63) so that with Eqs.e. at ^±AH>h. i.10. the "centre of mass" of the statistical distribution of A.)\2 ( = \{i.e.66) This means. and TA O C (10. this is practically the same as at time t for times t with \t — t'\ < T. H\tl>) = {H(H))\f).H}\f)\2. This means that for a physical variable A(q. in a different form: The eigenvalues of an observable are called "good" if [A. a position or momentum observation is independent of t.e.p) we then have 0=^M) = U[A. i. If r is the smallest such characteristic interval of time.2 > h. (10.65) and rA AA d{A)/dt' . (10. (A).
t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . the wave function ^ ( x . (10.e.tyEt/hdt = = = / J—oo dt J dEJrl>E.70) we can determine the function fEo{E) integration with respect to t. as is observed. i ) = J dE'^El^)eiEltlhfEo{E'). Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. This means. (x)/ E o (E')5(E .E1) 2irhfEo{E)4>E(x).H ) ( £ .70) by From Eqs. This means that in the case of decaying particles. /oo (10. for instance. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".184 CHAPTER 10. First we obtain from Eq.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x.69) and (10. Recall for instance radioactive decay. i. (10.72) = V'EoW 7/2h + 0i(EE )/h  oo . have around EQ the form ipEofct) = i. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.70) we obtain: oo / oo TPEO (x. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.)exp(iEt/h)). states with sharp energy (AE = 0).71) From Eq.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1.Eo(x)eiEot/he'^2h9(t).t) — ipE(x. But the number of radioactive (thus excited) nuclei decreases exponentially. (10. the wave function ipEo(x. (10..69): oo poo p / oo *PEo(x. the particle density diminishes in the course of time (t > 0). (10. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.(x)fEo(E')e«EE'Wh 2irh f dE'^E. Quantum Theory: The General Formalism 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 .
74) Thus the solution of this equation is the main problem. that** EQ. 185 i. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. The state with lifetime r — h/'y is not a discrete state. Eqs. 7 > 0.e. (10.11). (14. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. (10. i.t0). (C/(0) unitary).41). (10. i. where U is the solution of Eq. (10. of course.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use.e.to) = HU(t. Considered as a function of E. 10. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2.g. and is called "interaction picture" or "Dime picture".76) **For specific applications see e. Let U^°\t. i.75) (10.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). ^(*o))> then \xp(t)) for t > t0 follows from Eq. every unitary transformation of states and operators defines a possible "picture"). They are called "resonance states".10. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum.e. (10.105) and (20.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). . We saw previously that if the state of a system is known at time to.toM{to)).e. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula.45). ihU(t.45) can be solved only approximately (in actual fact. to) be an approximate but unitary solution of this equation. £/ = £/W(Mo)^'(Mo). This description contains effectively parts of the other two pictures and is particularly useful when Eq. W)) = U(t.e. i. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. We obtained the result (10.
186 Then CHAPTER 10.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.79) ^'~°i. (10. i.e.e.e. i.h^U> = at ( io.t0) 0 = l.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. (10.e. [/' has only a weak ^dependence (i. Quantum Theory: The General Formalism ih^(uWu') i. with Eq.e. varies only slowly with t).83) ih^U^ ishermitian. H(°Ht) = ih^uW.78) (10. . with Hamiltonian H = H^ + H'. Now let H^(t). (10.e. In addition we have (10.e.81) But [/(°) is unitary. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0.82) (10. £/(°)£/(°)t = 1. i. = HU<®U'. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).
We also set H^U^H'UW. (10.84) U^ where H' is also hermitian. 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^.78) at (10 85) (10. (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.85) = where we used Eq. lfe(*)}.85) multiplied by U') on the right of Eq.92) . i.87) = We define correspondingly as interaction picture state iM*)> = t/ (0).10. (10. (10.90) = U^H'\^s(t)). (10.U^HU^U' H'fi'.86) at = U^HU^U'.80). H = H{®{t) + H'. (10.e. U<®\Mt)) = \Mt)). (10. (10. (10.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.89) where ^s(i)) and As are state and observable in the Schrodinger picture. U^H'U^U' (10.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).78). and obtain from Eq.
cf. i. (10. ot i. (with ihjtAI(t) = [AI. near zero and \tpi(t)) is almost constant in time. i.H\0)}. so that with Eq.89)^0. ^ (0) l^(*)>.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.188 CHAPTER 10. a perturbation so that H ~ H(°)). (10. (10.90) and replacing the left hand side of this equation by the right hand side of Eq.e.e.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. dAs/dt=0) . (10.94) _Hf)Al 1 + lhdAl+AlHf). H'j = UWH'UW.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. Since H'j is assumed to be small (i. we differentiate with respect to time the operator defined by Eq.55)) at + imm<Ms_u(o) at dt _umH(. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. we have i. (10.e.60a) or (10. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). (10.e. A^t) = U^AsUi0). In order to obtain the equation of motion of an interaction picture observable Ai. comments after Eq. (10. (10.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).60b) with iff0) instead of H. i.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .94) Then we have (normally with dAs/dt dt (io=8o) — 0.e.e.89). the vector varies with time (different from the Heisenberg picture).
94) Hi = uW(t. H0\ipn) = En\(fn). We obtain the operator U^ from Eq. We have the Hamiltonian H = H0 + H'. ^ \<Pn){Vn\ = 1n (10.e. where according to (10. (10. (10. i. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.7 TimeDependent Perturbation Theory 189 10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture.e.98).e.t0)H'uW(t. i.i0) = e<Ho(tto)/ft. (10. H' = H'(t).93). ih±\Mt))=Hi{t)\Mt)).102) We assume that before the perturbation H' is switched on at time to.87) and (10. We assume again that the spectrum and the eigenvectors of Ho are known.10. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em.100) (10.t0). {ipn\<Pm) = Snm. (10.97) The equation to solve is Eq. (10.99) (1098) Instead of the original Schrodinger equation we now solve Eq. Since . i.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.80) as E/(°)(t. (10.
105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10.102) and truncate the series after the first perturbation contribution. 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.^ V ™ ) . the time dependence of \ips } can be separated as for stationary states.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). (10. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). 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'.190 CHAPTER 10.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.104 ) Here we insert Eq. we can write l 4 0 ) ) m = elHot/hWm) =e . (10. (10. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 . Hence we set ^> = £°«(*)l4 0 ) >n.108) .106) We insert this expression into Eq. (10.107) Jto With this we obtain from Eq. (10. (10.103) for m — n and use Eq. (10.
111) Different from Eq.En (10.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . Thus we set H'(t) = H'0(t). We assume that the perturbation is switched on at some time to. (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. (10. This is the case. where the momenta of the aparticles form a continuum. if in the scattering of a particle off some target its momentum p changes to p'.109) 10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. » — < (10. < KO?t . (10.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En .109) the transition probability Wmn(t) = jp Jo .Em)t} \{tfn\H'\ipv h 2h(En .10.Em)t' (<Pn\H'(t')\pm) = Wnm(t). but that otherwise the perturbation is timeindependent. A further example is provided by adecay.110) With this we obtain from Eq.112) for a^0. t o = 0.8 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^. (10. for instance. or if by emission of a photon with continuously variable momentum an atom passes from one state into another. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.
we would have En — Em ^ 0. though sufficiently small. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . 10.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2.E7m)t} = (10. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .192 CHAPTER 10.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.Em). (10. Hence lim St(a) = 5(a). Otherwise. so t h a t Wnm would be zero.Em)t} r i 7 TTT.2 The delta function for £ — oo. t—>oo Hence also (shifting t to the other side) sin2{^(ffn. and no other state would be nearby. . Fig. if the energies belonged to the discrete spectrum.— = o (En 2h En 2h5{En . > We see t h a t this quantity has the same behaviour (in particular for a — 0) > as the function 5K(x) we considered previously.2.io. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. 10.
3: Application of test functions to formula (10.118) where 8e is the separation of states around En which in the case of the continuum goes to zero.r Z x 1 ^ ) = 4>{Q). the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. (10. (10. Schwabl [246].117) in many applications.e./ t>oc TZ J_00 dx . for instance from a potential.10. . '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. so that initial and final states belong to the continuum. Fermi [92]. Fermi gave this formula the name "golden rule" .* In view of the usefulness of Eq. 75. pp. f See E.8 Transitions into the Continuum 193 as equal. F Jo .t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . Then.115) and (10. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. 266. (10. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0).117) provides the outgoing particle current (see Example 10. Pauli [227]. then Eq.4).116) The transition rate V is the transition probability per unit time.W . Let p(En)dEn be the number of states in the interval dEn. sin 2 (tx) ax i—( * According to F. (10. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .117) were derived by Pauli in 1928. in the article of W. i.113) Use the method of test functions to verify in the sense of distribution theory the result (10. In the case of the functional (10. 142. _ 2irh AE > > Se. If we are concerned with pure scattering. We evaluate the integral 0(0) = 1.113). p. Example 10.113) this is* hm . if p(En) varies only weakly with En.
e " 1 * 1 = 2 lim — .<p).4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. . x .tan p 2 + a 2 .117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. . .? 2TT / L \3k2dkdUn 1 /V(x). .. Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . ^o n . Ryzhik.. .( ^ V V • < ih 2 _ i k n ./ cfc a n p 2 +fc 2 .. . 2 T / ^ M. with v the velocity of the particles.194 CHAPTER 10.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . § We have r Jo \p\x dx sin ax sin bx  V. K = k 0 . The ingoing particle flux is therefore v / L 3 . 77^ + . Jin § m = See I.ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. L \3k2dkdUn It follows that T. 491. E x a m p l e 10. and is obtained as ih. Gradshteyn and 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). iknx . .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). p = 1 we obtain I = . Then ^ . 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. p. with n . . Then p(Ek)dEk Since Ek — h2k2/2[i.„I. formula 3. * ^ = . = 0 . . E = h2k2/2fi.a 2 ' i that for a = b = t . S. M. 2 . dEk = h2kdk//j. . V +(ab)2 a _x 2p6 7 m —^ '.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn .947.. 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. Hence the result is 0(0) as expected. Quantum Theory: The General Formalism with the help of Tables of Integrals.k. . = ( — ) dk= 2ir (—  2irJ k2dkdfl. 1 . and hence .
122b) S e e also L. 148.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).122a) (10. we present another derivation using the usual method of timedependent perturbation theory.10.120) In the case of Eq.121) .t) = ^On(t)^(x. (10. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. (21.x)^(E'.x) = S(EE'). 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). pp. We expand the wave function V( x . (10. Thus we set ^(x. by which we mean without leaving the Schrodinger picture. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known. see the discussion after Eq. (10./continuum = <San(£)</4°)(x. 193.^ We start from the Hamiltonian H = HQ + H'.69). (10.^°>(E. Schiff [243].t)^°\E^t)dE (10. For a possible application of this result.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.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. specifically in the case of the Coulomb potential.t)+ / n a{E.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. 10. 1 rfx^°)*(x)^)(x) = 8mn. . The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>).
(10.123) We now insert these expressions into Eq.127) we obtain from Eq.x)...x.196 CHAPTER 10. ^0)(E.E') = 5{E .. The second term on the left of this equation and the first term on the right cancel out.122b). = e^EE'^h5(E Thus with Eq.E'). . since f rJ>(°>(E.t)rJ><®(E'.„„. We multiply this equation from the left by </?£(x) and integrate over all space.122a) and (10. . V>(x.iJ5 »*/Vn(x). f 6kn in case of discrete tp's. Next we use the orthogonality relations .x.120). (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.e.128) . (10. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10. . (10.t) = eiEtlhy(E.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. (10.125) where an(t) — dan(t)/dt.x.i) = e.„. into dip (10. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. t) = San(t)<pn(x)eiE^h.... . (10. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). (10. Observe that these conditions do not contradict Eqs. / *. i. dnptMrnW = { ^ _ En) (10.126) + H')<pn^)eiE^h}.127) i n c a g e Qf c o n t i n u o u g £.
130) H'kn by \H'kn and at the end we allow A to tend to 1. (10.e.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h.120).130) This result should be compared with the Schrodinger equation (10. they are fixed by the initial conditions.130) we develop a perturbation theory. to perturbation theory in the lowest order.130) along with the parameter A in front of H'kn. 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. In the first place we replace in Eq. determine the state of the system at time t — 0. We restrict ourselves here to the two lowest order equations. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. (10. since it is equivalent to the latter. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. (10. Comparing coefficients of the same powers of A on both sides. In order to solve Eq. SH'kn(ty^E^l^\t) (10. i. SH'kn(t)e^~E^ha^(t).e.10. and we set ak°}=5km or ak0) = 8(EkEm). (10. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . i. let's say am / 0. (10.132) These equations can be integrated successively. (10. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.131) and insert this into Eq. 0.133) The numbers ak = const.134) . we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . Thus first we have ak0) = const.
ip) in the context of timeindependent perturbation theory.135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0. (10. (10.e.137) .136) we see that both "amplitudes" are Fourier transforms and look similar.e. (10. ) ' X V(x')dx'.198 CHAPTER 10. i.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'. 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. In this case we obtain from Eq. H'^y^"^^/hd£. (10. an expression of the following type AS?) = " 2 ^ 2 / e j ( k .136) Comparing Eqs. The amplitude / B o r n is a Fourier transform in spatial coordinates. Equation (10.111). We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. i.135) / Jo L Hfrfc ~ Em)/h Hence the probability to find the system at time t in the state n ^ mis (with Al) 22 i rjv I1 r i r IW a(l)m2 K ( j l = i^kl\t) = H'km e^E^dt> = H'km 6 \nnm\ ei(EnEm)t/h i(EnEm)t/h _ j [EnEmy 4g.135) and (10. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. What is the physical significance of Eq.Emy in agreement with Eq. so that ihakl\t = oo) = 0. whereas ak is a Fourier transform in time (a distinction which is easy to remember!).k . (10.e. (10. ifm{kl) = f J —oo (10.132).ml2sin2{^frn (En . ipm) into the state k.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.
z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e.2 Separation of Variables. r=  r . Z<i = Z) y(r) = _ M ^ . 199 (1L2) . The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. ( 1L1 ) We let pi.2.Chapter 11 The Coulomb Interaction 11. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.Ti. meaning the inclusion of the creation and annihilation of field quanta). The Hamilton operator is then given by w= jt + ji_^M. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. such as the electromagnetic field (the quantization in these contexts is often called "second quantization". Let r : = (x. 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.m. the position coordinate and the mass of particle i with i — 1. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.i be respectively the momentum.y. 11.1 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.
_d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. (11.5) The momenta P . (11.4) P = p i + p 2 . But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2.2/ mi+m2 The mass mo is usually described as the reduced mass. /'Pi p = m0 P2 A v. (11. Z) and r = (x. P = pi + p 2 . r} representation they have the differential operator forms Setting R = (X.4). z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. y. mi + m 2 iTi'2 r2 = R ^ —r. 2Pip2 m\m2 pip2 mi + m 2 . where m 0 = mim 2 .200 CHAPTER 11. so that as expected according to Eq. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • .U7&.3) for the individual position coordinates we obtain ri=R+ •^ —r.3) and centre of mass momentum P and the relative velocity p/mo by setting r» . Y. (H4) \mi m. mi + m 2 va\ (11. p are such that as operators in the position space representation or {R. . so that as expected we obtain the expression for p in Eq. 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 ' . r = rir2 mi + m 2 (11. Solving Eqs.
P on the one hand and those in r. .16) Although we write ri.+ ^(r) 2(mi + m 2 ) 2m 0 (H.j = x.Pj).10) One can convince oneself that the transformation from (ri. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii.pj]=iMij. Angular so that Momentum 201 ^ ^ .9) ft = ftcm + ^ = centre of mass energy + energy of relative motion.Pj} = ih6ij.z). Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. p on the other are completely separated. [Ri. Pj (which in the following we write again ri. r = —.+ ^2. (i. (11.pj.Ri.r2) to (R.8) and so W =777 v + 7 T . Pj over to operators ?i.= ^.13) * and = M ' 1V1 P = 0. etc. (11. (11.y.pj.Ri.+ P. we mean n = rx s x. for which we postulate the commutator relations* [ri.2 Separation of Variables.e. (11. Ri.2 It follows that p2 2 P2 n2 n2 n2 (H. with M := mi + vn^R = ^ P. p = W. (1114) m0 We observe that the equations of motion in R. The equations of the relative motion have the form of the equations of a single particle of mass mo moving in a potential V(r).11. The canonical quantization must always be performed in Cartesian coordinates. pj. (11. 2(mi + m<i) 2mo 2m\ 2m. r) is a canonical transformation.il) ""Op' i. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.10) we set H = Hcm + H. Hence Hamilton's equations apply and we obtain.12) (11.15) In accordance with Eq.
2mo (11. The Coulomb Interaction nc H = 2m. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). 0 < r < oo.r) = *(R)V(r).22) If.20) r2MAR_ $(R) = ER*(R). Thus the twoparticle problem is practically reduced to an effective oneparticle problem.21a) L*+v{r)] ip(r) = Ei/j(r) (11. (11.202 CHAPTER 11. i. as in the case of the Coulomb potential. ip: x z = r sin 9 cos ip.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. 0 < 9 < ir.r).Q 2M' + V(r). —n < tp < ir.23) y — r sin 9 sin tp.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . (11. it is reasonable to go to spherical coordinates r. where Hcm$(R) = and Hip(r) = with ^total — ER (11.21b) + E. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .e. = rcos9.24) . 9. r ) = J E t o t a i*(R.17) (11. (11. V(r) = V(r). (11.
e. . = —1.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11. i.i ' .11. i—>0 (1L25) (11. _ h2 \ .2.j.3. is the operator of the relative angular momentum. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r.)In Example 11.1 it is shown that pr is hermitian provided lim n/>(r) = 0.32) (11.12 — I2. (l + . i( i.2.k a clockwise permutation of 1. (11.29) . if i. j . that the above expression follows from l:=rxp=ift(rxV). = 0.27) We can now write (see verification below) p V = h2Ar^ where 2. (11. (11. (11. k an anticlockwise permutation of 1.28) .26) However.pr]=ih. (11._0di})\ x We now demonstrate that 1. with tijk — + 1 . k = 1. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«.2.3.3. otherwise.33) i.31) .2 Separation of Variables.5r = .j. = (p2 + ^ V (r + 0). where (r x p)j = ^eijkrjpk. 2 1 9 / 1 9 . We have I2 = (r x p) • (r x p). n ( .A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq.
40) (11. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'. r— = r • — or or (11. (11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.38) + y* + z .ih) — rpr(rpr With Eq.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. Eq.ih)pr and x 2 2 9 9 2 2 Q O (11.k ~ rjpkrkPj) (11. This means that l2 = so that 9 9 1 (11.43) r2(p2p2). (11.k Using Eq. (11.41) o = r pr.37) .27) we obtain r{prr)pr = r(rpr .ihr • p r2p2(rp)2 < + ih(rp).36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.ih5jk)pk (11.ihr • p ) .44) .42) (11. ih6jkrjpk (11.j. (11. (11. (11. * ' S Since r =xz we have (cf.(r • p ) 2 + 3ih(r • p) .25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .39) + ih).35) j.
(11.2TT].2 Separation of Variables. i. <p)r2dfls = .Q ' 2mor2 with the condition (11.6d0dip.7r]. 0. <p)*rip(r} 0.<p)drdna " r r=0 J \ll§r~(r4'(r. H 1 2mo pi + K) +V(r). h f l d —(r<l>*(r. The Schrodinger equation of the relative motion can therefore be written V1 V 2m. (pr4>. 6. ip 6 X>Pr C 7i.[ 4>(r.<p G [0.e lim rip(r) = 0.30) I2 is to be identified with the square of the angular momentum operator.45) oe Thus in Eq. (11. prip) = = I <j>{r. 11. 9. i.1 Separation of variables We deduce from the expressions of H and l 2 that [H. <p) d °° .<P) = Etl>(r.46) (11.(r.11.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).[ i J With partial integration with respect to r this becomes (<l>.e. 8.. <p)*Pri>(r.26). r—>0 il>(r.d_ r dr the operator pr is hermitian.6.47) Example 11.e. S o l u t i o n : Set d£2s = sin.! /').<p)*?(ri.<p) (11. <p))r2drdns % J r or r<t>(r. <p)* .<p))drdns.ip) = (<£.48) .d. pr = .6l. /'(r.2. r := r : lim rip{r) = o i . or J .e'iP^} .e. reR3.28) and (11.<p))ril>(r.Pril>) J / r= dQs rcj>(r. Then {<j>.e. 9 € [0. 6..l2} = 0. ¥ p)r 2 drdn s = (p r 0. (11.— (riP(r.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. Angular Momentum 205 Comparison of Eqs.e. Pri/Oi V0.
l2] = 0. (11. xpz] = y\pz. zpx] + [zpy. so that [lz. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0.xpz] = [ypz. [ly. Since 1 = r x p .e.e. g i = x. /z> *xJ = l <i\}ylx + tx'j/Jj [h.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions.l2} and correspondingly [lx.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l . z]px + py[z. (11.lx or l . [lz.e.50) Thus the components of 1 do not commute pairwise. H and l 2 possess a common system of basis eigenvectors.2: Verify that [h.lx] = ihly.206 CHAPTER 11. We therefore first search for a complete system of eigenfunctions of l 2 . (11. we have [h. for any two of the components there is no common complete system of eigenfunctions. Example 11.53) Either of these operators is the hermitian conjugate of the other. 3 .49) By cyclic permutation of x.lz (or l . z we have altogether [lx. (11.ly]=ihlz.ypx) = ihlz. y. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. However.52) This means that l .51) (11. they are incompatible variables. The Coulomb Interaction i. [lyA = 0. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.pz]x = ih(xpy .q2 = V. zpx . ?3 = z). These operators here play a role analogous to that of the quasiparticle operators . In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. 2 .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. i. ly] = [yPz ~ zpy. 11. 2 2 2 = 0.lz]=ihlx. i.
e. (11. y . (11. (11.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq.h~lz(1156) (11.l)hl\l). [i+M = 2%iz. (11.59) = 9) h(l0 r)\lr).54) On the other hand we obtain from Eqs.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real.+ ll.Z_] = [ l 2 .i+] = m+.r) := (Z_)rZ) so that lz\l . (11. In the present context the operators are called shift operators for reasons which will be seen below.54): lzl\l) = {llz . l\l) is eigenvector of lz with eigenvalue (IQ — l)h.11.r) (11 = {l02)h{l)2\l) (11. (11.57) i. Hence we must have Z+I0=0. (11. With the help of Eq.60) Similarly for l+ with the help of Eq. i + ] = [l2.58) {l0 : max.i] = m. (11. (11. [ig. (n. (11. Similarly we have lg{l)2\l) and more generally lz(l. (11.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i).61) But lo is by definition the largest eigenvalue of lz. A^ in the case of the linear harmonic oscillator.52): 0 = [ l 2 .3 Representation of Rotation Group 207 A. We write or define: / .54) l 2 = \{l+l~ + 11+) + ll = l+l.).49) we obtain the following relations: [iz. assuming a finite dimensional representation space. Therefore (lo + l)h cannot be an eigenvalue of lz.51). and with Eq. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. since lz is hermitian.lh)\l) = {l0 . .Y\l) = (lQr)h(lY\l).
208 Thus for l 2 we obtain: V\l) ' = 11. l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 . (11. (11. IQ(IQ 1 0 (11. (11.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .n). i.e. (11.l2\l) = l0(l0 + l)h2l\l).64) i. (11.From Eq.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz..n)}h2\ln).e.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.68) . (cf.\ll).e. [12.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.+ 11+) + l2z\l) = i+i. l\l . The Coulomb Interaction (l+l.n) (1 = 60) {l)n+1\l) = 0.h)\ln) {(lQnf(lQ. Eq.54) (n = 57) (11. (11. i. (11.67) i (11 60) = But according to Eqs.63) + l ) ^ 2 . (11...Z_] = 0.56) CHAPTER 11.y\i) l)h2\lr).55)) we have 12Z_Z> = l..\lr) terminates at some number (let us say) r = n (n a positive integer).60) and (11.66) It then follows from the second of relations (11.e. i.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. \l).
11.  U o . We write these* for integral values of IQ: \l.g..m\lz\l.l)h. As in the case of the harmonic oscillator we introduce normalized basis vectors.{lo . 17.l_. R. (11. . y/(l0±m + l)(l0^m)h.n){l0 . m + l)h.m) : Mo).71) can be established in analogy to the method used in the case of the harmonic oscillator. where lz\l.m). lz\l.69) Prom this it follows that IQ must be either half integral or integral.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .. The basis vectors \l).m).. The nonvanishing matrix elements of 1 = (l+.m) = IQ(IO + l)h2\l.. The normalization factor in Eq.Z 0 ).1).m) with — lo<m<lo2 (11. The phases are chosen such that§ l+\l.. y (lo .70) they satisfy the eigenvalue equations (with m = l0Jo l.m + 1)(Z0 + m)\l. The operators l+. Messiah [195].e.65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. (IQ — l)h... (11. Brink and G. (11. p.l)h \/(lo + m + l)(l0 .m) (l. According to Eqs.n)} = (l0 .m±l\l±\l. p.m .m)\l. (11..m) = = mh...63) and (11. (11. 24. A. \l — n) are eigenvectors of lz with eigenvalues loh. lo = l0n. II.l0): l2Z.n) 2 .lz transform the vectors \l).l.m) — mh\l.m y/lo(lo + 1) — (m — l)m\l.. we skip this here... lQ = . \l — n) into one other. M. Satchler [38].m) = mh\l.1).m) = = = = y/lo(lo + 1) — m(m +l)\l. lz) are then: (l.3 Representation of Rotation Group so that with Eq... Vol. i.m) l\l. —loh. We therefore write the 2/o + 1 orthonormal vectors: \l. The dimension of the representation space is therefore n + 1 = 21Q + 1.70) According to Eq.2..m). § See e. *D.n .m / + l)h .. In the following we consider primarily the case of integral values of IQ. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.
The Coulomb Interaction Z+Mo> = 0. 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. where IQ = 0.<p) = i0{io + lzY£(0.1. ip) = e(9)$((p). 11. as this is sometimes described. 9. . d2 sin2 9 dp2 _' (11. l_\l. CHAPTER 11. (11. mhY£(9.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip).75) the variables contained in the expression (11.yp.76) d2 (11.l0)=0.2. with i2Y£{e. $ oc eim*..4 Angular Momentum:Angular Representation In the case of the harmonic oscillator we arrived at the position or configuration space dependent wave functions by considering the position space representation of states..77) $((p) = m2<b(y). /o.210 Then. and m = —IQ.. (11.With the ansatz Y(9.<p) = i)n2Y^{e^). . —lo + 1 . for the simple "rotor".73) 1 d d sm0 06 \Sm9dd) + (11. i. as remarked earlier.e. We now perform the analogous procedure for the angular momentum or. In these coordinates we obtain: h l± so that h2 = xpv .74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9). as above. . ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . ..74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. ip. We . (11.<p).
* ? ^ . orthonormalized system of eigenfunctions.<p) Jo = 6mm'5w. (11.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.n>).79) . / 3 / 5 J — casO. is given by (2Z + l ) ( Z .77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.ypx = tfrK(11.\ = W . (11.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}. These functions are defined in such a way that they are normalized to unity on the unit sphere.V>)Yir'(8.(p) are called spherical harmonics.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.i». m > 0. Uniqueness implies immediately that m has to be integral.r o ) ! YT{e.11.t»'\Y7n((l'.0) is real and positive. so that im(ip±2ir) The functions Ylm(8.81) 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.76) and (11. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.^ = *(n .82) .=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. (11. We mention in passing that in spherical coordinates d lz = xpy . y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.80) y? Furthermore Y°l = V^fcosO). (11.(5cos 03cos0). (11. This is the case if one demands that l 2 and lz possess a common complete. and that their phases are such that Yt0(0.
85) (11.m(m + l ) ] i f = 0. f. exp(imip).g.. (11.88) y = ^aiXi+K.1(1 + l)]xK+i = 0.e. i.e. oo (11. in the replacement ip —> if + 2K.. of Ytm..x2)P[' . respectively..2(m + l ) x i f + [1(1 + 1) .2xy' + 1(1 + \)y = 0.p.88) we obtain for every value of x ^2 ai(K + 0( K + * .90) . 22 (11.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 . (11.e. Consider the equation for Pi(x). For the solution we use the ansatz of a power series. (11.m = 0.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .d. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. is also in use. we would now have to search for those values of I for which the solutions are square integrable. It is instructive to pursue the appropriate arguments.89) Inserting this expansion into Eq.l)xK+i~2 i . The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . i.1. i. (11.h.2xP[ + 1(1 + l)Pi = 0. (11.x2)Pf . the designation s. Instead of the values I = 0.x2)y" . i=0 (re lowest power > 0).2. (1 .x2)y" . i.212 CHAPTER 11. If we did not know yet that I = IQ = an integer > 0. Setting y = (lx2)m>2Pr(x) m .e.87) we obtain the equation (1 .86) y = 0...
l(l + l)h2 _2mo + 2m. ai(re + !)« = 0.93) where Xl(r) is the solution of the radial equation (cf.5 R a d i a l E q u a t i o n for Hydrogenlike A t o m s The common set of eigenfunctions of the operators H. For the solution in accordance with (11. l 2 and lz are the solutions of the Schrodinger equation of the form (11. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . but also those for a different series for a. M. We have therefore (multiplying Eq. Eq.78) to be square integrable. p. This is precisely the case.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.1(1 + 1)].1.94) ^For further details see H. (11.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0. (11.26)) yi(r = 0) = 0.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. Eq. they have to be polynomials (recall the orthogonalization procedure of E.94) from the left by r) h2 d2 h2 2 + H! + \)IL.Qr2 with (cf. Margenau and G. > 1. first line of Eq.92) This is a recurrence relation. Schmidt). (11.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. (11. these are of little interest to us here. .46)) p.*" 11. . when the series terminates after a finite number of terms.96) + V(r) . We see that this happens precisely when I is a positive or negative integer or zero. For \x\ < 1 we can obtain from this the coefficients of a convergent series. Murphy [190].91) from which we deduce that K = 0. However. (11.E Xl(r) =0 (11. 93.11. (11.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.
102) . (11. (11. He + .1. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). 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. oo) of a onedimensional space.1)). we obtain the socalled "indicial equation!'' s(s .e. L i + + . This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. The Coulomb Interaction (11. (11. i.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr.100) with (11. Obviously the irregular solution has to be rejected. In the first place we investigate the behaviour of yi in the neighbourhood of the origin.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. Z\ = Z2 = 1 in Eq. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. and substitute this into Eq.99) e ip(r) = Eip(r). We now consider the hydrogen atom with V(r) — —e2/r (i.96). s = 1 + 1.1)+1(1+ 1) = 0.e.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. which behaves like 1/r1 in approaching r = 0. and others. for instance. The equation also has an irregular solution. r m0 = meMp me + Mv (11. an analogous consideration applies to hydrogenlike atoms like. Equating the coefficient of 1/r2 to zero. nor the condition of normalizability for l ^ 0.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. since it satisfies neither the condition yi(0) = 0.214 with CHAPTER 11. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.
6 11. for E < 0 they > decrease exponentially. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. see e.v) vi(x) = 0. We can see from Eq.88.106) 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.107) equation. (11. Inserting the ansatz into Eq. hypergeometric (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.105) we obtain the equation for V[(x). T This equation is known as Whittaker's equation.6.g.105) 11. We set therefore Vi X J. W. *The factor 47reo appears in SI units. i.e.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11. It remains to determine the function vi(x) for a complete determination of the solution. (11. (11.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. Magnus and F. (11.e. (11. p. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . (11.108) This equation has the form of a confluent x$" + (6 .11.e. (11. i.105) t h a t for x —• oo there is a solution behaving like exp(—x/2).a * = 0.+if>x/2 vi(x). Oberhettinger [181].103) °y moe2 ( = 0.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. .104) W i t h these substitutions we can rewrite Eq.x ) $ ' .
(2) r ( z ) r ( l . However. meromorphic function.z) = ^f^. The function T(z) is defined by the integral /co T(z) := / eHzldt. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. (11. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. 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. (4) r ( i ) = V5F. particularly pure mathematicians.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. called dupZication formula.! ) ! = imlFI' c a l led inversion or reflection formula.J. and hence the *Some writers. i. (5)r(n + l ) = n ! . ^ ( 2 ^ ) ! = 2 2z z!(z .1/2)!.216 CHAPTER 11. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . more precisely.112) The gamma function is a nonvanishing analytic.e. 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. (11.(0_i)!(6 + n l)!n!' (11. and for both cases. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . the factorial for an arbitrary argument. (11. with many of these around in some calculation. (z)!(z .z(zl)\ = z\.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). it is convenient to use only the factorial notation. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. .106).
6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. 1 . which are also described as its nodes. One defines as the principal quantum number the number n. (11. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). (11.e.11. in our case: a = 0. 1=0 (11. 2 . d = 1. The degeneracy of the levels En is therefore of degree* n—1 1=0 n—1 V(2Z + l) = 2 V Z + n = 2^(n .v) T(l + lv) _ ~ K sin{7rQ/ . n' = 0 . i. = n = Z + l + n'. the orbital or azimuthal quantum number I can assume the values 0. every n.. .a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . With (cf.2. .1) + n = n(n .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq.e. The quantum number n' is called radial quantum number.1) + n = n2. For every value En. This is achieved. for z.1.101) to (11.113) This is a quantization condition. and so to the number of different values of m. . if the series terminates for certain values of u.. .104)) 1 n =v = K.V))T(v .» • » • » .I) r(vl p)sm{n(i/ I .e. i. Eqs.. (11. i. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression..114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. which is thus seen to be a consequence of demanding the square integrability of the wave function. Thus the infinite series must break off somewhere. must be a polynomial.110) as follows: T(l + 1 + p .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. .
218 CHAPTER 11.. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy.. I = 0. 11.. hence p). (c) The nucleus was treated as pointlike and not as something with structure.'.\ I = 0 or 1.1 = 0 (called sharp series. from n. (b) The spin of the electron was not taken into account... The following spectroscopic description of states is customary: nl —> ns.1 Hydrogen atom energy levels.. 0. np. The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig. / = 1. nf . respectively.. 2 . from an initial state n > 2. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. The spectrum of the hydrogen atom thus has the form shown schematically in Fig. stand for I = where n denotes the principal quantum number and s. hence d). . I = 0 to n = 2. and from n.§ One also writes n2l+1l. ' O n e describes as Lyman series the final state with n = 1. ne.1 = 0 (called principal series.. 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. from an initial state n > 2.p. 11.1.1. hence s).4. . nd. . and hence the fine structure of the energy levels is excluded.1 = 1 (called diffuse series.1 = 1 to n = 2.d.l = 2 to n = 2. .'. .
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. W.115) we can reexpress this function as Lr{z)  r\k\ p=0 \* p\ (rp)!(fe+p)! (U 118b) ^This definition of *LJ? is used by A.115) is effectively the associated Laguerre polynomial normally written Lk(z). see. Magnus and F. . p. the polynomial contained in y\ of Eq. We mention here three different definitions used in the literature. • fc. k + l.2. Messiah [195]. Unfortunately. Apart from factors. so that one always has to check the definition.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. however. p.2l { + 2. for instance. Unfortunately the definition mostly used in mathematical literature differs. For clarity and later use we denote the function used there by its normal form L^.414(3). formula 7.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). Ryzhik [122]. several different definitions are in use.1. • ( . it is given by^ *Lkr(z) = ^ ^ f 1 *ir.117) The function *Lk(z) is a polynomial of degree r. (11. I. 844. Oberhettinger [181].r = 0 . (11. . .r f . (u.z) = (r + k)\ Lkr(z). 84. S.x). • (11.114b) yields the binding energy of the states.= t . The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. in fact. With the help of Eq. and I. Each can be recognized by reference to the generating function. Appendix B. 2 .6. "1'f„+11" P] „.115) ' =0 (ri .2 (11. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). l . a + 2:*) = t v . 3 .p)\(2l + 1 + p)\ The spectrum (11. Vol.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. Gradshteyn and I.11. M. We distinguish between different definitions in use. . .118a) where Lk. 11. (11.
The generating function of the associated Laguerre polynomials — obtained by searching for an integral representation of the solution of this differential equation which is of the type of a Laplace transform — is for i < 1 and expanded in ascending powers of t (cf. Vol. Ll(z) = 33z + ±z2. pp. Sneddon [255].„M„ z"*L. It is said that this is the definition preferred in applied mathematics..** The associated Laguerre polynomial here. L\(z) = 2z. ^ F o r this definition and the following equations we refer to I. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. N. Messiah [195].220 CHAPTER 11.. 162 . and Ll(z) = l.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1.1. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . which we write ^Lk.z .d z—^z + {k + l~z)—+r Lkr{z) = 0. p. There at some points n + 1 is misprinted n + 1.„\*Kk. Example 11.2. L1(z) = l . **I.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. N. .t) + k l r)! E E (zY (i +(fc +h)\(ri)\ i\ r=0 L OO p OO i=0 E (k + r)\ *L (z) = J2L ( y. Sneddon [255]. r r Z r=0 v tr k k (11.164. . 163.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.121) In Example 11.3): e2t/(lt) (1 . (11. . Appendix B. is defined by^ dk (—'\\k(r^\2 (11. I.122) "See A. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. L2(z) = l2z + z2.
120).3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq. ^L\(z) = . . Example 11. (11.107) implies k = 11 + 1 and r = n + I..8): (11. *L\(z) = .124c) Example 11.119): ty"(t) + (a + 1 .124b) Comparing Eqs.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.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n . '"''Comparison with Eq.4z + . Then.1 .122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). i L1(z) = l . (11. N d rk 0. and obtain with this the normalization condition (11.z . and *L\(z) = . The limits will be determined later. (11.119) derive the generating function of associated Laguerre polynomials (11..120)) it)k& zt/{lt) ~*P p=k fc (11..1 8 + 18z 3z2. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.121).118a) and (11. and partial integration. (11.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp..4 + 2z. ^L2(z) = 2 .123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. (11. Jfpd>2s{p))eptdp. (11. Solution: We consider the following equation of the form of Eq.124a) (i*) pi The orthonormality condition is (cf.
124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. which removes [. p and one has (as polynomial of degree 6. It follows that 1 d i I _i. the only possibility is the last.222 CHAPTER 11. Then dp = dq/(lq)2.p ) « + i q c + i ( i . case (3) similarly. Thus in the case of Eq..p + 1 = 1/(1 . n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1).) 2 J J p 6 + i ( l . (3) around p = — 1 if a + b is a negative integer.( a + l ) p s ( p ) + 6s(p) dp 0. t > 0." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. 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. (11. Case (2) implies the exponentially increasing solution ~ e + t .q) and P~ t ( . 2TT_ Jq=0qb+l(lq)a+l 1 . we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . (11. with requirement of a finite solution. s bK> (lq)°+i (11. (2) from —1 to oo./ ( l — <?) —— = G(t).] from the above equation: C(p) := p(p + l)s(p)ept = 0. and (4) around p = 0 if b > 0 and integral.^ . . 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. j p = (P 0 "V?.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(.p ) o + i 7 t = 0 .a = 21 + 2. agreeing with Eq. Hence the expression in square brackets vanishes. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo..119) with a and b as there. We demand (to be considered later for integration contours) the following condition. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . since for n > b : (b — n)\ —• oo.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. this is the rejected irregular solution.
where Fnl(x) = yi(x) (11. The wave function of the state (n.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. Clearly only the second form is sensible. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . and 0 if cj=b. .(x).11. Jo (11. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.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. 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. 11.127) . so that I = (c + a)! 1 f .?).—i.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.6 The Discrete Spectrum of the Coulomb Potential 223 We evaluate the integral on the right with the help of the integral representation of the gamma function.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.111).112). /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .125a) and the factor a~ 3 ' 2 with a = Bohr radius. is introduced for dimensional reasons. I. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . (11. Eq.6.
130) We see that the mean value of r is the larger the larger n is.6(n + Z)[(2Z + ! ) . 11. 2 (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 + /„.v (6n2  6nk + A:2 + 6n .131) .3fc + 2). = 0) we have ) r /u =.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 . (11.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r). thus on the average the electron is the farther away from the proton or nucleus the larger n is. + . a (r)is = ~a. The Coulomb Interaction One finds in particular (see also Example 11.(2Z + 1) + 1} = ^[Zn21(1 + 1)].224 CHAPTER 11. This behaviour is indicated in Fig.1 + 1) (11.! ] + (21 + l ) 2 .8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .7..2.2Z .3(2Z + 1) + 2} 2 {2(n + 0 .n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ . (11.(l) n+l. For the ground state (n = 1. and for the explicit derivation of the second case below Example 11.
11. \f{r ) ~ n a. with n +1 replaced by n and 21 + 1 by k. i.134) *These are taken from I. v ^ V2n + 1 (11. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton .133) for large r.2 Hydrogen atom wave functions. pp. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.2 1 A r = V (^2) .( n . / ! J — . or (r ) ~ n a 2 2 ^ [ 3 n 2 .20.132) (11..j . In the case in which I assumes its maximal value n — 1. 173/174.6 The Discrete Spectrum of the Coulomb Potential r.J n 1s 2s 3s *r Fig. Problem 5. 11. = l a nl n N'nll rFnl.e. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 . S.(r) = x W 2 n + 1 =.l ) n ] =  [ 2 n 2 + n] an[ n + (11.2 I 3/2 Kl r 225 \2 1 R„.^ We obtain therefore for the quadratic deviation (r) . we have and tp becomes particularly simple. Sneddon [255].
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). we have pg = const. Thus. Example 11. Since 6 is a cyclic coordinate.e.e. in which periods are obtained from the BohrSommerfeldWilson quantization condition. the electron remains practically localized within a spherical surface of radius (r). Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. In order to obtain an impression of relativistic corrections in a simple context. = ngh.— > where ampler is the length of the semimajor axis of the elliptic orbit.4 the relativistic equation in cylindrical coordinates. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory".135) This expression becomes very small for large values of n. The Coulomb Interaction or TT = ^ = (11. the states Ytl are predominantly concentrated in the xyplane). identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.226 CHAPTER 11.3. \/v vT 7 v^ See also Example 14. and separating this in cylindrical coordinates r. i. but rather like classically in a plane (i. r. we consider in Example 11. p2 = p2 + — pg • eV. . „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. 6. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). pJ = —— and j> Pidqi = nth. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. i = 0. This means.r. (c) The nucleus was treated as pointlike and not as something with structure. However. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. for these values of I it is not distributed with uniform probability over the spherical surface.6.
(and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. 11..9. (p). Schiff [243]. it is possible to separate the Schrodinger equation in other coordinates. as also indicated in Fig. which correspond to a new choice of polar axis. p. z=^rj).11.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. which are related to Cartesian coordinates in the following way. <p = <p. . y=y/%qsanp. y — 0: V ^ c o s tp. (11. The deeper reason for this are symmetry properties of the problem. r\ = r ( l — cos#) = r — z. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r.(p . where r = ^/x1 + y2 + z2. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 .136b) £ = r ( l + cos#) = r + z.2.6.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £.* See L. ip are the spherical coordinates. In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9. where the constants are identified by comparison. where ip = 0. 2 2 Therefore the energy is no longer a function of n = nr + ng.136a) (11. ( v c + i ) . 86. Whenever there is degeneracy.3.V .T). 11.
. 1 . The Coulomb Interaction Fig.228 CHAPTER 11. In terms of the coordinates (11. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. there is really only one basic equation to solve here. .136b) the Schrodinger equation of relative motion (11. 2 . for uniqueness of the wave function. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . We rewrite the first of these equations with the help of the relation M (w ^  m + e) . (11. so t h a t the variables can be separated. with zaxis as axis of rotation.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. a separation constant. 11. . one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. and r\ = const.3 Parabolas £ = const.22) becomes — as we show in Example 11. = 0.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip.
We leave the calculation of the degree of degeneracy to Example 11.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). (11. . T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) .11.o. m0Z1Z2e2 ^ 2 = n (say). ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. . (11. 2 . . i. .114b).± ( m + 1) = 0 . n 2 = n" .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0.e. 2 . T h e n by comparison with the spherical polar separation the answer is as in Eqs. . .105)) Hence we write I = (m — l ) / 2 .5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. Adding now / .125b) (dividing by p 1 / 2 ) £ oc e . 1 .6. p = /?£ and cr = /^r?. we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". . . 1 .106) and (11. . ds2 = dx2 + dy2 + dz2. as well as the Laplacian A = V and the Schrodinger equation. In the spherical polar case we had (cf.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . To obtain a closer analogy with the separation in the spherical polar case. Eq. . E x a m p l e 11. " . . from Cartesian to parabolic coordinates.
136a). for m / .114c).l)a + (n .g<p 9 \ . Similarly we proceed with dx and dy. as t h e two signs of t h e exponential indicate. last chapter. Thus dz2 = . 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . for instance.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. we have to find the number of values of the quantum numbers n\. + l 4 {^^)drl f i + v \ .tx2 . The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. (11.2)d/2. i. > . dx = d{ y £77 cos ip) = . Eq.e.e. see Eq. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.230 CHAPTER 11. We consider first the case m = 0. all cross terms cancel in the sum of 7 the squares.g.. Thus we want to obtain the coefficient of wn~1 in £2 . 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11.2) and summing over m.2 = n — 1. i. / — cos ipdl. N2 where g^dS. Thus.l ) ( n . We then sum all possibilities and select the coefficient of ojn~1 in the result. and attach a factor to to each term introduced in the wave function.2d? dr)}.2)d} = (n.. . Magnus and F. etc. \— < / — cos ipdr] — \/£r] sin ipdtp. and so dz = (d£ — drj)/2. . Solution: For the wave functions ip = £(£)J\f(r])&(ip). Oberhettinger [181].2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 .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. 2 2V? V^ One now performs the square of this and obtains similarly dy2. Here we have (see preceding text) 711+77. / J„^2 . (11. The Coulomb Interaction Solution: We begin with z of Eq. " . z = (£ — rf)/2. We use the method of selector variables. i.136a). W.2 . </. 0 there are two. (n _ !) n = n2 §See e.E E n l „ . {g^Y. Observing that the highest possible value of m is n — 1 (i. n the double sum ]T]2 is TI — m. In the case m ^ O w e have n\ + 77. 2 + ^dV . Beware! For m = 0 there is only one mstate. . 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. 712 and m which give n.e. are elements of length. when m = 0 = 77. (11. n 2 (1UJ)2 The coefficient of a .1) ~ . 2 _ = (9id02 + (gvdvy + /„ . = e±im^e("+^'2{par'2L^{p)L^ (a).e.[df + dr]2 .1 in this expansion is seen to be n.
with Z\ —> Ze. these expressions agree with those of Eq.124c). with Eq. . 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. i. and determine the average value of v for a given state. .11. we can relate the expressions Iq to those of Eq. = j (2ni+m+l).™(p)] 2 By contour integration (see Examples 11. eF{z).e.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. I.drid<fi oc (p + o)dpda. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£..e. . ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.g. With the help of the wave functions we derived above.127) and setting n = r + k.e. fc — m. . In fact.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.•* n (l) = / Jo It follows that .(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. we would have to use degenerate perturbation theory and obtain the same result. which will then give the change in energy of that state to the first order (i.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). (11.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz.128). (11.8 and 11. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. However. i.128) if one makes there the substitutions n —> n\ (or 712) + m.6 The Discrete Spectrum of the Coulomb Potential Example 11. with the connection » (11.1 (cf. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. with z = r cos 9. 8. Z2 —> e. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . Thus e. if we expressed the perturbation in spherical polar coordinates.—r~n = 3(mn 2 ).6). (11. with RayleighSchrodinger perturbation theory).r. {p — a) — = dxe~*xkvLkn (z)] 2 .) = ^eF(p . Ikr+k. s 0/ . where F is the electric field. L ni\ ny. Sec. .a).
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.b. and is a special case of Example 11. s = at. In Example 11.124d) of the alternative integral representation obtained in Example 11.111) /*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 .g )(l..p ) ( l .3).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. c positive integers. The Coulomb Interaction E x a m p l e 11. (11.p ) ( l . . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . _ „ .0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l . tt„:_ (1 .7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.128).x ( n + m)! ( n . dt = ds/a. the integral / is the coefficient of pnqn in the expression _ / . (n + r . 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 . (11. the integral arising here becomes with Eq.1)! n\ This verifies the expression I™t in Example 11.9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.1 ) ! (n .a.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .( 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)!.3: /"OO I := Jo et[L^{t)]2tm+1dt.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.^~T^——^ ) = coefficient of (pn .9 we consider integrals with an arbitrary power of t.pK — — —L .9 )« Solution: Using the generating function for both associated Laguerre polynomials.232 CHAPTER 11.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n . ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .p ) r + 2 [(lp)(lg)]"l+1(lpg).l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t . E x a m p l e 11.
p ) ] .p ) 0 + l 9 c + l ( l _ g ) a + l ^ a ) ' where j(a) contains the integration with respect to t which may be evaluated with the help of the integral representation (11. c + b odd).1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.l ) * x coefficient of a* in the expansion of 7 ( a ) .a _ 1 = [1 + a ( l .e.^ ( l .(i) = i ! ( . we have to have y + (c + b — 2r) = 1. . — (cr)!r!(6r)! Kv. (lp)(lq) + a(l q)(l . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. so that = (q+fc±£) . Using the contour integral representation (11.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. where (a+**£)! c j n Jo = . ° ^ + a ( l .p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .L.124d) of an associated Laguerre polynomial.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. we have 2r = 6 + c = even.111) of the gamma or factorial function.w(r) = and / a ) = r\(P ~ r .e. (a) j / = 1. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l .r ) ! a ' .q)(l .p) .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 «.?(t)t a e. c + b — 2r = l (i.a t [Lg(i).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 .f ]. i. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.e.™( r > (1 + a)P r £^0y^w Jc+6_2r = . c + 6 — 2r = 0 (i.e. . c + b even) and/or j / = 0. i. 7£.11.
also with regard to earlier literature. the standard principles of nonrelativistic scattering theory.~* 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 . however parallel to the Coulomb case.)!' f^)(.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. + (!±<)«. ^A very readable source to consult. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. b\ V cited in Example 11. cannot really — and this means in a rigorous sense — be applied in this case.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. In view of the slow falloff of the potential at infinity. . t We have the Schrodinger equation of relative motion of particles with charges Z\e.^)(. Since such a rigorous treatment is beyond the scope of this text. 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. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r).1%^ 11.7 Continuous Spectrum of Coulomb Potential In this and the following sections we consider the scattering of a charged particle in the presence of a Coulomb potential. is the work of J. Taylor [267]. R. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111. 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.7. .
W. (11. (11.z)).M) = l + . v = ik(r .e. a(a +1) z2 ] (11.136b).e.108) and (11. According to Eqs. where ^. 86 .138) becomes 235 ^A + k2 .z). (11. „ az 1. what is the > behaviour of 3>(a.^ W ) = 0. pp. i. b..z) + W2(a. with the asymptotic expansions (11. . z = rcos<9. (11. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.b.. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a.144) *(a.143) and v = ikrj.109) we can therefore write the solution ipr = Aeikz$(i>y.11. Our question is now: How does ipr behave for r — oo.z) = Wi(a.z).140) The simple Coulomb potential permits particularly simple solutions.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). for r — oo) • ipr = Aeikzf(r .. Oberhettinger [181]. (11. .87 under Kummer functions. Magnus and F. that we encountered previously. (11. (11.146) See.b.141) Inserting this expression into Eq. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr).140) and using (11. i.+ ^  + . .142) This equation is of the form of a hypergeometric equation. A = const..b. (11. for instance.7 The Continuous Spectrum of the Coulomb Potential Then Eq.
The Coulomb Interaction W2(a.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.a) (z) na)"~ r(la) T(ba) n\ ' (11. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.150) ^(rz)]1"^ 1 + 0 T(«7) i.148) . i. (11.152) Since z = r c o s # . if tbr were „ikr ibr ex el"z + f{6) r (11.151) with Eq.b.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.(11.e. with 2. > A = Aeikz[W1(i1. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.z) + la)T(n + b.z)T(l .l.ik(rz))} =^ + ^d.153) The asymptotic solution represents t h e stationary scattering state (E = const. (11.i 7 ) (11. Without t h e logarithmic phase factors.e. a comparison of Eq. J\") ci(kr~/\n2kr) (11. (11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.^l*7 l + O (11.7 ln*(rz)) k(r .kz+l^ktrz)) C .236 n\Zzab CHAPTER ^T(n '£ fo 11.149) (11.l.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) . = r cos 0.154) .
The quantity /(#) 2 would then be a measure for the scattering in the direction 9. 11. (11.7 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. In other words. 11. the effective range of the Coulomb potential is infinite. also called phase anomaly. we see that this expression is the exact analogue of a current density in classical considerations.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. The problem of the logarithmic phase..11. (11. do dQ K 12 K Fig.155) Since the gradient is proportional to the configuration space representation of the momentum. z — — oo. that we encounter here is a characteristic of the Coulomb potential. 2im. For the incoming wave ipi = exp(ikz). the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. We define as current density the quantity h J := 2imo [^(v^)v(v^n. the current density is * h .Q hk mo v0. As a consequence of the masslessness of the photon.7.n . This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.156) .4 Variation of the observed differential cross section with 0.
154) we chose the prefactor of the incoming wave in (11.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.238 CHAPTER 11. the square 7 2 ). We make a few more important observations. 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. in the current this would in any case contribute only something of order 1/r. In reality this divergence does not occur. i. is called scattering amplitude. In the context of quantum electrodynamics one refers to "vacuum polarization" and relates this to the idea that a charged particle like the electron polarizes the otherwise neutral vacuum by attracting virtual charges of opposite polarity and repelling virtual charges of the same polarity.152) as 1.e. since in actual fact the Coulomb potential occurs only in a screened form. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). thereby screening itself off from the surroundings. In Eq. 'screened ZlZ<2e2 cr/r0 ° . This implies that scattering takes place for the attractive as well as the repulsive potential. We recognize the result (11. (11. That this formula retains its validity here in quantum mechanics is something of a coincidence.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq.e. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . (11.153): 4fc2 s i r  The expression /(#). which yields the scattering cross section. Correspondingly the other part of (11. We see that a depends only on the modulus of the potential (i.159) as the Rutherford formula which can also be derived purely classically. for 0 —> 0.
t). be given by ^ f dk AL.e.e. 2 (11162) where Vk(x) = ^ ( x .k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. t) = U(t. t 0 ). (k).161) where Ak0(k) ~ A(k — ko) and ^ k (x. t 0 )^k(x. t 0 ) .*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . (11163) (11.x° ' r j . / dk 7^340(k)4(xxo. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .166) Let the incoming free wave packet at time t > t$. (11. The particle velocity v is v = — .i i f (''°>/Vk(x. v0 = .^ ^ % ( x ) .t) = e . The wave packet is a superposition of waves ip(x. 11. 2mo V + F(x) k (11. 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. t0) = e. i. the wave packet in the absence of a scattering potential. (11. of continuum wave functions.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. and which provides the uncertainties arising in quantum mechanics. 11. and if H is timeindependent one has the stationary wave function ^(x. i.164) Ek~Eko + {kk0)v0h.165) ^ ( x ) = £ fe Vk(x). whose maximum moves with the particle velocity.e. which in the case of free motion (zero potential) are simply plane waves. of the equation r fi.4.8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig.t).yJ ^ j A k o W 727)3/2 e 0. m0 m0 and V. i.2 fr2 .11.
k = kez.169) The wave packet in the presence of the potential is obtained from the free form (11.171) The phase of fk(0.166). The Coulomb Interaction In Example 11.165) ko • (k . Then with k = ^(ko + k .e. With Eqs.x 0 ).154)) that for x — oo.ko) k0 (11.172) h h k0 {rv0(tt0)} (11. i.ik .163) we obtain </>ko(x .ip) is the scattering amplitude with respect to the incoming plane wave. ip) could be carried along but we ignore it.k 0 ) 2 ~ Jk20 + 2k0 • (k . (11. i.xo.169) and (11. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.t o ) / V k ( x . this solution > has the following asymptotic behaviour in which f(6.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11. From Eqs.170) A ko (k) e .iB *(**°)/ R °'tj + 1 pikr e * .t) = J ^ A k o ( k ) e .167) we have ^k0(xx0. (11. ^k(xxo) ^k(x) ~ = e.173) .<p) + Ol Akr l (11.e. (2TT)3/2 eik* + —fv(P.k0) ~ k0 + and the expansion of E^ in Eq. Eq.161) and (11.10 we show that the solution V'k(x) of Eq.^ ( * .167) by replacing the plane wave exp(zkx) by this scattered wave. the stationary scattering wave. (11. (11.x °Vk(x).i) dk (2TT (11. (11.240 CHAPTER 11.
11.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.8 Scattering of a Wave Packet so that with Eqs.e.174) r /k0(^. however. (11.5.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. Here ^ (x' — xo. U(T) = ^ V ( r ) . r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.y)eiko'Xoe'[fcor£.*0ko (x .5 Scattering of a wave packet with scattering centre O.r')l/(r>fc(r').x 0 .to) has t h e form of t h e wave packet in radial direction as indicated in Fig. *.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. the solution can be written iMr) : (2TT)3/2 ^dr'G(r.x 0 .11. Adding a suitably normalized solution of the free equation.z Fig. and determine the Green's function G(r. T h e scattering amplitude is t h e same as t h a t given by Eq.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . r ' ) = 47r<5(r . A particular solution of the Schrodinger equation is — J G(r. i. '.fco(tto)/ft]^0o)(x/xo. t) . Solution: We define the Green's function G ( r .171).to).(«)/ ipko(x . E x a m p l e 11. (11.r')V(rW).r ' ) . (11. r ' ) by the equation (A + fc2)G(r. 11. . t) f. (A + fc2)t/. with the energy of the maximum momentum of t h e wave packet.^y /k0^.r').154).167).
/2 dk'd(—cos 6)dip iki. 11. Thus we consider the contour integral taken around the contour C+.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane.Rek' kfc/2K Fig.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. Hence we write G(r)G+(r).=k+ie/2k 1 d f. e > 0 small.242 CHAPTER 11. displaced slightly away from the real axis as indicated in Fig.6. The Coulomb Interaction We find a set of Green's functions G(r. . one obtains another Green's function given by G_(r)= r eikr. These are the socalled outgoing and ingoing Green's functions. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r .6 The contour C+ in the upper half of the k'p\ane. G+(r) 1 d dk' lim e>0 TTT dr Jc+ fc'2 — (fc2 ~ik'r 1 d 2ni nr dr \k' + k+ ie/2kj ik r J__d_ 7rr dr 2ni N ^ residues k.fc ) oo dfc' 1 d nik r 7rr dr / .O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. dk' ik'*r 2TT2 fc'2 . Imk . we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). which is such that the contribution along the infinite semicircle vanishes. 11.
z) = W^a. 2ikr)] [Wi{l + 1+11. z). z) + W2(a.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). under the stated conditions 2/07 = Z\Zie2 = —Mo.0 + W2(l + 1 + iy. b. b. (11. whose continuation to infinity is again given by (cf. .£).145)) F(a.e. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. b.2Z + 2.g. i. Then yW (11.178) (11.e. 2ikr) s For comparison with calculations in Chapter 16 note that for h = c = 1 together with 2mo = 1 and hence E = fc2 the quantity 7 here contains a factor k. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7.z2 27fc 1(1 + 1)' yi = 0. (11. Thus we proceed as above.140) (previously we solved this only for the bound state problem) § . Before we introduce the general form of the socalled partial wave expansion. If we proceed from the radial Schrodinger equation.2l + 2. (11. e. Then £^<fc + (2l + 2 £ = ~2ikr.11.176) " 0—A ~ (I + 1 + iy)<t>i = 0.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. i. (11. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems.9 Scattering Phase and Partial Waves 243 11. Eq. 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.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. where Mo is a parameter in Chapter 16. (11.21 + 2. Eq.
+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.ii)' (11. The Coulomb Interaction Using Eqs.180) ^ ' (11.«7) „ . (2kryr{i" V 7 .181) so that eldlT{l + 1 .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 . T(2l + 2) (2fcr)*T« (Alliy ikr . .n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .ry) = e"^r(Z + 1 + ry). Then e2i5< = Reie Re~ J2i9 9 = Si.' .i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.146) and (11.i f c r + i 7 In 2/cr+i7r(i+l)/2 + We set T(l + 1 + 17) = r(z + i + i 7 ) ' r(z + I .i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .183) r(Z + 1 + 17) := Reie.244 CHAPTER 11.j syyll iy il( _ . (11.l e*"" + T(l + 1 + H ) T(l + 1 . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.
11..e.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). g = e«* = rSi + 1+ *V*ri (11.e._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).e. In the present quantum mechanical considerations I is. a positive integer.187) contains almost the entire physical information with regard to the Coulomb potential. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. (11.114a) reveals). . i. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element..184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. i.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7).n ' . That the energy E appears in Eq.. as we saw.2. where the continuum of the spectrum starts. considering I —• an(E). 245 (11. considering I as a function of E extended into the complex plane.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. However. i.188) in the form y/E.1. . kr) the regular solution with the following asymptotic behaviour: r 0 (11. (11. (11. with Eq.139) n' = 0. One defines as regular Coulomb wave Fi(j.e. i. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . For instance the poles of S are given by Z + l + z7 = .
246 CHAPTER 11. we obtain the partial wave expansion of the scattering amplitude as follows. (11.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. In these cases. 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. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. For the Coulomb potential a typical such trajectory is indicated in Fig. Singh [252]." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. 11.7.7 A Regge trajectory of the Coulomb potential. (11.191) .189) where as before the incoming wave in the direction of z is ikz ipin = e (11. The Coulomb Interaction and plotting these for different values of n. we obtain trajectories which are known as Regge trajectories. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. and thus in general. 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. 11.
eikr\P^cos9).r ^ V ^ Wostf).195) with the scattering matrix element S(l. It follows therefore that pikr ^scatt ^ f{0) = Aegeikz = ~ ^(21 + l)Lilneikr  rHe2iS<eikr\pl(cos9) eikr\pi(coa9) eikr " 2 ^ E ( 2 Z + l){e i ' .g.196) "See e. . (11. (11. The mathematical expression can be found in Tables of Special Functions (there cf. 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„ ' .193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.^2l + l^S^k) lik (11. (11.192) Thus the incoming and outgoing spherical waves have a definite relative phase. r e.~ J > Z + l)Lil*e~ikr .k)=me2iS^k\ in which r\i^\ indicates absorption.22. The potential disturbs this phase relationship (by delay or absorption or redirection). Magnus and F.i f c r .9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. 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. 21 . Oberhettinger [181]. pp. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.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. W.
This page is intentionally left blank .
Similarly difficult is.e. These newer methods will be presented in detail in later chapters. which in one form or another. however.* *See e. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. namely the instanton (or more generally pseudoparticle) method. was developed only in the last two to three decades.e.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. to explore first much simpler models in order to acquire an impression of the type of results to be expected. one of the main and now wellknown methods to perform such calculations. i. that the explicit derivation of such quantities in nontrivial contexts. Such a vital quantum mechanical phenomenon is socalled tunneling. i. in texts on quantum mechanics is rare. states of a finite lifetime.Chapter 12 Q u a n t u m Mechanical Tunneling 12. Razavy [236]. It is not surprising. therefore. in general. the computation of finite lifetimes of a quantum mechanical state. which a particle in classical mechanics would never be able to intrude. In fact. as — for instance — for trigonometric or double well potentials. M. can be found in any traditional text on quantum mechanics. For various general aspects and applications of tunneling we refer to other literature. It is very instructive.g. that quantum mechanically a particle may have a small probability of being in a spatial domain. 249 .
i.e.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics. t) was developed by Born in 1926.t) that the one particle concerned is somewhere in space at time t.t)\2 =1 (12. (12.t) (12.1) as the probability described by ijj(x. According to Eq.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x.t) = HiP(x.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj..2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. r (12. . so that p = T/>2 is the probability density for a spatial measurement at time t.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. Quantum Mechanical Tunneling 12.4) (12. ihr(x.250 CHAPTER 12.t). M^ Then dt + Vj 1=0. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.
However. Gosdzinsky and R. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). 2m E 2 V0 > 0. 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. even for discontinuous potentials. we can do this also here. 'Maybe the reader dislikes being confronted again with the onedimensional case. as shown by P. Tarrach [119]. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. although in general small. If there is no absorption. The Schrodinger equation for this case is* ip" + [A. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. pp.2a6(x)]il> = 0. This case plays a special role. In much the same way that we define there reflection and transmission coefficients. 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. If ^>(x) and V^>(x) are known at every point x. ~hr> k = n^ (12 7)  .3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. 29 . This effect is known as "tunnelincp. the equation can be integrated at every point. Nonetheless. 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. Schiff [243]. From this follows that both p and j have to be finite and continuous everywhere. as we mentioned above. Thus it is possible that the probability for the particle to be in such a domain is nonzero. higher dimensional delta potentials do not have properties like that in one dimension. the problem is analogous to that of the scattering process from a potential in a onedimensional case. t For the singular delta function potential a modification of these demands is necessary. 12.3 The ShortRange Delta Potential It is very instructive to consider in detail a potential V(x) = VQ6(X) with the "shape" of a onedimensional delta function. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous.6) . Delta potentials in more than one dimension do not permit bound states and scattering. which is the case we consider here. (12. 1 0 ~x o a= f See L. Such a region is for instance the domain where V is larger than the total energy E.12. where „ mpVo V(x) =+V0S(x).2 .33.
e. a — ik £Qr < { eikx a eikx •t" .14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.= 2a^(0).11a) ikC . 2 (12. (12. The continuity conditions at x = 0 are therefore for V : and according to Eq.: (i2iQ) Ce for x > 0.[^']o. i. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).ik(C .11b) 2aC. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx . [<//]0+ .' : It follows that ik .e 1 + B = C. a 2 — a 2 a — ik a2 + k2 ' .12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .1) .9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.13) ~A R= for x > °(12. .ikC = i. except in domains of the potential V. (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.2 a J 5{x)4>(x)dx = 0.8) (12.(ik .e. i. (12. e Q (12.252 CHAPTER 12.12a) (12.e. (12.9) for ij.ikB) = 2aC or 2a(l + B). i.
or h2 E = —a2. Omnes and M.2 o ^ ( 0 ) . Proissart [223] or other books on scattering theory like R.17) da#(s) 2 = 1.e. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. ( 1 2 .2 a C .19) Thus the delta function potential supports exactly one bound state.K 2 < 0.e. as coefficient of the outgoing wave.e «x = J \ CeKX for for x > 0. See e. z<0. .16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. i. and C follows from the normalization: OO A.(K)C = . oo (12. i.) and V>'(0+) . R.K C . possesses simple poles at a ± ik = 0.2 = . (12. f Ce~KX c. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. We write the solution ^ so that ^(0+) = ^ ( 0 .ip'(0) = . at k2 = —a2.g.i 8 ) . Thus ^ = Ce. when normalized to 1. . K = a. Newton [219]. 'Note that the physical pole has negative imaginary part. the quantity T.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. 2m 0 (12. = .15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential.a l s l.12.
12. 12. (12. or in the lectures of E. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.22) + De Feikx + Ge~ikx *We follow here largely F. x G [a.1. \x\ < a. p. V(x) V=E a v=y Fig. x€[a. 55. 58. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0. Quantum Mechanical Tunneling 12. oo]. (12. p.a]. Fermi [92]. .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.1 The square well potential. Schwabl [246].254 CHAPTER 12. We want to consider the case of stationary states for E > 0. —a]. The latter contains also specific applications in nuclear physics.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 set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.20) for for \x\ > a.
I (12. fc Af(a) = .28) Prom these matrix equations we obtain ( B ) = MWM(fl)_1 ( G ) ' where M(a)1 = v ^12'29) i / [i + n i K\eiKa+ika *)e (i t\ K\„iKa—ika fcje \ ] J fl2 30) • J ' 2 \ ("I — K'Jgita+Jfca Q _i_ K\p~iKaika \ . (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.24) yields F ^ G.12. (12.e.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . Fikeika .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. M(a) . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .iKa ? ikAe~ika and . ).f )e" t\ K\0—tKa—ika . (12. (12.ikGe~ika. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. ( V C n vD ).ikBeika = inCeiKa + DiKe~iKa. .26) K^gifoa—i/ca fc. (12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca..
i. = < Ce~lKX + DeiKX .i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 . . x € [a.32) [cos2«. (12.31) we have A AT(a) = )MFo B ^r)sin2na .4T(£)e ifci.a].34) into Eq. xe[a. (12. (12.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. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.e.a ] .256 CHAPTER 12.e sin 2na IJlika e B = r)F sm2na. we obtain A = F( cos 2na — . (12.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.(1 + e 2 ) sin2 2na 1 .22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.—2ika [cos 2/ta + \e sin 2«a]e~ (12.oo]. i.35) with with with x G [oo. \R\2 = T(E)^sm2Ka (12.a . Quantum Mechanical Tunneling Hence with the complex quantities IK (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.37) .
which is satisfied if either / = g or / = — g'1. formula 406.39) 2 i.12. in the domain of A imaginary and K real. 2m0\E\ h? ' (12. i. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. and k real. for VQ < E < 0. (12. nor when both parameters are purely imaginary.42) These equations have no solution for K.44) . However. (12. . (12. cot(fca) = i— or tan(fta) = K K . Li ( — H— I sin 2na = 0. (12.43) Thus we set k — —i ' H .34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.12. 2\k K. (12.e. 83.38) as expected.4 Scattering from a Potential Well On the other hand according to Eq.35) and using Eq. Dwight [81]. We see in particular from Eq.e. B. an equation of the type f~1f = g19.41) i.31): 1\T(E)\2 257 1 .\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12.e. (12. p. (12.40) we obtain cot(na) — tan(/«j) — i K IK . (12.e. i. they have solutions.
i. . The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.g. and is itself infinite there. 2Ka = nn.45) These are the equations that one obtains from the connection relations for the case of Eq. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. .47) These states in the continuum are called resonances. i. (12.49) Taylor expansion of the denominator in powers of (E — ER) yields . A = 0. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — .65. (12.c o . 64 .46) This condition implies the eigenvalues E^ER = n2^^ V0>0. (12.35)) for E > 0 sin2Ka = 0. Quantum Mechanical Tunneling .l . The remaining wave part becomes exponentially > decreasing. (12.258 and obtain the equations CHAPTER 12. Eq. i. pp. 2m0\E\ Ktan(«a) = — y——^— . (12.e. Schwabl [246]. Also. „ .50) See e. Then according to Eq. We now return to the scattering states. ± 2 . The amplitude of the transmitted wave assumes its maximum value where (cf. (12. 2m0\E\ and Kcot(Ka) = y—^—L. (12. ± l .+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .fl = ( . .34). .42) which then leads to bound states with associated even and odd wave functions.e. n = 0 . (12. T(E)e2ika = — r . k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. This can be understood. = 2/rn (EER). (12. At their positions cos(2/«z)£. around the values given by Eq.20).e. from Eq.l ) n . t2 2 (12.46).' We observe that T(E) has poles at those values of E which supply binding energies. F.48) We expand T{E) around these energy values. t&n(2Ka)ER = 0.
This result is physically very plausible.' ' 5a \V / Fig. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. V(x) = 9(x + a)VQ6(ax). T(E) has poles at E= ERi: (12. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. . These are precisely the resonance states. 2a<x<2a. = 0.2 The PenneyKronig potential. 12. A potential of this type is known as a PenneyKronig potential. 12. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. 12.e.^ l / . s \ V0 /• ~\ 'X . a 3 a.' ''' 5a 3a \ a ~~~.2. We restrict ourselves to the period — la < x < 2a. V(x) x=2a x= =2a . .5 Degenerate Potentials and Tunneling In the following we consider a potential which is very similar to a periodic potential and can serve to illustrate a number of aspects of the process of tunneling in quantum mechanics.12.52) Here ER is the energy of the resonance and 2/F the lifetime of the state.51) i.
. i. the differential operator is even in x. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.2. In view of the periodicity we have ^(x) = ^(x + 4a). . = EVip(x).a ) = 0 = ^(a).e. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.e.e There is the additional boundary condition (for Vo = oo) ^ ( .. HV^(x) = H.53) *w«w^( « j)'*^{i i ^ .Vo) n. E^En = ^ ^ t n = 0.56) 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.55) The Hamilton operator of the problem. (12. and we choose (12. and hence for = Eip(x).2 < a . (12. .1. The particle of mass TUQ whose quantum mechanics is described by the Schrodinger equation is free to move in the domains under consideration to the left and to the right of the central barrier. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x)... i. with V as parity operator.2 = W h 2 _„2 = 2mo(E ..260 CHAPTER 12. We set .
VV(°) = °. (12.2a.2a]. This wave function is no longer restricted to the interior of a box. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ .2).a]. This means that i>±(x) = <  = ±1. (12. We expect. (12.Be Ka (12. of course. however. 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. (12. = Ka ±Asin(k2a). xe[a. we shall not exploit here.. V'(O) = const. are ^_(0) = 0. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].63) = KBe~ F = . that these are such that for VQ — oo they approach those of the previous case.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . At x = ± a we now have to connect ip and ip' of neighbouring domains. (12. ^kAcos{k2a). so that (W meaning Wronskian) W[tl>+. ±Asink(3a — x) with x€[a.il>]?0.. V+(0) = const. — a].12. BeKX±Be~KX with xe[a. The central region of the solution is the classically forbidden domain. which. 12.59) Asink(x + 3a) with x & [~2a.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.2a]. — a].a]. We > obtain even and odd functions ip± by setting ^ = ±1.60) The boundary conditions that ip± have to satisfy.58) D sin k(3a — x) with xE[a. We therefore have different boundary conditions.
e.— (1 ± 2e~2A + O (—). Taking the ratio of the equations in both cases. (12. for VQ — oo.68) .2 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.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.262 CHAPTER 12.2 e . (with reinsertion of nir/2a for k) 2ak ~ nir . However.+ • • • 1! cos^(n7r) 2ak . i. i. say. + 0 ( e . k = fc0dd.66) Inserting these expansions into Eq. It is plausible therefore to use appropriate expansions. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.67) The value of k belonging to the upper sign.e.mr)2}. (12.even.nir) + OU2ak . = AA.e. In general one now rewrites the equations in matrix form. and is — as we see — less than that belonging to the lower sign case. we set tanhx = l .nir + 0[{2ak . i. 12. the odd solution with.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . (12. F ^odd (12. in the present case it is easier to continue with the above equations. is the one belonging to the even solution. For large values of Vb.3 Level degeneracy removed by tunneling.e.nir)2}} = 1 ± 2e~2Ka. say A.65) and we expand tan(2afc) about nir. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. (12. where Akoce'2™.64) we obtain ~[(2ak k . Quantum Mechanical Tunneling Fig. i.
. as indicated in Fig. 2m0E „ 2mnE .. Example 12. 57. 12. and is a reformulated version of a problem set in lectures of Fermi.4. b > a. y(x) = sin and y(xo) = 1. '. w(x0) •• = = mogy(x). exp r rb rb / •J a 2mo (V(x) .2 dxy/V(x) (12. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. but quantum mechanically there is a finite probability for this to succeed. Compute this probability. V ^ 0 are given by the equations ^o + " J . .^ 0 = °' Hence in dominant order •00 — exp . 12. 1pV 2ro 0 4>o '. Fermi [92]. . p.3.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. (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.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. .70) xo = (a + b) 2' . i.12.V{x)\i.** " W ~^7~ Fig. 0..e. Solution: The wave functions i/>o>Vv in regions V = 0.v = 0.4 A car meeting a bump in the road. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. 12. Classically the car is unable to overcome the bump. 2mn. ^v + ^[E.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. exp .
we obtain ( .2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.e.1. . ^ ) = 2 ^ . 950. : 2 ^ B ( ^ . p. and assuming a length of b — a = 100 m of the base of the bump. 14.l .71) The integral can be looked up in Tables of Integrals.72) Assuming a mass of 1 ton of the car (i. We find™ (with ( . p.**.2.807 m s . Jahnke and F. 369 and 8. M. one evaluates an approximate probability of exp[6. which give /•TT/2 dx sin'M .y) is the beta function (see Eq. p.4 x 10 3 9 ].17)). 1000 kg) and using the following values of the natural constants: g = 9.621.9191 ~ 1. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. We can now evaluate the probability.2 . Emde [143]. dx sin ' x •• Looking up the value of the factorial in Tables. n I . h = 1. Gradshteyn and I.055 X 1 0 _ 3 4 J s .1 / 2 ) ! = v^r) TT/2 /. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. (12. formulas 3. **For instance E.384. (15. S.1.264 we have ro / dxyfV(x) Ja CHAPTER 12. Ryzhik [122].
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. we consider the corresponding case of a particle above the flat surface of the Earth. This is the problem of Galilei in quantum mechanics. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.* In Chapter 14.3. in the book of G. 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. we consider the linear potential in three space dimensions that forbids the particle to escape. Then considering the probability distribution determined by the squared modulus of the wave function. * A highly abridged treatment of this problem is given. 265 .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. 13. in Chapter 15. 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. This is the quantum mechanical version of the problem of Galilei in one space dimension. Thus. each of which propagates with the classical momentum and energy of a Galileian particle. Finally. Example 14. whereas the first case permits only a continuous spectrum. p. Siissmann [266].1 Introductory Remarks We distinguish here between three different types of linear potentials. the last case allows only a discrete spectrum. for instance.
4b) .t) = . We consider here the case of Eq. 13.4a) f°° ${k.2) which is the case with a discrete spectrum.3) In general the momentum representation is of little importance.1) where g is the acceleration due to gravity and we can put Vo = 0.t) = mogx + V0. and yields rriQX = mog. We see " that the case g < 0 can be related to the reflection x — —x. The present case is an exception and a good example of its applicability. The equation of motion follows for instance from the derivative of the constant energy.e.1). (13. (13. (13. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle. from d_ m x2 0 dx + V(x) 0. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.266 CHAPTER 13.= l V 27T Joo dxelkxip(x. (13. (13. (13.t). V(x) = —mogx. g > 0.t) = . i. x > 0. Therefore we use the Fourier transforms il){x.t) = c\x\ + V0.1 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. c> 0.= l V 27T J_oo 1 dkeikxijj{k.2.t). Recall the classical treatment of the freely falling particle.t) = m0g with potential V(x. A different > "linear potential' is the socalled "confinement potentiaF V(x.t).
7) applied to every quantity in Eq.6). (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13.t).e. Equation (13. (13. We rewrite this equation in the following form th (13.8) With this result and the substitution (13. (13. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt. d \ ~f1 dkr[ k + m0gt ^r't (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.t) dk nikx k=oo '2K (im0g)il)(k. multiplication by a test function and subsequent integration). The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i. (13. Applying the total derivative d/dt to this function.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.t).6) the argument k is then replaced by this.6) and wish to solve it.9) . To this end we consider the following substitution in Eq.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.< (13.6) the latter equation becomes d ~( m0gt ^(^M<=+=?. i.13.t) the expression (13.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.d m +imo9 . • fd ~(' m0gt %h .e.7) In the wave function on the left hand side of Eq. (13.3) yields us now by inserting for ip(x.t) m0gx /2TT m0gxip(x. we obtain the differential operator on the left of Eq.t)im0g^k.t) h2k2 d ~J(k.t) dketkxiP(k.
1 we verify t h a t this solution indeed solves Eq. we have to reverse the substitutions and replace in (13.6). 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. t ) . (13.= k rmt. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.10) k by m0gt back : k —> k h The solution of Eq.10) Note the amplitude function ipo(k) in front.t In Example 13.11) solves the partial differential equation (13.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. We define the wave number kt as kt.11) in a neater form. Solution: We apply the two operator expressions on the left of Eq. (13.e.6) is then $(k. On the other hand the second operator expression implies dtj>o imog^—ip(k.11) *[k?f.6).t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. Our next step is to reexpress the solution (13.6) to the solution (13.11).6).i) .t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.6). To obtain the solution of the original equation (13. E x a m p l e 13. Thus we have first (the second term arising from the upper integration limit. (13. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.t). (13.268 CHAPTER 13.t ) ' U ( f c .12) . the following result h 2moi tp[k\ ^—.gh [ d t ' j f c + ^ ( t ' . i. We have thus obtained the solution of the transformed equation. (13.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.
o o 2 —  exp iktx h mZgH' .13.o ut = 2mo —kf 2mo A.= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . (13. m0g (13.19a) . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.18) h =k mogt p h S' h .gfci + 2 g2m0 2/i 2 ' (13.t) into this integral and recalling the property (13. we obtain rp(x.+^f (13. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * . (13.t) = .11).} fA ' _ _tkt 2 k+ t) 2 k .17) = J_/ C X ) h2 [It(k) .14) (13.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.2„2+2 mfaH 3^3+3 .I0(k)]. .11) can be written j'A ^.t) defined by Eq.15) we obtain / = . _ k t  h2 3 Inserting here the explicit expression for fct. Inserting ip(k.13) rn0_t+2 + mlg2t3 .— [It(k) ./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.4a).16) Next we evaluate the Fourier transform of ip(k. 27T ^ . we have ip(x.. .
The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.ut> = Jo t2 g2m0 i 3 k t . 7 u>t'dt' (13.18) into a • different form in order to be able to perform the integration with respect to k.270 and CHAPTER 13.o o Jo This result can be interpreted as follows. fm0gt\ \ 2h ) (13.23) .2. the relation (13. To this end we convert Eq.19b) h3 3 we can rewrite ip(x.t) =y= dkip0(kt)) exp i< fctx V27T J . (13. (13.22) We can rearrange these terms in the following way with the ^dependent contributions contained in a quadratic form: ht 2moi ht 2moi k mpx ht mogt 2h mogt\ 2h J + IX .v = kh/mo. (13.20) (13. These wave vectors vary in accordance with Eq.21) In other words.19a) exactly as for a falling particle. 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.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .t) as a decomposition into de Broglie waves. 13.t)\2 for t — oo (distant future or past). Linear Potentials f dt'. 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.The associated kinetic energy of the particle is mv+ = huJf.20) expresses the wave function tp(x.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.t) as 1 I00 ip(x. Such a decomposition is possible only for homogeneous force fields.
t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13.t) = 1 1 oo 271 27T /. integrate and then take the limit fi —> 0). We evaluate the Fresnel integral in Eq.18) thus becomes ip(x. fi > 0.QX 2 ^ 0 7 (m.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics. (13.t) 1 exp r .24) We now set fit / 2m 0 V" 0 V m0a. 2h h 24ift ' mogt^ r m.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.29) . It follows that ip(x.28) (for a verification replace a by a + ifi. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. being now independent of £.26) d£e ie £ kt^ m0' 0. /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. m0xcl ht m0gt 2h gt O v: °° 2hti (13. Then ip(x./oo dkip0(kt) exp l x ixI exp oo .e.2 The Freely Falling Particle: Quantization The integral (13. (13. rnogt m0g2t3 tX^r~./m 0 a.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ . TOO#A ft and.26) with the formula 1/2 (13.13. may be put in front of the integral. (13.
1) has only a continuous spectrum.dA 4>o(h (13.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. 1/2  ±^(2mlg/h^x^ / 2i (13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . (13. These are incoming and outgoing waves.1 A turning point at x = —XQ.e. For x exponentially increasing and decreasing solutions are possible. Linear Potentials Replacing here x by the large time value x c i. xc\).272 CHAPTER 13.31) and obtain from Eq. this expression can be brought into the following form: if *\ 0 ip{x.t) = eiEtlhct>{x) (13. (13. V(x)=m 0 gx V=E Fig.33) oo for E <C \mogx\. . For stationary states we set xl){x.t) ~ W —exp i V irlt m ktxd J cot. 13.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. kt and ut 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.e. at which the classical particle bounces back. The Fourier transform is given by the relations (13. 13.e. exponentially decreasing. t) = Eil)[x. E)dk. probability. i. t).00].t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m.4b). The point — XQ. cf. However quantum mechanically this is possible with a small. but also d 00 dt Eijj(k. Fig.e. at which E — V = 0.e. 1 f01 dketlcx^{k. (13. and we can establish for the resulting continuum solutions the orthonormality and completeness relations.34b) It therefore follows from Eq. is reflected. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. and hence is continuous and extends from —00 to +00. i.e.t). its momentum p would be imaginary there. which is not possible for real particles. i. Akx (13.35) It follows that /W tm0g \ 2m0 i.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. i. H2k3 4>{k) — cexp Ek (13.0 0 dkelkxE^(k. we . so that ih dtp{k.4a) and (13. t) = — = / V27T J .36) m0g \ 6m0 with c = const. J \ „ (13. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00.34a) Then d &k 1 f°° ih—il>(x.1.6) that also for 4>{k. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. i.t).t) = dt (13.13.e.
h? h2 .37) (eikxdk.39) dk4>*E{k)4>E.e. 13.M *K Joo Setting dt cos ( ts n Jo st). </>E(x) = .38) we have for <j)E{k): OO /"OO / oo (13.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.36) into (13. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). (13. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). i..36) (with c replaced by (13.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .41) J—oo Next we insert (13.a : ' ) . (1340) i. V 2lT J Mk) = .e. apart from a constant phase which we choose to be zero. / <>~ikx VZ7T J (13.40)) into (13. (13.L dxeikx(f>E(x).38). In a similar way we can verify the validity of the completeness relation.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.43) y=2^x+2^E. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk'). (13.42) The Airy This integral can be rewritten in terms of the Airy function function Ai{s) can be defined by the following integral: 1 f°° Ai(s) = — / eft exp i ( st ..L [ dkeikxMk). / dE(j)E(x)(t>E(x') = ^ s .£') = <*(#£')• 1 27rm0g' = /o .
.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.45) One can now derive normalization and completeness integrals for the Airy functions. / s M .^x + ^ E (13. 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= ^ .44) is then expressed in terms of the Airy function: .13. the function 4>E{X) i and setting s y = — = 1 .^ .46) This follows since with Eq. (13.J * . and we shall see t h a t one branch has periodic behaviour. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. (13.43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x) s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) r 2^rJ 2W dxe^Qe^^eU*3*3) _^Pi(Sy~tz)ei(s3t3) S(y = j£j**w I ^LJt{yz) 2TT _ We will encounter the Airy function Ai(s) again in Chapter 14 in the matching of exponential W K B solutions to periodic W K B solutions. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s./ 2m 0 E (13. whereas the other is of exponential type. . For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.
(13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . (13. (13.51b) . 13.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0. i.r 3 sin 30 ). We want the integral to exist and so desire that lim e**^ = 0. i.f dzJ*W 2TT JC with z = reie.49) (13. (13.. (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .276 CHAPTER 13.e.4 The Saddle Point or Stationary Phase Method The integrand of Eq.r 3 cos 30 J + i ( sr sin 9 . .TT < 30 < 0. .2 The contour C for s < 0 with ends in the angular domains (a) and (b).43) contains the phase (with x — z): > *(*) := sz . Linear Potentials 13.TT < 6 < 0 for r > oo. o (13..e. Thus we must have sin30<O.\zz. or n <6 <TT o for r > oo.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.
oo] as indicated in Fig.51a) and (13.55) s < 0. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. —i ${zs) = szs . 0 = &(z8) = s~zt s = za „2„2i6>.48).52) = exp(±i7r) Let s be real and s < 0.13. Then 6S = ±7r/2. (13. so that exp(±2i9s) — 1 and — s — rl.54) i ¥~s)3/2' (13. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.58) .2.S )V2 (13.e. provided its ends at infinity satisfy these conditions. Such a piece can be found in the neighbourhood of the socalled saddle point zs. (13.57) $"(z s ) = 2% s. and the main contribution to the integral comes from the a region around the saddle point. i. Then Ze TeC : V^se±i7r/2 = ±i\fzrs.e. (13. we choose zs = iy/^s.( \ / ^ ) 3 2z. 13. Inserting (13. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.51b). Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(. i.4 The Method of Stationary Phase 277 The integral (13. at which the phase becomes stationary.54) where a is the new variable with —oo < a < oo.55) into (13. (13.45) exists for the contour C. Then the ends of the contour lie in the domains specified by Eqs. where the derivative of the phase function vanishes.56) 2n(s)^Jc But I daefr"?*"^. and we set with a real: z := zs + a (S)V4' (13. (13.53) Since we shall have the contour C in the lower half plane.z3s and V^s + .
42) into an equation with a more appealing form. V^s / 4 V3 (13. This is an important aspect in the WKB method to be considered in Chapter 14. (13.45) 4>E(X) ~ Ai(z).60) The reason for describing the point zs as a saddle point is that the integral in Eq.2/32u = 0 (13.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). V 2h2 (13. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}.491. Gradshteyn and I. With the help of the substitutions (13. (13.44) we can convert Eq. Sec.62) This is the Airy differential equation with one solution 4>{z) = Ai(z)..59) In Example 13. S. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.32) therefore becomes dz2 + zd) = 0.278 CHAPTER 13.64) tSee I. For (—s) — oo the integral converges towards T/TC.44) and (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction.s ) V 4 exp \(s) 3/2 (13. (13. as one can see from the following equation^ ri' + Ffz •2^2. 8._ 1 . . (13. M. According to Eqs. (13.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.56) decreases exponentially around a — 0 for real values of a. > so that Ai(s) 2 v ^ ( . 2 s 3/2 . Ryzhik [122]. cos .32) with the solution (13.2 we show that for s — oo: > Ai(s) 1 .
the conditions (13. we obtain f° Joo dp=—[a+ i\a\]da._ . Thus there are two symmetrically placed real saddle points on the real axis. (13. Example 13.4 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) ~ . Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. Eq.59) and (13.51b) are satisfied. 27T Ic Jc 3 This time s is real and positive. changing to the variable p. 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 . P Introducing a new variable a by setting p = ai\a\. cos  .51a) and (13.60) we have thus obtained (with the help of the saddle point method) the asymptotic expansions of the Airy function Ai(z) which play an important role in Chapter 14.2: 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.13. * " ( z ± ) = =F2Vi.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).^ e ^ / V2 4 [°° Jo dae'2^. . Choosing the contour C to lie along the real axis.. Then.s 3 / 2  . kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = . with *"(z) = 2z.z ± ) 2 * " ( z ± ) .
the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. 13.Be 2 Fig.3. . one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. Thus here / ( z ) = — z + n l n z . n!= / e~zzndz= / e~z+nlnzdz. Linear Potentials E x a m p l e 13. z o = n .3 The saddle point at z = n + ix.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.e. (11. / " ( z o ) = —n/z2 = —1/n. and hence integrating as described above through the saddle point indicated in Fig. ^/2KJW1. 13. i.280 CHAPTER 13.111). Then.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). Solution: Briefly. since f'(zo) = 0 and (z — ZQ)2 = —x2.*. Jo Jo \ . ~ 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.
for misunderstandings pp.e. Kramers and Brillouin. Wentzel [282]. Froman and P. ^ The central problem of the method is the topic of "connection formulas". 316 . 292 . A critical overview of the history of the method. Kramers [153] and L. will there be that of a coupling parameter.Chapter 14 Classical Limit and W K B Method 14. § See e. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. however. named after its first promulgators in quantum mechanics: Wentzel.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. B. ' T h e most prominent earlier expounder is H.295. the linkage of solutions above a turning point to those below.317. where *G. Froman [99]. N. of misunderstandings and other aspects may be found in the monograph of Dingle. 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. *R.* The method had. A.e. and the role played by h in our considerations here.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. Jeffreys [144]. The WKB result is therefore frequently described as the semiclassical approximation. H. For the history see pp. Dingle [70].§ 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.g. been developed previously by others in different contexts. 281 . It goes without saying that such comparisons are very instructive. Brillouin [37]. i. O.
3) . • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E .g. A(r) and S(r) being real functions of r. For stationary states ^/. In this equation we set.t)\2 = \iP(r)\2.1) (14.t) = eiEt/h7P(r) and \V(r.282 CHAPTER 14. the derivation in Example 18. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).* We illustrate the use of this condition by application to several examples. Classical Limit and WKB Method E — V changes sign). the function S being called phase function. (14. 14.7. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles.V(r)]^(r) = 0. the energy has a definite value E with V(r. for which the observation of position and momentum is independent of the time at which the observation takes place. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. V>(r) = A(r)eiSWn.
11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.10) with a continuity equation by defining as probability density p:=*** = [A(r. (14. (14.t)}2.1) and separating real and imaginary parts.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.Q A 2i 8A 1 (14. These equations are equivalent to the Schrodinger equation. V(r.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.4) (14. h2 ih— = H$ A + V(r) * . (14. We can identify Eq.t)eiS{r't)/h.+ VAVS + AAS = 0.13) . h h (14. We set in this equation.14. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.* VAVS+AAS + V(r)A. (14.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14. we obtain the following two equations: E .V(Y) + and 2VA • VS + AAS = 0. (14.12) VA A (14.6) 2mo dt now with A also a function of t.10) m 0 ^ .
21) . m0 But now (14.9) for the phase function S. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.16) BA 1 moA—. Since J = — A2VS = PVS.n.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. .10) we obtain = — [V(A2) • VS + A2AS}. (14.17) With Eqs.01. (14. TTIQ (14. (14. Next we investigate the significance of Eq. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. mo <™> (14.15) and (14. the implications of this equation are also those of the equation of continuity.+ AVA • VS + A2AS = 0.10).J = 0. (14.. + h2 AA V = — ^ . ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real.14) A2VS. (14.284 CHAPTER 14. ^ + V • J = 0. (14.20) With this we can rewrite Eq.18) ot ot This is the equation of continuity that we encountered already earlier. (14. (14.9) as 9S 1 mo^ 2 + T. ie. we have mo—(A2) + V(A2)VS + A2AS = 0. (14. Since we obtained this equation from Eq.16) this becomes m 0 1 rr + m o V .
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. This means. i. which for very small values of h yield the dominant contributions. The equation of continuity then describes the stationary flow of a fluid. they are asymptotic expansions. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.e. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ.22) (14. This result is very general. the wave function \I/ describes effectively a fluid of particles of mass mo. 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.23) (14. a singularity different from that of a pole). with the help of Eq. so that V ( v • v) = 2(v • V ) v + 2v x rotv. where.2) for the probability amplitude. However.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. V • J = 0. that the quantum theory implies an essential singularity at H = 0 (i. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. Such series are typical for expansions in the neighbourhood of an essential singularity. we obtain ^ V 5 +  v ( v v ) + V V = 0. The particles have no interaction with each other.1/h. d — = (14. r o t v = ( l / m o ) r o t g r a d S = 0. dt 2 Constructing the gradient of this equation. We deduced this equation solely from the phase S in the limit h — 0. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. This expression with h in the denominator shows. (14.e. In the first • place we require the ansatz (14.14. (14.20). An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. and hence''' ^(m 0 v) + m 0 (v.V ) v + V 7 = 0. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv.24) d 1 dr d •— = d Lv _ . or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). .
is an asymptotic expansion.In A(x). h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).28) The calculations we performed at the beginning imply the following equations (compare with Eqs.e. i = eiW(x^h. p. Messiah [195].30) ax *For details see R.26) We observe already here that an expansion in rising powers of k. (14. We set.3 14. or h2/2mo(E — V).$ 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. There are also other more specific descriptions.§ y = A(x)eiS^'h where W(x) = S(x) + . The expansion in powers of h2.286 CHAPTER 14. As remarked at the beginning. Dingle [70]. the method is widely familar under the abbreviations of the names of Wentzel. 6.3.2.27) (14.1 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. i. Sec. §We follow here partially A. similar to our earlier procedure with A and S real functions. (14. . z (14. (14. Classical Limit and WKB Method 14.[EV(x)}y = 0. 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. We consider the onedimensional timeindependent Schrodinger equation y" + ^. such as "LiouvilleGreen method' and "phase integral method'.5)) and 2 ^ f dx ax + All=0.4) and (14. 318. Kramers and Brillouin.
30) corresponds to the equation of continuity (i. 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. .35) can be integrated. Eq. Since the expression (14.e.33) 3 (S")2 .14. (14.29) becomes (note that this is still exact!) (S')2 . (14.A = const.3 The WKB Method 287 Equation (14. + (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 . 3 / 5 .\ In S'.28). (14.10)). Li This expression can be substituted into Eq.2mQ(E .35) . (14.32) s" = sz + h2s'( + •••.V) and (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.36) Equation (14.2rn0(E V) = h2 lS'A 2S'0 (14. (14.31) \a.32) is therefore (S'0)2 + 2H2S'QS[ .V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .4 (S>)2 1 S'" 2 S> (14.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .
y/2mo(V(x) . . and instead is asymptotic which we shall not establish in detail here.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .40) rx dx v "7 W)\. 1 S e e R.39) and (14. (14. Eq.' For the expansion (14. The solution is seen to be periodic provided k(x) is real.19).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). Chapter XIII. p = h/X) k{x) :Then S'n +.37) For Eq. (1.33) to make sense as an asymptotic series.27) we obtain now with Eq. it is reasonable to set (cf.288 CHAPTER 14. (14. B. i.38) h y/2mo(EV)' (14.39) where a and /3 are constants.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. Classical Limit and WKB Method has the form of a momentum.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). Dingle [70]. (14. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. E > V(x). we expect that in any case So » ft2Si.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) .41) We now ask: What are the conditions of validity of Eqs.and So = ±h / — — A J k(x (14. (14.e. (14.
42) [(^). Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.1 / 2 ]" [§0SS). E > V(x). Since k(xy 1/52\S'0 (14.36).45a) . (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.3/2 s&T + 3 {So Il« we have (cf.14. we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small. Eq.
*2 A(z) = V / 2m 0 (£ .V{x) ~ Or . i.2 Turning p o i n t s a n d m a t c h i n g of W K B solutions Let x = a be the value of x at which E = V(a).44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. h = 1) . at which E — V{x) changes sign? We note first that in this case near x = a: E . (14.48) behave in the neighbourhood of such a point called "turning point'.48) The (dominant) WKB approximations are exact solutions of this equation. V(x)) (14. around the classical turning point.a)V'{a). We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.290 CHAPTER 14. and hence (with UIQ = 1/2.e. In the neighbourhood of E = V(x).46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. the WKB approximation is in general invalid (see below).V(x) = V(a) . Roughly speaking one therefore requires E to be large in both cases. (14. (14.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . Classical Limit and WKB Method we can rewrite the inequality (14.3.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. 14. In terms of X the original Schrodinger equation is y" + = y = 0. How do the solutions of Eq. In a similar way we have for E < V(x): l'(x) <C 1.
We assume we have a turning point as indicated in Fig. i. in the domain E < V the WKB solutions are exponentially increasing or decreasing.1 there are WKB solutions on either side of the turning point and some distance away from it. (14.14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. 14.1 The regions around the turning point at x = a. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for .48). (14. On one side.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.1.3. 14.3 after stating them first here. since this requires different approximations there.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. in the domain E > V the WKB solutions are oscillatory. 14.e. and on the other side. 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. V=E Fig.1. 14. We shall not derive these conditions here in detail but rather make them plausible in Sec. The transition is provided by matching relations. 14. As indicated in Fig.
52) For a potential rising from left to right.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. Note also the extra factor 2 in yi. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . 14.292 CHAPTER 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 / —. Thus. has the same asymptotic behaviour in the domain i « a a s Ay\. in order to make the formulas independent of whether the potential is rising or falling.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. (14. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. as in Fig. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. one could simply replace everywhere /•a / pa • • • b y / Jx J x . (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. a interchanged.50) x S> a.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . (14. 14. however with the limits of integration x.H— J =^ v/exp + m (14. This means the asymptotic part suffices only in t h e exponentially decreasing case.
3 The WKB Method 293 In that case Eq.Airy overlap* Fig.54) .3.14. is singular at a turning point x = a. the matching relations have to be altered accordingly.2 The overlap regions around the turning point at x = o. (14.51) — always with the decreasing exponential on the left — is valid in both directions. y" + Toy = o. Therefore we consider now this original equation in the domain around x = a. 14. the equation V approximately linear around turning point V=E Airy solution WKB .48). (14. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. (14.e. 14. If higher order contributions are to be taken into account. i. as we saw. 2 1 X (14.a).53) Since at x ~ a. whose exact solutions are the dominant WKB approximations. 1 ~ (x .3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. the equation becomes approximately y" = (xa)xiy.46).
— a ^> 0. Cf. (13. (14. (14.291. (14. A. In the following we require the asymptotic behaviour of the solutions Ai(—z). Abramowitz and I. Stegun [1].a)X\/3.56) We have discussed some aspects of solutions of this equation in Chapter 13. B. p.294 CHAPTER 14.57a) and (14. (14. formulas 10. This is given by the following expressions.^ 4 I 7T x — a <C 0. A second solution is written Bi. . Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. or M. as > may be seen by referring back to Eqs. 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. > for a.57b) UV2 3 11 by J dx .59). J (X (i4 58)  Replacing in Eqs. we have r x * f x ' /2(x a)i/2d*=^(x . Dingle [70].5960. R.4. p.60). In particular we encountered a solution written Ai. (14.Bi(—z) for \z\ — oo. 448.57b) To the same degree of approximation as the equation y" = (x — a)xiy.55) the equation becomes the Airy differential equation d2y dz2 zy.1 / 4 by \^. i I 772 \3 + > for ia>0. With the substitution z = (x .57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. (13.
l v/2m0(EV(x))dx 7r = (2n + l).2.. As an illustration of the use of the WKB solutions we apply these in Example 14.3 The WKB Method and 295 •.l.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = .51) and (14.g. 14. As we approach a limit of this interval in going away from the turning point. n = 0.2. The latter domain is a small region where the solutions from either direction overlap. In this sense we have now verified these relations. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. are only valid in a small interval around the turning point at x = a as indicated in Fig. and hence must be proportional (in view of the uniqueness of the solution there). by Ai(z). however. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig.14. e. we enter the domain of validity of a WKB solution.(zf2 by J we obtain the matching relationsX (14. In this way different asymptotic branches of one and the same solution are matched across a turning point.3 The anharmonic potential well. These solutions.52).1 to the quartic oscillator.59a) . Example 14. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). We summarize what we have achieved. In the direction to the left as indicated in Fig. 14. (14.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution. 14.
. .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 . yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. Here we are interested in the discrete states (the only ones here). . The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.. Let E = V(x) at x = a and x = b.y o dx TT T provided 6 _ 4 /. Classical Limit and WKB Method where x = a. (14. regions 1 and 3 in Fig. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. j/i. Then around these points the potential would be well approximated by a linear potential (as we saw above). We argued earlier that the WKB method is suitable in the case of large values of E. . the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J.2. 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. Hence we search for solutions which are exponentially decreasing in the far regions.3. n = 0 . 1 . 2 n = 0. .59b) Solution: We consider an anharmonic potential well as depicted in Fig. (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. (14. According to Eq.51). In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. as in the case of the harmonic oscillator itself. The condition for this is ya{x)=yb{x) for xe(a.e.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). 2 .296 CHAPTER 14.60) Using the matching relation (14. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. 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 ( . 14.44) this implies large values of n.l. 14. (14. (14. b are the two turning points.3. 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. ^ A = (2n + l ) . i.b). /n = s u ^ TJcos^ya TJ+cos^a T J .(x) = c ' v Acos I / — I for a < x < b. Sometimes this distinction becomes imprecise. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule.
64) (14. 14. 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. This relation is sometimes wrong. 2 .1. n = 0. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. Dingle. equivalently. (14. with n — 0.e. . 2. In this socalled "old quantum theortf (i.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. B. Bohr. .x number of turning points h. Sommerfeld and W.63b) *The author learned this in lectures (1956) of R. A.x number of turning points TTH.63a) or. . (c) V Fig. Wilson by supplementing classical mechanics by Planck's discretization.14. . ri2 I Jqi :pdq n + 1 — . . (14. The corrected form* is. pdq = n + 1 — .4 Three cases with different pairs of zeros.1.4 BohrSommerfeldWilson Condition 297 14. and knows of no published form or script.
4(a). this is a trigonometrical zero. 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. i.298 CHAPTER 14. where n —. Case (a): The case of two trigonometrical zeros. together with q — — oo. this is an exponential zero. In quantum mechanics we have dq2 + Ki/j(q) = 0. where /(g) is positive. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. 0(g) ex .e. i. i.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. (14. (14.e. 14. 14. i. / • n an integer > 0.66) J<1\ Thus in this case the socalled old quantum theory is correct.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. (14. There are three cases.e. Jqi ^ fpdq J =2 pdq = (n + l)h. We have the case of trigonometric solutions. as illustrated in Fig.4(b). Before we continue we recall from Eq.65) We use the knowledge of these solutions to find the eigenvalues of the equation. the case of tp(q) — 0 at q = gi. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. if /(g) is negative. Since sine and cosine have zeros spaced at intervals of TT. this means dqyf—f = (n + l)7r.67) Case (b): Next we consider. where /(g) is negative. In this > case the wave function for g > go is .
67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa .e.63b).63b) to obtain the eigenenergies of the quantized harmonic oscillator. Therefore their arguments differ only by (n + l)ir. Solution: As in the case of the quartic potential. For the wave function to be exponentially decreasing to the left of qo and using Eq. or Pdq (n + l ) " 2 h.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.3. the sines must be proportional.70) We can therefore summarize the results in the form of Eq. ri\ / J do y/f(q)dq (n + 1) 1" (14.14. i. Hence we have pdq= i n +  . (14.63a) or (14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. Thus in this case the socalled old quantum theory is wrong.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). (14. Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. E x a m p l e 14. 14.69) pdq = (n + l ) " 4 h. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.68) (14. Since both cases refer to the same region qo < q < q'0. i. this is the case of two turning points. and • • for the wave function to be exponentially decreasing to the right of q'0.4(c).2: The harmonic oscillator Use the relation (14. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . As illustrations we consider Examples 14. (14.2 and 14.e.
^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. where g is the acceleration due to gravity. permit this although with rapidly diminishing probability.5 A particle above a flat Earth.28 x 1 0 . . g = 9. Example 14. Calculate its quantized energy.5. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B .63 x 10~ 3 4 J s .300 With potential V(q) = 2ix2mov2q2.807 m s . 14. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. 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. 4 / x CHAPTER 14.71) For m 0 = l x 1 0 _ 3 k g . as illustrated in Fig. and h = 6. :: V=m0gq Fig.2 3 J x (n + 3 / 4 ) 2 / 3 . in principle. 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. one obtains E ~ 2. 14.V(q))dq The result is E/u. so that E = hu{n + 1/2). Thus there is one turning point at E = mogq.2 .
2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). d dE V)={n+)n. Solution: The proof is contained in Example 18.7.4: 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 £ = .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 ( — = .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 .) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB. (14.5 Further Examples E x a m p l e 14.5 Further Examples 301 14. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' . ra 2/ = 0.14.( 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 — = .= — <*— = h(n+\.f V ft h JXl h 2TT (14.75) which verifies the formula immediately.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.l.2.72) \\2dxp=Qdx^{Eh Hence. with p = modx/dt. E x a m p l e 14. Chapter 16) one obtains the quantization relation (for Mo = const. In the case of screened Coulomb potentials (cf. 1 . 9 . (14. and u> is the oscillator frequency in either well.
78) (a) We determine first the three turning points at qi = <?i. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.~ \ (14.302 and T CHAPTER 14.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. 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 . where . The q term in this equation can be removed by transforming the equation to a cubic in y.6. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.6 The cubic potential.79) . (14.y3) = 0. (11. which determine the imaginary part of the energy. Example 14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q).y2)(y . i. 14.a . (14.77) Determine (a) the turning points. where 1 2 • 1 2 y = Q.133)). Eq. 14. Fig.<?2i<?3 for E > 0 given by E — V = 0. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. q = y+ a .e.
6 = 0 i m p l i e s E m a x of E q .c o s O + cos(6» . B y r d a n d M . D . p .= s i n t 3 v 3 (14.78)) c o s 30 •• 4P3 ( ' \ aeb2} . 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 . q3 ~  . D w i g h t [81]. g 2 ~y{3}>0. H e r e cos(<9 ± 120°) = .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 . 92.cos 0 + cos(6> + 120°)] = a2 92 . F .gi = 1. f o r m u l a s 236. we set 0 = 60° — e.< 1. (14. . _ \ / 3 — tan(9 2 V92 . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . (14.ui=K(k). (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 — 0.120°)] = . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . M i l n e . V3 + t a n 0 2 . V cose' + sin 8 (14.1 / 2 .k 2 2tan# \/3 + tan! .T h o m s o n [196]. 78: sin 120° = v ^ A c o s 120° = . 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.81) m a 91 . tan60°=%/3.\tan0^ v/3. (14. 37.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. § T o h e l p we cite H.2 c o s ( 0 + 1 2 0 ° ) ] . i.3/2.04.9 3 — = = 2 " 3 " [ . a n d t h e a n g l e 0 = 0° t o E = £ m a x . q3 = a2[\ . F r i e d m a n [40].c o s ( 0 + 12O°) + c o s ( 0 .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 .9 2 [ .6. ' P .2 c o s 0 ] . p .93 2a2 [ . we h a v e qx = . 212.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . gi ~ y {V3e} < 0.0^6O°.a 2 [ l .84) 92 .g i ) ( g 2 . (l_fc' 2 a + fe'4)l/4. T h e r e f o r e E = 0 c o r r e s p o n d s t o fc2 = 1.q)(q .2 c o s ( 0 . 93.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 .9 l ) f f / du s n 2 u en2 d n 2 u . 14. p .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.e.' : ± 6 0 ° w h e n E = 0. 92 = a2[l . 91 < 93 < 9 < 92(14. 0<6O°.{ V 3 e } > 0.08.93). 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 . 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. M .2/3 i n t o q = y + a 2 / 3 .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. e > 0. p .g i < 1.80) I n s e r t i n g t h e r o o t s yi. 80 a n d 361. ui=sn 1 \l— — = l. (14.14. B .79) a n d t h e n s e t t i n g m o = 1.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.
the integrand of the integral across the well is effectively the inverse of this momentum. formula 233.qz) (14.^ V3 / V V3 y a / c o s e J. sinV = q3 qi ~ = 1.91) a2[cos0^] y/3tanfl ? L . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . 2 2 . (14.304 CHAPTER 14. Thus Iwen=lgF(ip. . 1+72 = and J o s 0 + ^ = (1 . E(u) the incomplete elliptic integral of the second kind. as in (14. V?2 .= —^ = k a2[cos0+^] v ^ + tanfl V=".87) Inserting G(fc) together with the expressions for the prefactors into (14.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . D.85).00. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . E(u = K(k)) = E(k).?)(?3 . with K(k = 1) = oo.fc'2 + fe'4)"1/4. A:') = # ( * ' ) • (14./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —.^ + ^ ) (l + fc'2)2 . E(u = 0) = 0. 91 < q < q3 < 92 bJc. Whereas the integrand of the above barrier integral is effectively a momentum.) . Next we evaluate the required integral across the well from 91 to 53 at energy E.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.92) ''P. 72. p. G(fc) =86aS (ITS£ li[fe.91 F(TT/2.84) we obtain Carrier = 2a 2 sin6> 2 b( . F.sm t %/3 £\. We can evaluate this integral again with the use of Tables of Integrals. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. (14. (24. The WKB Method elliptic integral of the first kind.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . 0 (14.S .k). g= 2 .2(fc2 . Byrd and M.86) 15 tan. Eq. . Friedman [40].~ y ^ ) a< [ COS 5 + .32)). V 93 . (14.91 .
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. S. so that even in the case of the ground state the zero point energy will contribute to the prefactor.113. we obtain E a6b2k'4(l + k'2). formulas 8.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . (14. M.93) where w is the harmonic oscillator frequency in the well.114.** However.2. Ryzhik [122]. (14.l. Gradshteyn and I.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. we obtain > . With q = 1/z the integral (14.63b). Thus one obtains from (14. (14. to represent a physical decay rate.96) Expanding the coefficients of the elliptic integrals in Eq. 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.4 2n + 1 fc' = 4 — ^ — . pp.88) in rising powers of fc' . 32 (14. (14.14. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. a°b n = 0. and hence at E = 0.3..3 and 8.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . (14. . 905. 906. we obtain 15 . 2a 5 b K(k) + 2 E(k).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 (14.85) (14. .80). To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.
178)) call the Furry factor set equal to one. 2n + 1 — / 26a5b \ .102) .fc'2 + fc'4)1/4 * ± t ^ . (0) i hw En = E\> . (14.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq./^~^l^ab ^aub . (20. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.96) for the imaginary part of the energy as in the Breit.— (14. _ T T • (14. +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . n ~ V27T ..7 .Wigner formula (10.„.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results. this has not yet been done. 15 a .J + 0(VoJb).. the result agrees with the ground state path integral result (24. 7 ~ —. i.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . so that we obtain 4 . It follows that (for one complete orbit back to the original turning point) . V (14.e. We observe that with \j2je as 1. 8 Bfc . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .73).e.ba / 25a5b\n+2 ±l_l eT5 a f >. .e x p [ .101) (14.j . _ exp[2/barrier] >.2 7 b a r r i e r ] Z Z7T where (cf. i.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.. The WKB Method For k' —• 0 the first and the third terms dominate. ..100) 'barrier = ~0°b In [ ^ — j .> tf = = 2/wfiii = ±tv8a°b——P. also Eq. (18.306 CHAPTER 14.
307 . K. K. indications from the nonabelian generalization of quantized Maxwell theory. now known as quantum chromodynamics. Eichten. the Coulomb potential. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. i. This potential became widely popular in the spectroscopy of elementary particles. B. at least indirect. [231]. which permits only bound states. D.e. [233] and H. Gottfried. The study of this potential. Quigg and J. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. Quigg and J. Kinoshita.M. the classification of the various states of nucleons and mesons and other particles. Rosner [230]. Yan [82]. describing free fall under gravity. that this is indeed the case.1 Introductory Remarks The particular linear potential we considered in Chapter 13. 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. Rosner [269]. Lane and T. T. led to a spectrum which contains only scattering states.g. C. C. L. along with inclusion of angular momentum and spin effects and their interactions. There are numerous. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. L. after the realization that quarks as their constituents might not exist as free particles and.Chapter 15 Power Potentials 15. Thacker. See particularly E.* 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. in fact. [232].
3) with the boundary conditions''' u(0) = 0.V 7 = const.r acting between two particles is constant and directed towards the origin (of the relative coordinate). In order to have a clear starting point.4) Instead of considering the linear potential. we return to the Schrodinger equation in three dimensions.£ ] * ( r ) = 0.e. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. and u'(0) > [u(r)/r]0.e. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). (15.4>)ru{r) = Ylm(9. h2 2n V 2 * ( r ) + [V(r) . This force maintains this value irrespective of how far apart the particles are separated. (15.2) tt(r) = Ylm{9. and so u'(r) . i.6) = (I + l)u(r)/r = R ~ rl.1.1) i. We leave consideration of the logarithmic potential to some remarks and Example 15. m\m<i M= : . A > 0.u ~ r i + 1 (/ + l)R(r).5) u'(0) u[r) r=0 = R(0). and the normalization [ dr\y(r)\2 = 1. </>)R(r). 15. where \x is the reduced mass of the two particles of masses mi.2 The force The Power Potential F = . ~ (I + l)rl (15.m2. (15. v > 1.308 CHAPTER 15. For a central potential V(r) we write (15. Power Potentials general power potentials. mi + m2 and r is the relative coordinate. These particles therefore cannot be separated by any finite amount of energy. R(r) = ^ . f dr{u(r)}'2 1.
4) in the threedimensional case.9) Threedimensional case: We transcribe the above considerations of the onedimensional case into that of the threedimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!).. (15. We therefore proceed as follows.2.. N = 0.8a) ^ JO = (2Ar + l ) £ . which has been reduced to an effective onedimensional problem in the polar coordinate r. and there identify the Swave states with those states of the onedimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq.63a). This means that we transcribe Eq. (15. . (14."MA. 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) we can write the quantization condition (14.V(r)) = (2nl) +1 N.. 3 .9) into the threedimensional case in the following form: / Jo C dry/2n(E .7) k(x) = h/y/2m0(EV(x)). d5. or f l H " ^ ) ! .2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. Then k(x) = k(x). (15. We also assume that V'(x) > 0. x > 0. .2. (15. (15. odd integer ^=(nll)vh. In order to to obtain the . (15. is to be interpreted as the principal quantum number in three dimensions.8b) This condition implies that we have only one pair of turning points. .15.63b) for the onedimensional Schrodinger equation.63b): 2 and V(0) = 0 (15.10) Here rc is the turning point and n — 1. so that with Eq.1. and the turning points are at a — — b := — xc. We now want to find a quantization condition for the threedimensional problem.
(15.12) and (15.13) dt . (15. (1516) (15.17) The integral on the right is the integral representation of the beta function B(x.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .14) Prom Eqs. E A fv. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.\rvc = 0 . c (nj\irh. s.11) rc = (15.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. Power Potentials eigenvalues E —• En for the radial potential (15. dr = .310 CHAPTER 15. Now.'(1*)1/2= J1F\lt)y1dt.y) defined by (xl)\(yl)\ y ^ (x + y . _ —vr" *dr. .15) I ' = ^ 1 (( 1) B{x. (15.13) we obtain _ X_ „ _ so that A£_ (15.
(15.15.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) .* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.1. 15.1)!. Weinstein [281]. and n — — ]7r/i. One can convince oneself now. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. (12. for instance simply graphically as indicated in Fig. $T(z) = {z.1 Approach to square well with v —> oo. The case v = 4 is that of the pure anharmonic oscillator also discussed by M.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. . i r ( i ) = r ( l + i ) = ( i ) ! . T(z + 1) = z\. ^ '"Compare with Eq.55). 15.
(15. 15.312 CHAPTER 15. the case of the linear potential.> r ( 2 ) A V 2 in \)*\W \it^T& (15.41) for the .oj = ^/2X/JX. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. we obtain En = (n . (6. Physically it does not make sense for the probability amplitude to have a discontinuity there. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. since these are the ones we are interested in (cf. (6.21) (4nl).23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. discussion at the beginning of this section).1 + \)hx> = {An • l)£fiw. The potential V(x) = X\x\ (15.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. But for our present purposes this applies only in as far as the solutions (j)(x) which are exponentially decreasing at infinity vanish at x = 0.25) "Comparison with Eq. Power Potentials In the case of v = 2.mo = \. eigenvalues (JV + h)tkj — (2n .61) h2 = I. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .g — —2.2. > and with Eq. 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). (13. x > 0.oj2/2. For the following we set in Eq.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). the harmonic oscillator. In the case v = 1.2) implies A s fj.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones. we obtain E„.
27) i.x)\x=0 = 0.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . i.59) applies. \(s?» (15.e. (15.60) applies. Ai(s) "v^ G ' cos s 3/2 _ t (15. (15. Eq.29) . In the domain —a < x < + a . and for s —> oo the trigonometric behaviour of Eq.2 Behaviour of an odd wave function at the origin.x). E > V or s > 0. (14.e.e. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. i. i.26) For z = s := E — x —> —oo. s = E. i. 15. (13. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E .28) In particular at x — 0. (13. this wave function has the required exponentially decreasing behaviour at infinity.e.15. Ai(s) 1 2v^(s)V4 exp x —> +oo.e. dz2 According to Eqs.56) and (14.
e. i. Table 2. It is possible.3.= ) = «.08181 5.28) is really only valid for s large and s — oo. but for large > values of E Eq. Quigg and J. The expression (15.03914 11.51716 6.00767 11. Power Potentials Table 15.04017 11. with permission from Elsevier.33811 4..32025 4.e... = im — (7r/4). It is interesting to note that in the dominant approximation both methods agree.2. (2/3)E3/2 or 3 / 1 2/3 E — En — .1 demonstrates the quality of the WKB approximation by comparison with the exact values. L.78445 7.82878 1 2 3 4 5 6 7 8 9 10 2.30) ! ^ / 2 .00852 11.08795 5. to obtain the zeros of the Airy function numerically.^ = !(2nl). [231]..94249 9. . (15.e.En from Ai(£„) = 0 2.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. Co.93528 12. Table 15. the eigenvalues En are determined by the zeros of the Airy function.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .29) implies that cos (  ^ .02265 10.31) (an odd function has an odd number of zeros!).94413 9. *This Table is reprinted from C.82814 i.02137 10.52056 6.78671 7. 201. p. Quantum Mechanics with Applications to Quarkonium. i.314 CHAPTER 15. (15. (15. n = l.93602 12. of course. Rosner. so that n — 1/4 ~ n. copyright of NorthHolland Publ.
• •) by a mean value.e.g. We saw previously (see for instance Eq. van Royen and V. e. The decay widths T.34a) **The most frequently quoted reference for this result is R. See also J. (14. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. e. the charmanticharm meson ^ . ee.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i.15. In order to be comparable with experimental data these investigations.g. r <53) 1 3  dr cos 2 () (15. *Note that this makes N a W K B normalization constant.3 The ThreeDimensional Wave Function 315 15. Jackson [140]. can be shown to be proportional to the modulussquared of the particle wave function at the origin. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. of course. D. cos2()^— / 1 fr° 2 > w . which are inversely proportional to the lifetimes. i.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. The resulting masses agree in most cases very well with those extracted from experimental observations. We shall encounter WKB normalization constants at numerous points in later chapters. P. F.e. i.e.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. have only a finite lifetime and decay into other particles such as electronpositron pairs. Weisskopf [239].** r ( # ^ e e ) o c *(0) 2 . in Chapters 24 and 26. . In general these quarkantiquark pairs.
Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( . (15.10) with respect to n implies (which is permissible for small separations of neighbouring levels) 2/i ldEn rrc dr ^/2ii{En V) 2 dn J0 = irk..32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.34b) VMEvy Variation or differentiation of Eq.1 )n .35) u T (•T=2TT/OJ dt cos 2 cot 2' (15. (15.2) and (15. i.rcu w X{r)dr = Kh . (15. & (15.316 In accordance with t h e relation CHAPTER 15.38) N : sin V ^ Jr dr' 7T (15. (15. (15.l N n .4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .37) (15. A = (15. From Eq. Power Potentials this averaging implies a numerical factor like 1/2..36) \i dEn irh2 dn According to Eqs..39) in / .40) VWY . so t h a t approximately l N2 / Jo k(r)dr . dE a—— / dn J0 so t h a t with Eq.2.\ \ * n = l.e.35) N2 .
Hite and H. G. K. § S. E. . J. S.37) we obtain: 1 (2M^)1/2 47r ft.* Using the result (15. 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. the expression (15.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e.l ) n . . W. S.g. 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).38) implies 317 (15. for instance. for various potentials: (a) For a confinement potential of power v. MiillerKirsten [28]. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. J.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. Bose. . W. The case of arbitrary positive power potentials as above has been considered by R. 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^"^. Hence Eq. K.41) I*(0)2=_L^)^. and hence with Eq.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. MiillerKirsten [29]. MiillerKirsten [147]. J. . Bose and H.1 ^ ( 2 ^ „ ) 1 / 4 . (15. 3 .44) (15. W. we have ^'(0) = ( .23) for the energy levels of the linear potential.15.2. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. Kaushal and H. (15.
133).. (11.318 (c) and for the Coulomb potential^ CHAPTER 15. Power Potentials *(0) 2 oc . ip = exp[(z — c)/2].L 2 + U{z)]<f>.I l n f ^ p . *=^f. as well as Expanding U(z) about its extremum at ZQ = —1/2 and using the perturbation method of Dingle and Muller derive the expansions N)2 = ^^ 3 . q = 2n + l.46) and the classical Kepler period of Eq.1. .g ^£+(<*Pl»riy = 0. (14. 7 With the substitutions r = exp(z — c). Bose and H. M.46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14. . MiillerKirsten [29].2. n = 0. E' = E + gln(r0). {S = ^ > 0. = (I + 1/2) 2 . K. (15.. = 1(1 + 1).j . and there Eq. U(z) =  dz zf3e2a/?e2*. . J. Landau and E.5.— oo < z < oo.L2 h4 = 4/3exp[(2a — /3)/f3]. Eq = i q . We observed there in the case of the Coulomb potential that the period is proportional to n 3 .1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).75). D. S. Example 15. Lifshitz [157].e. W. in agreement with Eq. i. Solution: Details can be found in the literature.c = —a//3..^ ( 3 ^ + l)55p^«(3^l) + . rewrite the equation as ^ z + [ .. (15.^ +•••." 1 11 L .
In reality. it will always notice the latter's presence. A very similar potential is the Yukawa potential V{r) = g2 — .Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . The exponential insures a rapid falloff of the potential at large distances and for this reason the parameter ro can be looked at as a measure of the range of the potential. This implies. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. 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.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. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. and in fact. it is.Chapter 16 Screened Coulomb Potentials 16. so that like charges are repelled and unlike charges are attracted by each other. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. Historically this potential arose with the realization 319 pVr . The result is a screening of the Coulomb potential which thus attributes it a finite range ro. This effect sounds unphysical.
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.ii k . Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. Heisenberg [133]. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. one frequently resorts in calculations to the expansion of the potential in rising powers of r.x j 36 x 4 (2TT) k + p? 4 2 (2TT) J d x e^xl 4TTX ' We observe that the Fourier transform of the propagator is effectively the Yukawa potential. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). (16.1) We encountered Green's functions earlier.3) where for the real potentials we consider here. all coefficients Mj are real and independent of the energy E = k2. Since. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. 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 . A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. and that.320 CHAPTER 16. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. in fact the parameter [i represents the mass of such a spinless meson. i. (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.2 ) or (p = hk) 1 /" . however. the mathematical expression with an exponential is not so easy to handle analytically.e. . and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. from the relation Pi + P 2 + M2c2 = 0.
mo being the reduced mass of the system. Mandelstam [185].1 in the above. .§ 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.3) in the radial Schrodinger equation for the partial wave ip(l. S.r)=0.16. See also the other references below.e.Q. £+**PVV t/. (16.e. Frautschi [97] and E.k. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. 2 . Froissart [223]. Lovelace and D. (16. i.k.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. C.3) have been considered by various authors. 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. i. Kinoshita [21]. Cheng and T. bound state eigenenergies and the 5matrix. .1. Bethe and T.r). l+n+l = ^P. if>oo. They are usually written I = an(E).1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. J. to be of the order of 1/k. physical) values of angular momentum as functions of energy E. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. Wu [47]. . As discussed in detail by Bethe and Kinoshita^ one can start by arguing that a countably infinite number of Regge poles may be defined in the region of large negative energies E = k2 of the radial Schrodinger equation by requiring I + n + 1 for n = 0. Masson [180]. 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. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. A. T.5) where K — ik and A n is an expansion in descending powers of K. "These were first considered by C. f H. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. however. i. (16. . (16.e. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.4) where E = k2 and h = c = 1 = 2m. Squires [258]. only for the cases of n = 0. Potentials expandable as in Eq. Omnes and M. S S.{l.e.
The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. W. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. A. Regge [31] and E.k. 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. W.15) and (16. J. (16. < const. M. Bottino. for all n. .z) = ez2/2zl+1X(l. Under these conditions the S'matrix is meromorphic (i. J.k. Proceeding now as in the case of the Coulomb potential we change the variable of Eq.)/j.19).6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. 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.9) the cut starts at E = k2 = 0. Squires [258]. *See in particular A. one can assume an expansion of the potential V(r) in ascending powers of r. 'In the case of the Coulomb potential (cf. (16.z) Then x is a solution of the equation VaX= (16. J. o rV(r) I dpp\V{peie)\ regular at r = 0.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. < oo for all 0 < vr/2. H. In the following we follow mainly the last two of these references. 16. Schilcher [203]. J. tf H . Screened Coulomb Potentials as a function of the energy. Miiller and K. starting with the power r _ 1 of the Coulomb potential.322 CHAPTER 16. in the present case at fc2 = 0 or k2 — Mi = 0.** The energy is later obtained by reversion of the resulting series.ndp. Longoni and T. as may be seen from Eqs. W.4) to z = —IKv and set TP(l.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.e. Muller [204]. Section 11. Muller [201] and [202].
a + j)Smi(a.j l) + (a + j.5)) a = + 323 .9) we obtain m l)$(a . Sharma and H. (16. a + 1) = a — b+ 1.11b) ^ The associated boundary conditions are: So(a.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.b. (a.0) = 1.10) zm$(a)= ^ j=m Sm(a. j)Smi(a.z) = Ha). all other So(a. a) —b — 2a. . and all Sm(a.j +(a + j + l. (a.j)$(a + j).a + + l).z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.7) is seen to be of order 1/K.j) may be computed from a recurrence relation which follows from the coefficients (16. (16.9) (16.a+ l)$(a + 1) + (a.i 0) = 0. Eq.1). K.11a) The coefficients Sm(a. where $(o. W. We also observe that the ansatz (16. The function <&(a.a)${a) + {a. anharmonic and cosine potentials have been derived in L.b. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a). J. as in the case of the Coulomb problem. MiillerKirsten [249].j) for \j\ > m are zero. (16.2 Regge Trajectories where d2 Va = z^ and (cf.a — l) — a — l.awhere (a. which means that the hypergeometric series breaks off after a finite number of terms. (16.16.5) is equivalent to a = —n. By a repeated application of the recurrence relation (16.* Recurrence relations for coefficients of perturbation expansions for Yukawa. {0) X = Ha.j) (16. so that to leading order we have VaX{0) = 0.a + j)Smi(a.10): Sm(a.j) = (a + j l.
<&(a + n) on the right hand side of Eq. 1 .12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higherorder perturbation terms.1] 2 . and hence Va$(a + n) = ra$(a + n). + (16. . [a.324 CHAPTER 16. when n = 0.' — . We first cite the final result .[a>a + J]i+i*(a + 3). Evaluating the first few terms of the expansion (16.[ a + l. (16.An{K). (16. Va+n = Va . i R »] = 2K[a.aj 2 (2K)A _ [a. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .e. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. i.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.n.13) one obtains the quantity An(K) and hence with Eq. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. Any term /j. 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). 0 < \j\ < i. [ a . (16. for the coefficients M^> of the expansion Details can be found in the references cited above.12a) where [a. (16.aj 2 \ [a. [a.o]i = M 0 .5) the Regge trajectories I = ln(K).a + l ] 2 [ . (16. the latter can be written . of course. oo _ .a + j]i+l = MiSi(a.a]l^a) + ^2(2K)i+1 Y.a\4.7).13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. a . This follows from the fact that D a $(a) = 0.j).l. #(a + ra) fjL$(a + n) V.
l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . P. (16.§ This is an important observation which indicates the equivalence of the methods. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .^ 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. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.1. 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.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n .1 by evaluating the first two terms.2. of course. . we obtain the expansion for the Regge poles.5). (16. With numerical methods one can achieve more. G.16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . However.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . including an exploration of the domain of small energies E. Tate [6]. Thus the expected appearance is that shown in Fig. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. 16.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n .e.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .^ p [ 3 M 4 ( n . i. Inserting this expansion into Eq.1) + 3M3M02 + M22n(n + 1) (16. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). .15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16. Obviously the expansion — which. Ahmadzadeh. Burke and C.
For a resonance with a long lifetime. .11b) we obtain Si(a.a) = b .326 CHAPTER 16. (16. satisfies the relation ajAO ~ h. where the imaginary part of I is very small." Re I 1 0 1 2 Fig. Frautschi [97].O) = Other terms vanish since So (a.^ . a] 3 = M 2 S 2 (a. However.13). From Eq. ai is small and A6 is large. (16. 0) and Prom Eq.1 Typical Regge trajectory for a strongly attractive Yukawa potential. In fact. Similarly the conjugate variable to angular momentum is angle.a)So(a. through which the particle orbits during the course of the resonance.0) = (a. [a.0). 16. and the lifetime At of the resonance satisfies the relation TAt ~ h.1). this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig.2 a = In See S. p. For a bound state a j = 0 and the orbit becomes permanent. 0).12b) we obtain [a. (a. . one can expect a resonance with the lifetime determined by this imaginary part. just as a decay width T represents the width of the resonance in energy. 16.—" + In = — . C.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. 114. Solution: We evaluate the first three terms of expansion (16. The conjugate variable to energy is time. a] 2 = M i Si (a. and the angle A#. so the imaginary part ai of I — an represents its width in angular momentum. Hence Si (o. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. ±1) = 0. Example 16.
0) + (a + 1.a + l)S0(a. (a.13) we obtain 0 (Mo . A" An An A" = = = (a. a)Si (a. We obtain the quantities S i ( a .2 Regge Trajectories 327 Analogously we evaluate S2(a.al)S0(a.2 a ) .65) 2m dz V'A + 2Bz + Cz2 • Z— 0.n ( n + l Inserting these expressions into expansion (16.4) 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. i ) again from Eq.0) (a .1 ) + (6 . (16.6 + l)a + A. a)Si (a.65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 .0)+ 0 + 0 = (a1). (a.11b). .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .16. a)Si (a.00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I. . 0) + a S i ( a .(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . 1) Hence 52(a.0) + 0 + 0 = ( 6 . Solution: Ignoring higher order terms in r.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .2o)Si(a.b)(a .1. using the Cauchy formula (6. we have to evaluate (cf. 1) (a . 1). Thus (with terms which are 0): Si(a. . One sets z = 1/r so that.6)Si(a. (n+l)U+^pJ .0) = (a .1) + (a .1) Si(a.6 + 1 ) .2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6. Sec.a)5o(a.0) Si(a.0) + 0 + 0 = ( a . 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 . 14.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .
±2HK in agreement with our expressions above. *J.3 The 5Matrix It is clear that if we now work through the usual procedure for the derivation of the S'matrix we pick up a logarithmic phase as in the case of the Coulomb potential.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Screened Coulomb Potentials Mo . however. Example 16. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. 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 . This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 ..328 Thus here dr K' CHAPTER 16. n = 0. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.5) together with the expansion (16. I.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ .l.* 16. we can use Eq. The turning points in this case are at r = 0 and r = oo. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. . (16.t First.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. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. Here we do not perform this procedure.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A.2. Singh [252].(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. (16.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . i.e. Boukema [34] and [35]. by expanding the right hand side in rising powers of v and evaluating the individual integrals.15). Solution: For details of the solution we refer to papers of Boukema. tSee Chapter 11 and V.
14). such as. we can now write down the S'matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e r(/ + i^#>)e ' (16 18) ' We observe that the poles of the S'matrix are given by l + l + ^p = n. Miiller [204].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\. n = 0.l (two additional terms are given in the literature). i. H.18) of the Smatrix now to explore further aspects.17) and is therefore real for real values of the potential coefficients Mj. ...16. Miiller [211]. 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 .17). the behaviour of the scattering phase in the domain of high energies. W. This has been done§ and one obtains .g. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16.. One can use the explicit expression (16.19) we obtain the energy. § See H.19) +2M 3 M 0 (3n 2 + 3 n . W.* 16. for instance.l. the following *See e. which are precisely the expressions yielding the Regge trajectories of above.l)n(n + l)(n + 2) + 2M2M02 . J.2. We observe also that the S'matrix is unitary as a consequence of the result (16.e. Paralleling the case of the Coulomb potential in Chapter 11. J. Now reversing the expansion (16.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.0[(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\.
i. C. VahediFaridi [212] contain as special case the expansions of the first pair of authors.< 3 M 4 M 0 ( n . the transformation was introduced in the form given below by G. p.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . in which the leading term is the usual expression of the Balmer formula for the Coulomb potential: K2 = Mi+ 1 ° l4n(n + l ) ^ ( / + n+l)2 4(7 + n + l ) 2 Mio +4(2n + l ) ^ ( / + n + l ) 3 4(Z + n + 1)4 ( l) ^ — . Analogous expansions for specific energydependent Yukawa potentials have been investigated by A.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. B.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.11) + 2M22M03(9n2 + 9n .e.1) 10M 3 M 2 M 0 2 n(n + l)(3n 2 + 3n + 2) + 20M23n3(n + l ) 3 30M 4 M 2 M 0 (n .3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). E. Mendelsohn [135] and E. in the following expression with an "See in particular G. p. Sommerfeld [256]. Iafrate and L.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n .2)(n . Large coupling expansions derived in H. J. Frautschi [97].l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Warburton [279].1) + 2M2M. J. 'According to S.^ 16. MiiilerKirsten and N.* We recall from Chapter 11 the definition of the scattering amplitude F(9. Watson [280] in 1918 and later resurrected by A.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . Screened Coulomb Potentials expansion.330 CHAPTER 16. A. .f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n .l)n(n + l)(n + 2) . N. Zauderer [286]. W. 282. 107.
25) Setting Note the minus sign in the argument of the Legendre function. For this purpose we consider the following integral taken along the contour C shown in Fig. (11.k)~l] = ~ k" PiSiW sin 5i(k) and S(l.184) to (11.. n = 0.2 The integration contour C.22) = \F(9.. Actually 5j lrr/2 as in Eqs.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. Eq.k) — e 2i5[(k) ~" p—iirl (16.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. (11. (cos 9). The scattering amplitude possesses the partial wave expansion F(0.26. . c = 1. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.k) 0ikr (16.24) Si(k) being the phase shift. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).187).k)\2. ~:' sin nl (16.iJ 11 —[S(l.1..fe)P ..k) = ^T(2l + !)/(*. I = n + x.k) = 2zfc L " v "'' v .23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.''' The idea is now to consider each term of the expansion (16. z=o in which (cf.187)) (16. 16. 16.2.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.kmcose). \x\ <C l .23) f(l.
as indicated in Fig. for instance. Thus the direct reaction may. 16.—. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.k) Zi Jc TT{L . be elastic scattering of a off b.26) P((+COS0) = 2 ( 2 n + l)/(n.332 CHAPTER 16. a + d — • c + b.28) .3. In order to see the relevance of Regge trajectories in a reaction of particles. a + c —> b + d. The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d.n) oo {l)1 Pt{cosO) v ' (16. 16. n=0 which is the usual partial wave expansion. d(q') (16. u={pq')\ (16.3 Reaction channels and their respective Mandelstam variables.27) a(p) © b(q) Fig. 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'). t = (qq')2.3. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. 16. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. For the momenta indicated we set (with metric +. in which case c = a and d = b. we obtain m® = ± I dl^±^f(l.—.—) s = (p + q)2.fc)P„(cos0).
N. one can write the amplitude as depending only on two. A(s.g. Treiman [25].g. u.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. One now makes two hypotheses. F. Goldberger. t. § See G.et) (16..The variables s. Chew [48].t.32) t = 4(q? + ml). N.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.t.t. T See e. if s describes the square of the total energy in the s channel. Thus.t) = VtFt(t. In a Regge theory on the other hand.(<fe)^(cos 6t).t.u) (if any of the external particles has nonzero spin. there will be several such amplitudes which together describe all three processes). In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).u are not independent. e. which is an analytic function of the variables s. u = . 16. L. A(s. one then has A(s. . Blankenbecler. t = 2q£(i .u are known as Mandelstam variables. the momentum k there would correspond to qs) s = ~4(ql + ml). if we had been considering the s reaction originally. (16. the variables t and u would describe momentum transfers in the crossed channels.16. S. Ft = £ ( 2 J + l)F. the only kinematics we require in the following is given by the relations (e. Khuri and S. except for poles and cuts which characterize the reactions in the three channels. (a) Mandelstam's hypothesis: All three processes (described by the s. B. (16.9s)\2.c o s 0 s ) .t).g.4.31) i l Taking all particles to have the same mass mo. M.§ Since the three variables s. s = 2qf(lcos0t)(16. the dynamics is described by the exchange of quantum numbers. Mandelstam [184] and R. For a Lorentzinvariant normalization of the initial and final states.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. Symbolically we then have the situation shown in Fig.t) = y^Fs(s. etc.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s.0s). the quantities carrying these quantum numbers are the Regge trajectories.2 ^ ( 1 + cos6 s ).
It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. Lehmann [162]. 16. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. However.)e~ •fj. 16. .334 Yukawa CHAPTER 16.5 The Lehmann ellipse.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. the Legendre polynomial Pj(cos 9) does not possess a cut.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. in the domain of schannel physical values of the kinematical of the ichannel). But > > wave expansion X)(2Z + l)F. 16. the partial wave expansion is not valid in the physical region of the (i.4 Yukawa versus Regge theory. But for integral values of I.5 for Yukawa potentials V{r) u: lm cose dr a(/j. Screened Coulomb Potentials Regge oc(t) Fig. For s — oo : cos 9t = 1 + s/2q^ — oo. So what can one do? 1 H.e.
Moreover. 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.16.^ B 1 A Fig. „ i. sin nl [16. N F(6.33) .6. to use the SommerfeldWatson transform. contour C / / D ^ ^ . and there is only a finite number N + 1 in the domain of 9W = .e. 16. i. 16. all poles of Ql > 0. Thus consider now the same integral but taken along the closed contour C shown in Fig. 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.t). Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. B Iml 1 \ i All \ r\ 0 / closed .26) but for a different choice of the contour.6 The integration contour C".5 The SommerfeldWatson Transform 335 T h e answer is.1 / 2 . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l.e.
In particular. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. as we discussed briefly in the above.35) This result is valid for s — oo and t negative (cf. Regge trajectories have. this behaviour has been confirmed in high energy hadronic reactions. Bose and H.e. Predazzi and T. It follows that under these conditions. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Eq. at high schannel energies. . the amplitude can be represented as a sum over ^channel Regge poles. for high energies in • > the schannel).e. Foissart [223].t)&B(t)sa°V. (16. if ao(t) is the Regge pole with largest real part. p. Omnes and M. i. 42. 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. W. for plots and discussion see R. of course. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. K. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. J. 16. Apart from more refined details. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. Regge [229].e. Limic [177] and E. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i.32)) and finite. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. i.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. MiillerKirsten [29].336 CHAPTER 16. we have (16. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo.
a composite particle being one with structure.g.6 Concluding Remarks 337 important parameter in string theory. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. Related aspects are discussed in Sec. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. 20. and also as socalled "fixed poles" . MiillerKirsten [209]. The way to do this is similar to calculations in Chapters 18 and 20. and is composed of quarks. . For an overview see e.** Whereas the former are related to absorptive properties of a reaction. the latter are related to the distinction between "' elementary and "composite particles".16. The above treatment of screened Coulomb potentials is incomplete. Singularities appear also in the form of cuts in the plane of complex angular momentum.1. J. H. called "Regge cuts". like a meson or a baryon. W.
This page is intentionally left blank .
19 and p. however.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered.g. since many analogous relations hold. motivated mainly by the regularities of the crystal structure of matter.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. 2K or AK. the most immediate candidate is a trigonometric form like that of the the cosine function. p. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. are not much harder to handle than trigonometric functions. The Schrodinger equation with this potential. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. 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 reason for these difficulties is. Arscott [11]. See F.* These elliptic equations involve Jacobian elliptic functions like sn(a.0 < k2 < 1. and are themselves periodic with period e.) which are functions that interpolate between the n. example 14. 339 . called elliptic modulus. K being the complete elliptic integral of the first kind. 25. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. depending on the parameter k. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. M. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). In essence these periodic Jacobian elliptic functions.Chapter 17 Periodic Potentials 17.
For large coupling. Miiller. The Mathieu equation. We might mention already here the much less familiar but more general Lame equation. Like the trigonometric functions sine. W. which does not possess any polynomial solutions. J. The first of these conditions is already tMathieu equation: R. The most important formulas for our purposes here are collected in Appendix A. 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.1.340 CHAPTER 17. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. each with its own characteristic eigenvalue. the Lame equation possesses 2 n + l polynomial solutions.. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. In fact. This is a significant difference compared with the Mathieu equation. [205].7. J.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. those of integral order. Periodic Potentials vide periodic potentials like trigonometric functions. W. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .. ''The reader who encounters these Jacobian elliptic functions here for the first time.2 . B. 17. i. Lame equation: H. which one looks up in books like that of MilneThomson [196] or Tables when required. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. The cosine potential of the Mathieu equation can therefore more precisely be described as Mathieu or trigonometric potential and the potential sn 2 (x) of the Lame equation as Lame or elliptic potential.2h2 cos 2z]y = 0. Both of these domains are important for a host of other considerations. Muller [73]. or Sw&ve Schrodinger equation with elliptic potential.1. See also the discussion in Sec. 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. c n z and d n z are handled with analogous formulas.1) where h2 is a parameter and — IT < z < IT. (17. 8. need not be afraid of them. which is conveniently written ^ + [\2K2sn2z]y = 0. the three Jacobian elliptic functions s n z . and to relate these to the weak coupling bands or regions of stability. (17.e. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. 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. cosine and tangent. . Dingle and H.4.
and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). the semiaxis A > 0. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre.e. a bounded function. we see that in a plot of h2 versus A. i. h2 = 0 belongs to the domain of stability. i. Salem and T. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. 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.2.1 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. as in Fig. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. the case of h2 small.1. i.e. Although the mathematics literature on the subject uses the physical concepts of stability and instability. In classical mechanics stability is treated for instance in connection with planetary motion.^ 17.e.e. i. Thus the solution of the appropriate Newton equation is essentially a periodic function like cos 8 (in the case of planetary motion this yields the polar equation of an ellipse).2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. . we compare the For interesting related discussions see M. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.17. Adding the negative term — 2h2 cos 2z to A. y" + \y = 0(h2). Looking at the Mathieu equation for h? —> 0. 17. Vachaspati [242]. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. 17. these concepts are rarely explained as such there. since the solution there is of the form cos vAz.
d2y(z + 7r + [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).3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. d(z + TT) cos 2z We see that there are solutions which are proportional. Periodic Potentials equations ^ p + [A . 2/(0)' yfr) . (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). a = const. i. y~(z) = 6sin vXz.e. i. Setting z = 0. we obtain y(vr) y(2?r) a = — — a = —^^r. ° ^ [1 + 0(h2)}. sin V Az or linear combinations. — • .342 CHAPTER 17. (17. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. y(2vr) 2 a = —7—^ = 1. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). y+{z) . (17.e.2h2 cos 2z]y(z) = 0.2/i2 cos 2(2 + vr)]y(z + vr) = 0.B and a. TT.
7 y\z) = ay'+(z) + (3y'_{z).W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. with constants a and /3.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. 2/_(0) = 0. y{z) = aty+(z) + 0y(z).y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\. This means. yV(0) = 0. (17. the determinant of the matrix must vanish. 2/'(7r) = a ^ W + ^ . \y+>y\ = y+{ir)y'{ir) . This expression is equal to the coefficient of a2 in the quadratic equation for a.4). [y+(v) . The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4.e.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above.17.aa][y'_{n) . and by'+(?r) = —ay/Xy{n).y+(7r)j/_(7r)} = 0.a2b2\ a± ~ 2a6\/A .y+(7r)y_(7r) = 0 or abVXa2 . i. y'(z + n) = a[ay'+(z) + py'_(z)}. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. Using Eq. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) . Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. we choose these with the following set of boundary conditions: y+(0) = a.oWX] . (17.3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. We can write therefore. y'_(Q) = bV\ with Wronskian W[y+.{ay'_{ir) + bV\y+(ir)}a Now from Eq.
.5) a The parameter i/ is known as Floquet exponent. we have <7± — CHAPTER 17. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . (17.. <T__ + <T_ = 2y + (vr) . Periodic Potentials g= 2aby/\. Prom this condition we determine later (cf. this implies y+(Tr)/a = ± 1 . (17. (17. (17.22)) the boundary conditions of periodic solutions. i.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. i.1.= — . **This point is not immediately clear from mathematics literature.8) "Observe that for integral values of v (in lowest order of h2). 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. Setting cr+ = e .7.455A2 + 1291A . ** An important consequence of Eq.344 Setting / = 2by/\y+(Tr).e.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. we have for the sum of the roots a+ + <7_ = 2 cos TTV or" C O S T T I ^ ^ ^ C O S V A T T . .6) We can derive various terms of this expansion perturbatively with the method of Sec. 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. cr+ = 1/<T_. (5i/ 2 + 7)/i 8 vz + — . T h e result for A is the expansion A = h4 2 . <7__cr_ = 1. _ / ± v 7 ^ V .2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .e. 8.0 This series may be reversed t o yield the Floquet exponent. This calculation is demonstrated in Example 17. where the constants are usually taken as unity from the beginning.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). Eq.
Meixner and F. = J—^ + o(hr). (17. These cases therefore have to be dealt with separately.11) or smvz or e±ivz.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. Solution: We write the eigenvalue equation A = v2. . p. It follows that in these cases i? = n2[l + 0(h2)].„„ x n.17. we must have a2 — 1 and hence e 2iv/A7r = 1} i. as will be done later. (17. Example 17. (17. For these integral cases of v we demonstrate in Example 17.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. (17. cos TT\/A 32A(A1) 2 . Schiifke [193].1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12).7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. n an integer.12) S e e J.2 how the expansions for small values of h2 may be obtained perturbatively.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.5) and (17. (17.1) 15A2 .35A + 15A2)/l8 /. (17. 8. W. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . VA = ±n.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec.10) A = n2.35A + 8 . 124.e.3).jl2. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0.
3 and 19. or A = 0(/i 4 ). (17. i. v) = 0.!/) = 2A We now observe that DvVv = 0. . i. ("• ^ + 2) (o) 2(2^ .2. Schafke [193]. . We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . where except.v±2) = l. h2(u. up to 0(h2) we have A = 0. 17. i. and (v. Next we treat terms yv+a in j / 1 ' in a similar way.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions.h2). = . see Example 17.e. (17.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . v . me±v(z.16) *These smallh expansions are convergent.e. for instance. of course. v)yu + {v.e.346 CHAPTER 17.(2) + . e ± i ^ + 2 ' 2 + e±i("2)z. W.„ + £ h2i £ P2i(2j)yv+j (17. '.2) which applies when v is nonintegral. The recurrence relation can also be obtained by substituting the right hand side of Eq.16) directly into the original equation in the form of Eq.2) _ __ _ (y. h2). With our perturbation formalism — as demonstrated in cases considered in Sees. (17.2)j/„_ 2 + {v. se and me as* oo i y = „«>) + yW + i.3. (17.. Proceeding in this manner we obtain the solutions ce.15) Hence a term fMy„+a on the right hand side of Eq. by J. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. (i/. v .e. The right hand side of Eq.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) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. when a = 0 or 2v + a = 0.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.14) therefore leads to the following nextto leading order contribution (v. siniy + 2)z + sin(i/ — 2)z. (17.v + 2)yu+2]. since j / 0 ' = yv leaves unaccounted R\. Meixner and F.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v.h2).14) = = = cos(y+ 2)z + cos{y— 2)z.11). sev(z. as shown. (17. (17. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17. . Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.
See e. 0 = h2(u. (17.15) (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)+. one has to deal with each integral case separately as in the following example.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A.v)\ . For v = 1 we then have yi = cosz = y~\. .u) + h4 (". the solutions are not yet normalized. p. The higher order terms naturally require a little more algebra.. In the case of integral values of i>. (17. Meixner and F.7). 121.1 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]. . i..e.17. W.17) Inserting 1 for the stepcoefficients of Eq. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y.. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. (17. Example 17. . „ . we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero.2—r——— (i/ + 2.2 ) v(y2.g. Following the first few arguments of Example 17. thus determining the quantity A.2: The eigenvalue A for u=l and h2 small Use the perturbation method of Chapter 8 to obtain the perturbation expansion of A for v = 1 and h2 small. + y~i) (17.R{ . J. v) ' ( 2 v + 2) (17. 2 ( 2 i / . . fo^ + 2) .18) Again applying Eq."2).1 with A = v2 — 2h2A but with v = 1. Solution: We start as in Example 17.14). * Note. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. Schafke [193]... however.
Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. In the above we considered the case of v an integer and calculated the eigenvalue A. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z.24c) below) . 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 . Periodic Potentials h4 r ••• . *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . h2). see below!) one obtains the additional terms given by Meixner and Schafke [193].348 From this we obtain 2h2A = h2 CHAPTER 17. . 123) (there the solution is called cei(z.n integers. i. J. cos z h cos Zz 8 h2 h4 ( cos 3z cos 5z — cos iz A 1 8 4 l 8 48 h4 (cos 3z cos 5z 1 4 I 16 48 (17. The reverse situation is later of importance. We consider this case again in an example.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. h4 (17. Schafke [193]. (17. Eq. and v is to be found in ascending powers of h2. Meixner and F. the case of A an integer.e. We introduce a normalization constant c and rename the normalized solution y then yc. 120. W. m. 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. 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].19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai).* The associated solution is ym+ym. With normalization (not to 1. p.
J . fe47T2(A4) 4%/A(A . We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .1) (A .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.('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 . MiillerKirsten. ^S.9b) about A = 4.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A ..9b) v — 2 + <5.3 1 x 4 + 8) 274233 (.3 5 A + 8)TT2(A4) _ 3 / / 64(A . Gubser and A.• 1 + 5n h 2 s 2932 Setting in cosm.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . J. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . Manvelyan.l ) 2 (A . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.4)Av A2v A ~ 32A(A .17. Liang and Yunbo Zhang [189]. (11A2 .4 ) + .1)2/A TT2 +h8 (15A2 . Hashimoto [124]. (17.2 Cosine Potential: Weak Coupling Solutions Example 17. 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 .1) .3 1 A + 8) ( A . W. Substituting these expressions into Eq. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. H. (17. .9b) and considering the approach A — 4 gives > (A4)2 + .Q . (17. on the left hand side of Eq. S.. . o 1• • • .
These are defined by specific boundary conditions.21b).10) applies.w^.W4W=UtWSta±I(. i. in F. 17. . may also be found. We now derive these boundary conditions. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.6. as also conditions (17.350 CHAPTER 17. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq.e. "These conditions.2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir. 17. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. We verify these for the large/i 2 solutions in Example 17. Taking in particular Eqs.1 Boundaries of domains of stability. bV\ y+(0) 2/40) 07.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir.2. solutions for integral values of v.. Arscott [11].4) into account. Fig. for instance. (17.21a) and (17. M. = and s+w 2tW±I. 28 . pp. (17.29.
1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. which after some rearrangement can be written y+(vr) _ .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. we obtain y+(7r) _ i  2y_(f).17.4(f) (17. Then y+ y 351 U . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. o y+(fy(f)ob\/A _ = 1 + 2I&A/X . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . . .21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. /7r\j/!_(7r) y L l 7TJ . ^\y±M /vr\yL(7r) . /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.
(17.T + . There the second term on the right is nonzero. 19. 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 .2.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . (17.' fh2 V2ceo = l .21a) and (17.. (17.352 CHAPTER 17. A *• b2n+2(h2). . . (17. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. in this order. 17.21b) must be ± 1 . A > b2n+i(h2). A 2 y • ce2n=q0i.2): hA 7h8 29h12 + Y l28^04.24c) 8 h2 sin 4z \ .3. of v is even or odd.• a2n(h2).+ . (17. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. the case of ai with solution cei (of period 27r). sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . 2K.— — — • • • .2 we require this relation for nonintegral values of v. These may be subdivided into classes depending on whether the integral value n — 0.23) We can now see how the domains of stability and instability arise.5) that the left hand sides of Eqs. ce± = cos 8 64 h4 5h8 b2 = 4 1 . se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .. 17. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. IT. y > se2n+i=qo. y > ce2n+i=qo. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.24d) v 12 ' **To avoid confusion we emphasize: In Sec. was treated in detail in Example 17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2. Mr7nA (17. 27r. (17...1.. This is shown schematically in Fig. (17. These are defined by the following boundary conditions.1).24b) 8 h2 z —— cos 3z + • • • .. We see that there are four possibilities and hence four different types of functions. and whether the function is even or odd.22) * a2n+i(h ).
z3 ce2 = cos 2z . The cosine potential cos 2z depicted in Fig. 17.> (17. 17.24e) 17. Dingle and H.2 The cosine potential.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials.hz — 12 .3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . .1 that the boundaries of regions of stability merge to lines. 13824 .2cos4.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. B. .3. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator.3 17. 17. Miiller [73]. of course. J. We can see from the inverse of Fig.17„. In the case of the cosine potential. Thus in Fig. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain. The main objective of the present section is therefore the calculation of the tunneling effect in the form of a splitting of the otherwise asymptotically degenerate harmonic oscillator levels j t Fig.3 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. 353 . the walls are not infinitely high and hence tunneling occurs from one well into another. .2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.17. W.
This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1.2. there is tunneling from one barrier to the next..3. In the case of finite heights of the potential barriers. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .3) and is.2 and 8. Ion (17. as in the case of the cosine potential. 4 (17.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. We therefore wish to expand it about these points. 8. (17.2 The solutions We insert Eq. Periodic Potentials minima at z — ±ir/2.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . simply the onedimensional Schrodinger equation for the harmonic oscillator.354 CHAPTER 17..1. (17.2/i cos z V = 0. but only approximately so. we therefore set 9 A A = 2h2 +2hq + —. where A / 8 is a remainder.. Sees. n 0.26) 17. and we set this equal to q.. Taking the remaining terms of the cosine expansion into account.27) .i r ) ~ 1 — 21 z ± 7r Changing the independent variable now to the equation is approximated by A + 2/t2 dx2 + 4/i 2T y = 0. of course. Thus in this case the quantity on the left is not an exact odd integer.
17. 16/i/ (17.30) D A): COS « = 4z + 1 iq sin 2. 17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig. (17. (17. and substituting this into Eq. can be . (17. the linearly independent one.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) .29) is known. z).3 Schematic picture of Mathieu equation eigenvalues.31) We make the important observation here t h a t if one solution of the type (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).28) A" ±4hcos zA' + 2h[ q =f sin z \ A. 2 (17. another solution. Choosing y(z) = A(z)e2hsinz.27) we )A = 0.
The solution Aq(z) of the first order equation Dq Aq(z) — 0 is c o s ^ 1 ) ( \K+\Z Aq{z) = sml(9+i) ( ivr + \ z ) COS 2 9 ( D I\n+\Z (17. With some algebra one finds that 16^ + where (9.Q)Ag + ( ? .e. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. (17. i.9) = 0.4 ] . .4 ) ^ . 2[(92 + l) + A]. (9l)(93).356 CHAPTER 17. 1 + 4 ) ^ + 4 + (^.\z To lowest order the solution A is A^> = Aq. Aq+4 2Jh ._ 4 . when i = 0 — the terms not involving Aq in (17.32) cos^'+i) [\K .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. ? .4)A.30) amounting to Ri ] q = . of course.g + 4) .35) not yet taken care of is set equal to zero. (17.33) (17.35) But since and so a term iiAq+n on the right hand side of Eq. 2 /i D^Ag = 0.94) = (9 + l)(9 + 3). ? .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g.^ 6T [(«>« + 4 ) A ? + 4 + (Q. (9.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.9) (9. 2 A A = («.9 + 4) = (9. (17. (9.94) \ — Aq4 (17. D{q%Aq+4i = 0.36) Then ^4. ^ ) A + ( ? .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. (17.
37) Here all the terms.37) and so on — left uncompensated — must vanish. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. . Q . q . .8)A.<J+4 Aq+& + 1 2 1 . (17.. h). q (17.q 8)Aq+8 (o)' ^±V(q 1 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. +4. q. (q. (17. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.)+ & ! = « ( .4. h) = A(z. ) ^ 4.35) and (17.g + .q + 4) (g._ 8 (17.4..q 4)Aq+4 q + 4 1 q + + + + + <^(.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. Thus 0 = (g. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .g4) 4. q . h). except those involving Aq. .4) 1 1 (q.39) which is the equation from which the quantity A and hence the eigenvalue A is determined.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. q. to satisfy Eq.38) Aq8 + 1 2 Clearly we can continue in this way.40) . Thus in particular A(z. (q. q. A(z. (17. q . (17.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.17.4 . A..q + 4)(q + 4. Then for A = A(0)+A(D+A(2) + .g + 4)(g + 4. q. the sum of the terms in Aq in Eqs.4) (q . h) = A(z. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17.
" Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (91) H (17. 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' (17.42) (17. and discussed there. .43) 9 2eh . i.i ) We encountered Hermite polynomials earlier in Chapter 6. it is found that D(B) z+ — ]B = 0. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0.67). Before we study the eigenvalue equation in more detail.) i ( Q . We observe that for the large values of h we are considering here. (6. where Hen{w) is a Hermite polynomial when n is an integer. 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.45) {B) We recognize Dq as the differential operator of the Hermite equation. 2 w2 + H* (17. Note different definitions in use. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). The differential equation there is Eq.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 .44) where D(B) 9 = dw2 $_ !)• (17. (17.e. We return to Eqs. (17.41) so that the early successive contributions decrease in magnitude.358 CHAPTER 17.d2B w dw2 + w(l w dB_ dw d dw 1.46) 2s(«.
.q + 4 ) £ g + 4 + (q. with the equation C" + 4h cos z C + 2h ( q ..47) where the coefficients are the same as before. as in Eq. the expansion for the quantity A being identical with that of the previous case.17.e.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.e.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17.50) C = 0..7 T H 1 2 Z 4 <1. (17.48) This shows that this solution is valid in particular around z = ir/2. i. which is an amazing and unique property of the Mathieu solutions — cf.51) . Finally we obtain a third pair of solutions. q)Bq + (q. A{wz (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.34).e.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ . The solution with contributions B^°\B^>.. i. 1 COS I .4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. (17. i.3 ) \He* ! (9+1) H.qi)Bq_4. also the same coefficients in the case of functions C below.e. y — Cexp(2hsinz) and Cexp(—2/isinz). (17. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. 16hJ (17. 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. i.
q + 4)C 9 + 4 + (q.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. . ie 1 COS I . (17.w2)He'n . can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. (9+3) (2n 2 + 2n + 1) + .q. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). (17. 2 J where n=(ql).4)C g _ 4 . (17. (17.n2(n l)Hen2 . Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq.54) where the proportionality factor is complex.q. We have thus determined three pairs of solutions of the Mathieu equation.(w2 + A)Hen.59d) of Hermite functions Hen(w).53) This solution is valid in particular around z = —7r/2. We note that C(z. show that the perturbation remainder of Eq.4: Hermite function linearization of perturbation Using the recurrence relations (6. each of them associated with one and the same expansion for the eigenvalue.44).4.52) and the coefficients are again the same as before. A closer look at that expansion is our next objective.q. q)Cq + (q. 17.360 CHAPTER 17. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.A Hen . These domains of validity are indicated in Fig. 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. Example 17. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.h) ocB{z. 2' the remainder 72.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . He'n(w) = wHen(w) Hen+i(w).7 T 1 2 Z 4 <1. (17.h). The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.
1i .. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.q + 4)Bq+4(w) + (q.17.4) = (q . l)Ag4 + P2(q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.0 and —4.57) . 2)Aq_8} + • (17.q4) (17. (q. with (. (?) 1 1 + (9.g4)(q4. (q. (b) The allowed moves are +4. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.' 361 in 2*f*1>[i(gl)]! = (q. 2 + ^ M 3 + (17.)Bqi(w). q . We can write the solutions obtained previously.l)(q .j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . 17.3 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).. r j=r.g. ) . the solution y = exp(2/i sin 2)^(2). q + 4) = (g + 1)(? + 3).q 4.55) M2 = M.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . no intermediate move to or from q being allowed. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q. .l)[l(q _!)]. e. q)Bq(w) + (q. ^ 1 Hv" '^ ^' ^ ' . q) = 2[(q2 + 1) + A]. in the following compact form with coefficients Pr(q.i ) ( w ) 2l(. (q.3). .q + 4) (q + 4.g) 1 1 and so on.56) (94.3.q + A){q + A.
W. and in general* M2r = J]j[P2(g. j) = (g + 4j .3) ^3(9. .g + 4)(g + 4. j) +(g + 4j + 4.g + 8)(g + 8. Miiller [73]. 2) 1 2 ' ' ^ ' ^ 1 (g. 0) = 1 and all other P0{q. J. M2 = P 2 ( g .j). Periodic Potentials PlM)=(M+i). P2(q 2) Pl(g.g + 4) ^3(9.j) = (l)rPr(q.4. and the contribution from a sequence of r moves from q + 4j back to g.j)Pr+1(g. Dingle and H.P 2 ( g .j). p = (<?><? + 4 ) ( g + 4 >g + 8) .g + 4) 2 1 (g.g + 4)(g + 4.j). x)_ (g.l ) .59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.34) we deduce that Pr(q. j .g + 4)(g + 4. the factor j removing a duplication of factors in the denominator at the junction.j)] = 2j]jPr2(g.g + 12) 1 2 3 (g. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g._1) = Mzil. for instance.j ^ 0) = 0. (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 j ) P r . g + 4j)P r _!(g. . From the coefficients (17. Thus we have. (17.362 Clearly CHAPTER 17. l ) .58) with P 0 (g.j)^2(9.e. B. j + 1) (17.j).1) + (g + 4j.i (g.j). i. g + 4j)P r _i (g.g + 4 ) ( g + 4.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.g + 4)(g + 4.g + 4) + 1 1 1 and so on.g + 8)(g + 8.60) M2r+l = = *R. as jP r 2 (g.
J.( 3 ^ . This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. as a result of tunneling between wells. Miiller [73].62a) f R . remains to be determined next.5 the following proportionalities there: B[w(z)] = aA(z).3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 .284g + 57) + (17. of course. S.17. L. where solutions overlap.4. the solutions and eigenvalue expansions again have the same type of symmetries. In fact we demonstrate in Example 17.^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. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. Up to > > this point.2 * +3) + ^ 2 (9g4 .(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . Its precise deviation from qo. 17. [116]. Ince [136]. Dingle and H.92q3 + 70g2 . as pointed out earlier. [137].4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. B. W. . *E.* 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. Goldstein [115].3. 17. a= (8/i)3(9x) [1(91)]! 1 . It will be seen in Chapter 18 that in the case of anharmonic oscillators. h — —h. We observe that there are regions.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.^ .
62b) *z domains of overlap Fig.32).5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA. and determine the constants a and a.(17.7T H Aq(z) = = C O S ^ .1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . a .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . Solution: We begin with the solution containing the function A{z)._ ^ l CHAPTER 17.4 Domains of solutions and their domains of overlap. The first term of the expansion of A(z) is the function Aq(z) given by Eq. We write and expand this function as follows: . 17.1 4 .COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .1 ) ( 7T + .364 and C[w(z)] =a~A(z). Example 17. Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.
22). these are conditions imposed on even and odd solutions at the point z = it/2.92q 3 + 70q 2 .2. (17.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.53) for the expansion.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. with e. i. (17.284g + 57) + • The other relation is obtained similarly.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. 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 .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.i ) 1 (g. s .2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .17. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2._i)]! w*te» 2l(91 (. Since the solution with A is obtained from that with A by a change of sign of z.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.63) In Example 17. (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.g. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. 17.e.36). (17.61) for the connection and Eq.g + 4) (q + 4. (6. we have to impose the appropriate boundary conditions.!)(. As we saw there.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. A*1) given by Eq. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73].46). \ql)/2(w) 2i(9D[i(. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. we obtain Bq[w(z)} H. the coefficient of the dominant factor c o s ( .q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . with (8h)li''1) (3g2 .
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.59d). W. Oberhettinger [181].22) and to correlate the solutions to those for small values of h2. and then replacing 2n by i>.63) to the domain around z = ir/2 by using the proportionalities (17. give the following expressions for polynomials with argument zero: ~19)"(^)!. p.e.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. a ± C[w{z)}c_2hsinz^ a (17.g. He'v(0) = Hev+1{0). The functions defined as in Eq. With the help of the duplication and inversion formulas of factorials. He'u{w) = wHev{w)He„+1{w). Periodic Potentials We can extend the solutions (17. Thus y+ = ce and y_ = se. He2n+1 (0) = 0. i. (17.62b). . 80.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. one obtains the first of relations (17.» a \dw _ C[w = 0}c_2h a w=0 = Q (17. Magnus and F.66) ^E.366 CHAPTER 17.e. Then y± oc ^ M e ^ i n .66) then follows from the recurrence relation (6.62a) and (17.22) are therefore given by the following boundary conditions: 1T y'+ v+fao I'=° y a a\dwjw=0 B[w = 0]c2h a a 1h dw\ ! ' .66). The second of relations (17.
» 2 6 /i the condition reduces to the equation cot{ .11.7.67a) g 106g 87 2 14 /i 2 (17.17.106g .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 .39 2~^h? (17. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative. and using the reflection formula (8.69) If the right hand side of this equation were exactly equal to zero.136g + 9) + • • • .g0)3].65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.40g3 + 18g2 . .87 4 2 ¥w (17. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .1) 7T .67c) 4 2 \dw)w=Q C[w = 0] [\(q . we have cot j ^ ( g ~ 1)} = ~\<(l .82g2 .67c). the solutions would be given by q = go = 3.67b) <f ~ 82g^ . f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .68) Inserting here expansions (17.! ) ] ! [ .39 (17. and similarly for C and its derivative. Hence expanding the cotangent about these points.Qo) + 0[(q . (17.\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .67d) Considering now the second of conditions (17.( g .67a) and (17.
e.73) We now return to the eigenvalue X(q) and expand this around an odd integer qo. . The remaining condition (17.40g3 + 18q2 .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .528gg + 3307Q . Thus we have 2 (16/i) qo/2 T2 4.9. we use \(q)~\(q0) + (qq0)(^J . .5.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.72) This leads again to the result (17.71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .40gg + 18gg . Considering now the first of conditions (17..136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .65) yields the same result except for a change in sign.70) + ^ 2 (9?o .40gg + 18gg .(17.70) except that now qo = 1. (17. i. = _a a 6 _4h (17.136g0 + 9) + • Clearly the last of conditions (17.408g0 + 1089) 2 /i +• (17.65) leads to the same result except for a change in sign.136q + 9) + .... Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .120g£ + 467g .
120.3 we call this difference 26E„n. The dominant contributions were also found by S. The tunneling effect.e. example 3.1. J. B.2.. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. quite apart from applications. just as any solvable problem is important for the insights it offers in a concrete and transparent form. i. 9o = 2n + 1.87) = Azpteo). .3. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential.61). In the literature one finds frequently the statement that the splitting is a nonperturbative effect. 20.17. Dingle and H.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.. (17.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w .t From Eq.88q0 . We see here that *R. 17. M. (17. p. (17. *In Sec. (17. In the literature that we follow here* in all cases several more higher order terms have been given. Arscott [11]. though approximate.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. See also F. Miiller [73]. Goldstein [116]. and not the separation of these as a result of tunneling. + ^ 2 (9<?o + 89o " 78g02 . plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig.. and that for the odd Mathieu functions se qo+ i and se9o (lower. n = 0. minus sign). W.74b) The above results for the cosine potential are in many respects important. i. this becomes 369 = (16/i)^+1e4h l .74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . The complete.A_(g0) ^ .e.3. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. . made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. results from the boundary conditions.
^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). (17. T y'_(0) y'A±*) y{z). From Eq. The splitting can also be obtained with some methods of large order of perturbation theory. results from the imposition of boundary conditions. evident through the exponential factor exp(—4/i) in Eq. the Schrodinger equation with the cosine potential is a special case of several other more general cases.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). . y^{z) = Aq(z)e~2hsin *. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. _3/+(z) + ____2/_(z). Thus we do not reproduce every step.370 CHAPTER 17. The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential. 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.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . Periodic Potentials this nonperturbative effect. Naturally one expects the solutions there to be related to those of the periodic case considered here. The explicit derivation of the level splitting (17. (17. As mentioned at the beginning. Then we obtain first y±(z) = yA(z)±y^(z).74b) is also important for various other reasons. the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). Example 17. 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. . yA{z) = Aq(z)e2hsin '.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. 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.63) — and using only the leading terms — derive the following set of equations with shifted arguments similar to those of the small/i 2 case of Sec.2. 17. M±TT) . We begin with the solution containing the function A(z). y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. and one can compare the methods. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion.6: Translation of Solutions For the largeh 2 even and odd solutions (17. in effect.74b). The exponential factor represents.
where Aq(0) = Aq(0) = V2.%/2(g . etc. Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .) cos < 7r(q 1) [ = . = ±ir).4ft.g A ? ( z = ±TT).VA DV 371 1 V±.c o s  ~(<1 T 1) f y .. The Lame equation has not been a widely known equation of mathematical physics so . Aq(z) The derivatives can be handled with the help of Eq.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+. ( .4ft) cos i TT(<J + 1 ) 1 . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).• Then y'+ (0) = 0.4 17. we obtain the conditions stated at the beginning. A' (z = ±ir) = (.( ± 7 r ) = c ° s  .17.7 r ( g ^ l ) U + ( z ) q : i s i n { . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) . Similarly we find y(z±ir) = = and by putting z = 0: V.4.7 r ( q = F l ) } j / .1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials. which is easiest by comparison with the foregoing treatment of the Mathieu equation.yV+\z^ = [j/+(0)] 2 (<? . and y+(z = ±TT) = V2(q .y]z=„ = [y+yL .q).T ( ? =F 1) p + (0).( z ) From this we deduce for z = 0.z ) . A ?q( Z = ±TT) = . 17. Replacing sines and cosines by the functional expressions. the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r).[A'qM + 2hAq(n)] = iV2(q .31). (17.4ft) ~ 8(q . but also to enable a brief familiarization with the Lame equation.2hAq(n)] . we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .3 The Ellipsoidal Potential Replacing VA. y'_ (0) = \/2(4h . (l^ifr^i^O + Cl^ifo^j/aCO cos  .( 2 ) .4ft) sin j ^n(q . y+(0)~2V2.Ah) ~ 32ft.
W. L. J. § F .* and the work of both prepared the ground for presentday investigations. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials. The most conspicuous difference compared with the Mathieu equation is. k) for k = 0.. 17. 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. Liang.* As Arscott [11] (p. however. 194) remarks to parallels with the Mathieu equation:". p. H. .372 CHAPTER 17. Tchrakian [165]). so that Floquet's theorem cannot be immediately applied .5 The potential sn2(z.0. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here).75) ' T h i s means double well.1. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940." Fig. 194) remarks..K2sn2u . one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . H.75.. Ince [138].. This line of investigation.4sn4u]y = 0.5 < z < 5. 19.0. Arscott [11] (p. *A. Periodic Potentials far. Erdelyi [85].Q.. MullerKirsten and D.. as alluded to at the beginning.fi2A. the nonoccurrence of the Floquet exponent. f E . inverted double well and cosine potentials (J. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane. M. But more recently it has been observed to arise in various contexts.0 and . (17..98. Arscott [11]. and was continued by Erdelyi. but also cubic potentials.5.
In order to distinguish the above equation from that with the Lame potential. N = 0. The range of the independent variable u is 0 < u < 2K. . where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.e. cmx and dnu.78) We follow in this brief recapitulation the description in H. is the elliptic modulus of the Jacobian elliptic functions snu.K2sx?u}y = 0.K) = qK+^±.76) which can be looked at as a Schrodinger equation with periodic potential K2sn2u where snu is one of the Jacobian elliptic functions of period 2K. 17. MiillerKirsten. If we put fl = 0. (17.17. Although the results have been calculated for the ellipsoidal wave equation.1.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions.77) into Eq. the usual factor —h2/2mo.\k\ < 1. The second step is to insert (17.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). (17. in the case n2 — oo. .77) where q —> go = 2iV + 1. The first step is to write the eigenvalue A as A(q. in front of the second derivative has to be kept in mind (mo being the mass). for very > high barriers (harmonic oscillator approximation around a minimum of the potential). (17. J. . W. 17. i. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects.5 for several values of fc. .2k. In comparisons with the Schrodinger equation.e. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. (17.2 = n(n + l)k2. The function sn2u is plotted in Fig.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K.75) reduces to Lame's equation and one writes K. i. K being the complete elliptic integral of the first kind. Jianzu Zhang and Yunbo Zhang [206].4.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. (17. 2 . Eq. and of the derivation of the level splitting'. the equation ^  + [A . we consider mainly the Lame equation.
77). A second solution is written y = A(u) exp < / nsnudu >. more precisly for dnu = cnu F 1 » . Thus in the third step two more pairs of solutions B. i. Periodic Potentials .q. A(u. by transforming the equations for A.K)=A(u. C replacing yl.q. \/8K (dmz = cn« \ ' F . determine their proportionality factors) in domains of overlap (their extreme regions of validity). _ Solving the resulting equations iteratively as before. one obtains again the same expansion for A. (17. (17. ^ (17.e.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed).80) The domain of validity of these solutions is that away from an extremum of the potential.e. i.K — > A(u)=A(u + 2K). dnu =F cnu . / z J . one has to derive new sets of solutions there and match these to the former (i. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. B and C. but solutions B. A into equations in terms of the variables z(u) = ^—{k' . k' = Vl~ k\ \dmi±cnu/ 17.374 where CHAPTER 17. (17.82 .K).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. the other in terms of those of an imaginary variable.Es(ii). .. . one pair in terms of Hermite functions of a real variable. C which are valid for dnu±cnu <1.e. /dnu + kcnu\ \ dnu — kcnu J For large values of K the equation for A(u) can be solved iteratively resulting in an asymptotic expansion for A(u) and concurrently one for the remainder in Eq. A. A are derived. which are respectively even in u (or snu) or odd. dnu± cnu K Thus one can construct solutions Ec(u).
Tricomi [87]. (17.e. Miiller [205]. 17.38g2 .3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. p. who obtained this expansion for the eigenvalues of Lame's equation (i. Es(2f0 = Es(0) = 0. W." 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) . Oberhettinger and F.75).^ ^ { ( 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£.63)} + ••• . This expansion is found to be in the more general case of Eq.e.6(5g4 . i. a of _ _ B = aA. £2 = 0). Erdelyi.384n2k4(q2 + 1)} q r {(l + A. F.85) The first three terms of this expansion were first given by Ince [138]. 64. Magnus. W. the ellipsoidal equation. G. (17. .17. J. as well as \duJ2K 11 (17.4.83) 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.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. C = a~A. (17. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . *A.
88g0 .g 0 ) ( ^ .90) + 3.4k2(q2 + 2g0 + 5)} + • • • ]. 2K and 4K respectively. W. and the lower to Es*>+1 or Es*5.j 26K2 (1 85) = A(g0) + (q . 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 .Es^ 0+1 . Muller [205].l n 2^+ i J + 0(«i) (17.40g03 + 18g02 .Ec£ 0_1 and Es2° of periods 4K.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)*. . Periodic Potentials These conditions define respectively functions EcX°.87) (17. 2K. Finally expanding A(g) ~ A(g0) + ( g .:J {3(1 + k2)2 (9q* .376 CHAPTER 17.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i . 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.go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .g0 is obtained by expansion around zeros.i Here the upper sign refers to Ec*> l or Ec£°.2 /c +256fc2g0(g2 + 5)} (17. J. (17. from which the difference q .136g0 + 9) n 3. go being an odd integer. 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.89) one obtains the eigenvalues from which the level splitting can be deduced.78g^ . + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .
90) one has in this limit: /l±±\ ~K/U = e .74a).87) agree with those of Ince. W.e. Hence the conditions fi = 0. Miiller [206]. (17.Ah2 sin2 u}y = 0. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . that calculations with the Schrodinger equation are not only easier.4.* Apart from the choice of notation. Dunne and K. One can see.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation.17. L. (17. A = 0. 17. (17. V. Ince [138]. K2 = h2 (say).2h2 . *H.4 h ( i e M 9 0 7 2 \. (17.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite.86). it = x ± ^ . Rao [79].2h2 = A. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. Without going into details we mention again (as at the beginning) that there is also a specialization from the ellipsoidal wave equation and its solutions to spheroidal wave equations and their solutions. . A . J.93) (17. + E. however. Replacing u by x ± IT/2. K = 2/i.* M G .J1 O.2h2 cos 2x}y = 0. .t One can verify that under the conditions stated the results of this case of the ellipsoidal wave equation reduce to the corresponding results of the Mathieu equation. Thus in the case of the dominant contribution of Eq./2i and hence A^ A/ A AU I2 . i.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. but also more easily generalizable. h . the conditions (17.f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. this equation becomes y" + {A . in which they dubbed the classical configurations Lame instantons.
Thus. D.Y. i. A.W. This is an interesting problem.^ The equations considered above also appear in diverse new problems of physics. W. Ouvry [276] and Jianzu Zhang. N. M. J. inverted double well and cubic potentials.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. for instance. H. Kan and K. J. J . MiillerKirsten and D. 1 1 See Z. W.Q . 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. Cho. in the problem of two parallel solenoids the lines of constant electromagnetic vector potential  A are elliptic with the Hamiltonian separating into a Mathieu equation and an associated Mathieu equation. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. . See e. Khare and U.e. MiillerKirsten and J. Gu and S.5 Concluding Remarks The potentials we considered above will reappear in later chapters.378 CHAPTER 17. A generalized associated Lame potential has been considered by A. Tchrakian [165]. Sukhatme [148]. Shiraishi [49]. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. **Y. Quian [123]. all basic potentials. double well. H. A host of related elliptic equations has recently been discovered and studied. § . Periodic Potentials 17. S. See Chapter 25. de Veigy and S.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). Liang. Ganguly [103]. Rana [287]. A.g. H.
Achuthan. 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. ' T h u s 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. MiillerKirsten [163]. T. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. Weinstein [281].. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. This seemingly simple problem revealed extremely rich internal structure . 379 . Liang and H. T.Q. D. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems.. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Priedberg and T. MiillerKirsten and A. W. [19]. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. Wu [18]. The recent work of R. Wiedemann [4] and a revised version of parts of this reference by J. In particular the investigations of Bender and Wu. Bender and T. Wu. f C .t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory.Chapter 18 Anharmonic Oscillator Potentials 18. H. J. M. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Bender and T. W. J. both of which make the calculation more difficult.
which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. 18. V(z) = \\h*\z* + \\<?\z*. *~z (1) (2) (3) Fig. with the potential described as an inverted double well potential. 18. V(z) = \\h*\z2 I 2i 4 \cr\z . 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. 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. (3) Complex eigenvalues with tunneling: In this case.380 CHAPTER 18. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. These contributions may be given different signs.1 The three different types of anharmonic potentials. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly .1. described as the case of the double well potential. and thus lead to very different physical situations.
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. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. that the general applicability of the method becomes evident. (18. In this case our aim is to obtain the aforementioned complex eigenvalue.18. as is sometimes hoped. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. however with decay as a consequence of tunneling. 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. 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).86) below. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. We do not dwell on Case (1) since this is effectively included in the first part of Case (3).175) below. (18. In the case of the double well potential our aim is to obtain the separation of harmonic oscillator eigenvalues as a result of tunneling between the two wells.1 Introductory Remarks 381 normalisable wave functions. We begin with the latter. since the potential decreases without limit on either side of the centre. The result is an expansion in descending powers of h?. The imaginary part will be rederived from path integrals in Chapter 26. The level splitting will be rederived from path integrals in Chapter 26. . The eigenvalues of this case are given by Eq. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way.e. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17. except for a change of sign of c 2 . Case (3) is seen to be very different from the first two cases. The shift of the eigenvalues is best calculated with straightforward perturbation theory. 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. it will not be possible to obtain the exponentially small contributions with convergent expansions. Thus we are mainly concerned with the double well potential and its inverted form. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). The eigenvalues of this case are given by Eq. i.
The potential in this case is given by V{z) = v(z).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. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.2 The inverted double well potential with (hatched) oscillator potential. h 4 2 2h1> ~1/2 2c?. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. a point which has to be kept in mind in comparisons. 18.1 refer to the two cases. 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 . v(z) = ^h4z2 + ^c2z\ (18.1 and more specifically in Fig. Anharmonic Oscillator Potentials 18.2 18. We take here h = 1 and the mass rriQ of the particle = 1/2.3) "1/2 harmonic oscillator h8/25c2 .2.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. (18. (18. If suffixes 1/2. 18.382 CHAPTER 18.1) for h4 and c 2 real and positive.2.
the harmonic part of the > potential dominates over the quartic contribution and Eq. The positions z± of the maxima of V(z) on either side of z = 0 in the case c2 > 0 are obtained from h2 v'(z±) = 0 as z± = ± — with (18.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. (18.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. 4 . to match these in domains of overlap.2. The problem here is to obtain the solutions in various domains of the variable.5) becomes .1.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.1/7  Vg(w)y(w) = o(J^j.. _i\(w) and q = qo = 2n + l.18.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. (18. \h?\ — oo and c finite. 18.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0.d2 . and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.e. and a variable w defined by setting E= b* 1 + A A and w = hz.. The result will be that derived originally by Bender and Wu.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w).2. y{w) oc Di.4) we can rewrite Eq. n = 0. /i).. (18. (18.. i. (18.
11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.14a) 1/2 VA( ) = A(z)exp (18. We observe — before touching the square roots in Eq.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) .+ ^ .12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. (18. as we shall see. This observation allows us to define the pair of solutions yA(z) = . Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. 1 d r A!{z)±iA{z)±{ 0. h) is obtained from the perturbation expansion of the eigenvalue.9) z =0 these Thus these c2z4 1/2 + • 2 (18.9) we obtain 1 h4z2 c2z4 A L2 (18. Before we return to solutions we derive a new pair which is valid in the adjoining domains.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 = 0.4(2. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.14b) .13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. •A~2 hrz" 2^\ c^z V2 —— + (18. In order to arrive at solutions we set in Eq.10) into (18. as encountered and explained earlier.384 CHAPTER 18.2. these solutions are not valid around z = 0. (18.10) E= qh2 ^ j. Inserting (18. be and A = A(q.2 We are concerned with the equation d2y(z dz2 where ^ 1 .2 A (18.
/ c z Aq{z)+0 1/2 . (18.1 and as a verification again in connection with the solution y# — as the other solutions.19) We see that one solution follows from the other by replacing z by — z.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z). i. 1 (18. (18.Aq(z) approximate the solutions of Eqs.34) below) — to be derived in detail in Example 18. Clearly Aq(z).zh2 =()*2"(18.20b) .20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. (18.16) For large h2 we can write the Eqs.13) and A(z) that of the lower of these equations.18. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq.4 .15) where A{z) is the solution of the upper of Eqs.e. we do not pursue their calculation. We take the square root by setting z2h^1/2 +. (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. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. /H 4 Z h +~C Z 21.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18..17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18. 385 (18. (18. Since these higher order contributions are of little interest for our present considerations. 1 2 4 2 Aq(z) + 0 (18.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.
. Anharmonic Oscillator Potentials i.! ) ^ .21) Example 18.. h2 —> —h2).e..386 CHAPTER 18.17) and (18.2 . 03 = . .20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.[yA(q.e. 128 8 16 A = i. away from the central minimum.1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"^ The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z). (18. z)). As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). c*i = 1 1 . z) ± yA(q. (18. procedure. (18. 1 . A = . h2. we write > V±(z) = .e..20b)..• Using Eqs.18) we obtain A » = H^rL^z) 2 z .20a) and (18.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A.20a). one observes that with £>. ft = . zA'(z) where the expansion coefficients are given by ao = l.:=*£ +i(5l). 2 . X » / A q + 4 i = Mg+4i . i.f . W i t h proper care in selecting signs of square roots we can use the solutions (18. Vq+ii = Vq + 2i. h2. (18.v ' = o (« . 04 = 5 . a2 = 2 1 . to obtain in leading order the eigenvalues E. A A _ 3 .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. (18. i.l)A(z) 1 = tfA"(z) A . * = — .13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . „ .2 ( * ) 2 and from this or separately A"(Z) : 1 (q .
. In this way we obtain the next order contribution A^1' to A^°>. 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.23) . (18. those in Aq(z).'2H9i) ^faDfr™) and . give an equation from which A is determined. w = hz. The solutions yq(z) of the equation Vq(w)yq(w) = 0. Rq '. when i = 0.5). (18.. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .. +1J±iw) (observe that Eq.(q + 3)(q + l)Aqvri(q '^ ' .22) are parabolic cylinder functions D i . of course.w — > > ±iw) or functions R B w q( ) .18. 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= . (18. i. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B.. _1J±w) and D_\. 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). (18. = •>— r. [1(9 + 1)]! .2 The Inverted Double Well Potential Thus a term /j. r ( CJw) = ^ .—77 [i(9l)]. ^f( g + i)(±H2^ + 1 > \ — 1.22) is invariant under the combined substitutions q — —q. We return to Eq.e.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.
(q±2)(q±S). l(q2 + l).3.0) = = = ±(q±3)(q±7). Tricomi [86].0). Vq+4j = Vq + 4j.q}=A + c2S4(q. and for j ± 0 : [q.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 + .q + 4j] = c2SA{q. Magnus. . 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 .^ For higher even powers of w we write i w2l V<i = Y.l)2/?4. since Vqyq = 0.27) Now. pp.24) The extra factors in Eq. (18.Aj)(18.q + 4j}yq+4j. Oberhettinger and F.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.4j)yq+Aj. Actually these factors can also be extracted from W K B solutions for large values of q. F. Erdelyi. 115 . we also have Vq+^yq+tj = 0. (18. S2i(q.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. W. (18.123. and so Vqyq+Aj(w) = Ajyq+Aj(w). (18.25) where in the case i = 2: SA{q.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. (18. G. j=i (18.388 CHAPTER 18. Comparison with our notation is easier if this reference is used.29) ' A .( < ? . See Example 18. As an alternative to Bq(w) in Eq. (18.±4) S4(9.±8) 5 4 (g.3)y g _ 4 .
This type of invariance is a property of a very large class of eigenvalue problems. .34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. h ) = qh . .3 °) For the sum y{w) = y(°)(«. (18.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. Equation (18.(!) («. h2 —> h2. so that the entire expansion is invariant under the interchanges q . i.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. and c 2  replaced by — c 2 . (18. (18.[rp ) Y.—A(q + 1) .) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. 2 .32) Evaluating this expansion and inserting the result for A into Eq.) ty^^w) to be a solution to that order we must also have to that order [q.18. (18.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq.31) Proceeding in this way we obtain the solution y = yW („.) + j. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. p i(9»9 + 4 J'WK7». .q]=0. Thus the next order contribution to yq is y H = ie 2^ —z.e. We can now write the solution y(w) in the form y{w)=yq{w) + Y. (18. A = ^(q2 + l)c 2 + o(J^.10) we obtain E(q. n — 0.• q.e. i. ( 18 .35 ) .1.—v*+*j ( 18 .^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J. and even powers of q in combination with even powers of 1/h2.
38) [yB(z)]argz=7T = [?/s(2. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. H. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. Wiedemann [4] . q + 4j) = 0. We thus have the following pair of solutions VB(Z) VB(Z) = B w . MiillerKirsten and A.390 where for instance v J CHAPTER 18.)]arg2=0 These solutions are suitable in the sense of decreasing asymptotic expansions in the domains They are linearly independent there as long as q is not an integer. i. w — ±iw. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. and for j ^ 0 all other P0(q. J. q) = = 1.37) For further details concerning these coefficients. ? T 4 ] [ g T 4 . o. Achuthan.i 1 + 4j) in complete analogy to other applications of the method.—. their recurrence relations and the solutions of the latter we refer to the literature. 2 4tPi{q. > "p. W.11) is invariant under a change of sign of z.q±A) = MM!i±MM+[M±8][g±8. We observed earlier that these are obtained by making the replacements q —> —q. (18. we may infer that given one solution y(z). Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q.36) with the boundary conditions p o(q.g±4] ±4 ±4 [g." Since our starting equation (18.e. there is another solution y(—z).argz=0 (18. g ± 4 ] T4 ±4 and so on. Again we can write down a recurrence relation for the coefficients Pi{Q. ^VO(<L<?) ^i(?. q + At] (18.
2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.18.The solutions yc. 18. with normalization since the normalization constants are also asymptotic expansions. J. Mauss [192]. Such contributions arise. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. (18. in fact. one has to stretch each by appropriate expansion to the limit of its domain of validity. like w of Eqs.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.g.4)). (18.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . Aq(z)) does not appear in any of the higher order terms.g.z2 . We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. the solutions of type A being valid away from the minimum.q + 4j) as in ys.g. Integrating and expanding as follows since h? is assumed to be large. In order to be able to extract the proportionality factor between two solutions. h2) in which odd powers of q are associated with odd powers of h?.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. Thus in the transition region some become proportional.2. 4 . .h^ih.39) are with the same coefficients Pi(q.40) "Variables like those we use here for expansion about the minimum of a potential (e. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.h^ih 391 (18. see e. Vc(Z) = [V!B(Z)}q^~q.
.41) with the solution ys(z) of Eq. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.43) W + 1)]! w^+V A . Comparing the solution VA{Z) of Eq.42) H W2 (ql)e$W I „ .0x ! [ £ ( g . 4/T 4 Prom (18. The function Di.7 r . eh* /12c* e\z*h? (18.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. . y^(z) we see that in the direction of z = 0 (of course.(«.w yB(z) ~ B.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw. Anharmonic Oscillator Potentials Considering the pair of solutions ju(. 2 ^i![i(g4il)]!(2«. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . (18. with z in ys(z) replaced by — z. > n.45) .43) (since z — — 2 implies > argz = ±7r). We do not require this at present. (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.l ) ] ! ( .2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.4 i .44) In the solution y~g(z).392 CHAPTER 18. 41) The cases of the solutions of types B and C require a careful look at the parabolic cylinder functions since these differ in different regions of the argument of the variable z. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\.2 « . we would have to substitute correspondingly the expression (18. (18.42) we obtain for the solution ys(z).z).
49) d yc{z) Again there is no such simple relation between yA(z) 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.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 h2.21) as even and odd about z = 0. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.42) into y~c(z) w e obtain Vc(z) = (h*..9. (18.46) However.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .yc{z)pansion (18.. large probability for the particle to be found thereabouts. hence we have to . we see that in their common domain of validity yA(z) where a = =Vc(z). the ratio of yA(z)iVB(z) i s n ° t a constant.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0.2.e.1 ) e 12c 393 Kgl)]^"1) l + O h? (18.18. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.7. We proceed similarly with the solutions yc(z). (18. i.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. The first of the conditions (18.41).50) will be seen to imply qo = 2n + 1 = 1. (18.y'_(0) ^ 0.50) and y+(0) ^ 0.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq. At z = 0 the solutions of type A are invalid.. Recalling the solutions y±{z) which we defined with Eq. 18. and the second qo = 3.5. Looking at the potential we are considering here — as depicted in Fig.11.
e. we obtain i V7T24( g .54) dio q(w) .(iu)]. Stegun [1].( g + 3)}.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.SxW) = \ i»fl(0) .53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin.i ( g + 3)]l (18. Abramowitz and I.F ? .1)('») + ^ r + l)]! ^. IT 4 dio In fact. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.394 CHAPTER 18.(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)}. .57) [i( 9 _l)] ! 2 i(9i) B. M. — C. 1 [7(93)]! n «(°) = . y c ( 0 ) . + 1)]!' D ' j . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.i ) .g.s i n { . Expressions for C 9 (0).2: Evaluation of yB(0). Z (18.5 ( 9 + !!)(*"')• i ( 1) .55) Solution: Prom the literature.(0) = and htoi)(°) ^—'.oi(. A. [C.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.(0) (18. We leave the detailed calculations to Example 18. one finds that [ 1 IT (q + 1)1! = .l)(°) Thus with the help of the reflection formula cited above: D B.i)H = e. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. Example 18.fc(0) Thus we obtain the equations (18.^{(9+1)}. with the reflection formula (—z)\{z — 1)! = IT/ simrz.! ( ./o^Tpif (q+i) . y'B(0). ( 9 + 3)]! s i n { .l ) ] ! [ .' '. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.(«.58) diu [  ( g .51) and 0 = V(0) = Jim i»xM .) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18. (18..( g + 3)}.56) D 4(.2.(0) + ^ ( 0 ) l z—>0 2."*(«!) ^ ' ~ [ ..(0) = . C.. [ ..
we obtain 2 i(»+i)[_i(. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.i ( « + i ) [ _ i ( . 4 r T (18. A __f7r \4 'J We rewrite the right hand side of the derivative equation again with the help of the inversion and duplication formulas and obtain _ (^)g/4(_l)j(g+l) e /> 8 /6^ _ p (/i4)g/4(_1)(g+l) _^_ . + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.18.3)}. isin{f(g + 3 .46). (18.sin{^(g+ 1)} = J .62) . (18.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^.)[i(g3)]! and ^i(.49). Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g ._l) From this we derive C. ~ ^ 1 5(9+1) 2i(''.53) in dominant order and insert the appropriate expressions for a and a from Eqs.i .+i)(°) [ _ I ( .(0) = *>i(„+i)(0> 2 .2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.56).+1iH]«=o = .sin{^( 9 . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .1/2)!. (18.6 ) } ^ — i t a n < — (q + 3) >. W . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(. _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.
_ e =*. z2 > z2.go) ^ (1) It follows that we obtain for the even function with q — go = 1..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. . 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.9. .63) Proceeding similarly with the second of relations (18.. (18. (18. we again obtain (18.63) and (18. (18. 18. and the left hand side of (18. 7.L ^ i . This is Case (1) of Fig.63) vanishes for q = go = 1. 18. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.53) becomes 8 m(I(. by the rotations E > einE = E. (18. Recall > the original Schrodinger equation (18. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. (18.396 CHAPTER 18. + 3)} = .64) about go — 3. (gg 0 )^±^ y e5?. In fact the left hand side of (18.11. . With a Taylor expansion about go the left hand side of (18. we obtain cos((g + 3)) = .. .. Thus we have to determine these conditions first..65) Expanding similarly the left hand side of Eq.64) In each of Eqs. z^ ein/2z.5.5..2) with potential (18.2. i.( g 0 + 3)  + • • • ~ (g .53). . equivalently.64) the right hand side is an exponentially small quantity. the vanishing of the wave functions at infinity).i . (18..63) becomes (? . 7 ..e.. 1 1 .. ^ ..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.9o) ^ cos  . the energy E is real.^ .^ >^_>m e J*.1.9.64) for q = g0 = 3. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions.1).
1/2. 2 c2>0. 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. or — < 6 < . Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied.e.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0.18.e. the resulting wave functions vanish at infinity and thus are square integrable. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — .e. (18. i. if TT/3 < 6 < 0. i. Now. r / c 2x 1/2^3 = exp{ ± i — —\. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.68) . we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0.66) z3 = \z\3em. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. —}. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. in the domain — TT/2 < 36 < > 7r/2.cos 36.e. (18.e. the case of complex E).67) Rotating z by vr/2.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. y{z)~exv\i[) 7T dV'V. we set z w r = \z\eie. i. i. 6 6 In the case of the inverted double well potential under consideration here (i. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. replacing sin 30 by " 0 + ^) 1 IT = .
3 The inverted double well potential with turning points ZQ. We do this extension with the help of WKB solutions. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. 18.Z\. (18. we match the WKB solutions . Looking at Fig. Anharmonic Oscillator Potentials V(z) 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) 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. We have to remember that we have various branches of the solutions y(z) in different domains of z.21) to + infinity and to demand that they satisfy the condition (18.e. (18. Eq.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).21)) were defined in terms of solutions of type A which have a wide domain of validity. and there impose the boundary condition (18.68) for c 2 > 0. This is the boundary condition also used by Bender and Wu [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.398 CHAPTER 18. i. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. Our procedure now is to continue the even and odd solutions (18. We therefore explain our procedure first.Z\. for z > Too. Our even and odd solutions y±(z) (cf. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. 18. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq.
291. 6. The distant turning point at z\ as indicated in Fig. 4 2 T ' i. . I [195]. t To the right of the turning point at z\ these solutions match on to (r. Messiah.70) where z < z\. p. *R.2. 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.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. z\ ~ .4) for E and ignoring nondominant terms) t o . f R . Vol. (18.71) We now come to the algebra of evaluating the integrals in the above solutions. 18.4 h + c z 2 1/4 < . equations (21). B. ^2? 1 2qc2 /2? . (18. h2 (18.4.18.3 is given by (using Eq. Dingle [70]. A J24 1 ? 2 • "^ •z h \—c z ~ — q h .2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. (22) or A.zi)/ \ 1/4 ywKB\ ) z — g/i* . In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. Sec.^ T rz\ 1 + yh* .=c z 4 dz 1 L2 9 + 2 4 1/2 x exp .*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) .69) The W K B solutions have been discussed in Chapter 14.^ 4 1/2 + icV 7T Jz + (18.B.70). p. Dingle [70]. i. We begin with the exponential factors occurring in Eqs.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. 291. z2h4/A > c2z4/2..e.e. and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\. ^ 2 4 •K + —(r.
211/2 1 \ /j.73) . Anharmonic Oscillator Potentials E± = exp = exp .40)) E± = exp _/^_2 f ± In A 2c2 .2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq. CHAPTER 18.2 \ 1/2 .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf. . In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.\qh2 + \z2h± .±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18.^ i ^ ^ c 2 .400 i. Thus the above factor yields '2c2\l/2 h4 so that (cf..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. (18.2.e. zi 2q 1 c 2 [1 . (18. (18.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g . Eq.2.40)). Eq. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). (18.^ I 2 2 1/2 zV = F 8c2 Jz [i .
\z2h4 + J Z\ nl/2 \M fZ . > » Returning to the even and odd solutions defined by Eqs.78) Inserting this into the solutions (18.71) and these into (18.20b).76a) P P r 2/1 2 " 9 ( .Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) . In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.2 The Inverted Double Well Potential 401 Thus at the left end of the domain of validity of the W K B solutions we have (l.21 1/2 .*l) (18.74) Comparing these solutions now with the solutions (18.18.20a) and (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.*l) (3y^(z) (18.(J*i) VA 0 ) = PVwKB (Z C.79) +S_(±)exp . h2 — — h2 is maintained.76b) apart from factors [ l + 0 ( l / / i 2 ) ] .2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .77) Now in the domain 2 — oo we have > \qh2 . (18.
Eq. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( . i. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J . (18.^ j ..^(q2 + 1) .84) Expanding about q = qo we obtain £(g.34)..e.82b) This is our second condition along with Eq.81) Inserting expressions (18./i2) = E{qQ. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.2. (18.5. (18.. (18.h2) + (qqQ)h + (18.65).402 CHAPTER 18. 18.^ § ( 4 g 2 + 29) + O ( ^ ) .. (18. E(q. (18. i.e./ ? = 0.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.85) + .76a) for (3 and /3..h2) + {qqo)(^pj = E(q0. we see that we have to demand that 5 + ( ± ) = 0.3. (18. . i / 3 ± .6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues. (18. h2) = \qh2 .68).
h2) of the relations KWKB (Z) =PVA(Z).h6/6c2 E = E(q0.36) of their Ref. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.~p{q. zo} means "to the right of the turning point ZQ" .2 . ft6 ^ = e. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/.yA{z). Since this is not without interest (e.ZQ) .g.l ) ] ! > + = ^ 2 . above) (1. S W K B ( 2 ) = PVA(z). o 2 + l)^(4. Example 18. !*WKB(Z) ~ 2 . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1.18. where the superscript {r. The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.3.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. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 . in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).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 . 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.h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o . i.e.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.g ( . h2).86) will be dealt with in Chapter 20.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q.
zo)/ \ qh2 2 ZO + z2hA 4 2 .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . 2 z 2 ) / 2 e \ . 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. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . f x Zn 4) h2 2. 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.] { hz combined with the particular constants contained in y ^ ' ^ g ./ ^ 2 X 1 / 2 exp(ifi.ic224 2 4 qh2 + i^fe 4 . Expanding the square root occurring in ^ W K B W ' ^ W K B ^ ) ' Igft2_I22h4+lc2z4 2 4 2 and we see that 1/2 = 7 ^ 1/2 2.1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation . the (approximate) solutions* 1/2 exp • ± i ( . cf. Anharmonic Oscillator Potentials where zo > z.404 CHAPTER 18. ." / 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.g In other words..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. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. Schwind [247].O. N.
i. In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e.J ' i J . Then .3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. „ / M 1 / 2 i . ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.3.1)! = ( 2 7 r ) 1 / 2 e .18. in fact. (18.zo).^ 1 ..3 18. The last expression can be made acceptable for q = odd integers by applying Stirling's formula in the denominator of the second of the previous pair of expressions. 4" ) (h?z2)^q+1) for q j£ odd integers.41) for yA(Z).13) of Bender and Wu [19].87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. But there are significant differences.yA{z) with expressions 27WKB i J'WKB ' that the factors p.g. we shall employ basically the same procedure as above or." / 4 ( ? « / 4 .kyil(.z ± ) = .89) ._ . (j .e.23)) appear quite naturally.DfJ»v we see Comparing now the expressions (18. as in the case of the Mathieu equation. (18. (18.88) + [EV(z))y{z)=0 (18.1 The Double Well Potential Defining t h e p r o b l e m In dealing with the case of the symmetric double well potential. h4 > 0. in Eq.5 "l _ o f .(r. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . We can reexpress these relations with the help of Stirling's formula.
(18. the previous equation (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 . Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. (18.z Fig. 18.z±) y = 0.V(z±) l{z z±fhA + 0[(z .406 CHAPTER 18. (18.95a) .4 The double well potential.i>5.91) becomes T>q±(w±)y(w±) = o(^)y.z±).90) V(i)(^)=0.91) (18. .94) (18. vW(z±) = ±Qch2. (18.93) w± = h±(z .
(18.99) 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) = 0.95b) with the equation of parabolic cylinder functions u(z) = D„(z). (18.3.e.18.93) for E into Eq.. 18. Thus again.100) .3 The Double Well Potential where ^ 407 d2 •« 2 2  d8. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)]. (18.87).96). (18.1. given one solution.. we conclude that in the dominant approximation q± is an odd integer..97a) U(z) = (zz±)2 + 0[(zz±)3}.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.. (18.2 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 +  . n = 0. qo = 2n + l. i.97b) Our basic equation.i 4 f W » = o. d2u{z dz2 + v+ 2~r u(z) = 0. is again seen to be invariant under a change of sign of z. we obtain d*y (18.95a) and (18. we obtain another by replacing z by —z. Eq.95b) By comparison of Eqs.2. Inserting the expression (18. (18.
e.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .101b) To a first approximation for large h2 we can neglect the right hand sides. 8hw (18.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18.101b). We leave the calculation to Example 18. The result is given by A and E(q.h2) . we observe that the solution y^iq. these equations can be rewritten as + (qz+z)A{z) = .101a). z) = yA{q.yA(z) are associated with one and the same expansion for A and hence E.408 Since CHAPTER 18. i.103) Looking at Eqs.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. i. Integration of Eq.h2.105) "8 + 4 ^ . (18. Vq = {z2+z2)— + (qz+z). h2. yA(q. h2 h4 2c' with g+ = q. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0.1 6 ) . ~h2.c W + 1) .102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. z) may be obtained from the solution yA{q. (18. (18. Aq(z) = Aq(—z) = Aq(z). but the solution is a different one. (18.^ A"{z) + 2c V2 *A(z) (18.4. (18.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). n± = V2h2.106) . z). Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.104) Both solutions yA{z). h2.e.^q(l7q2 6 4/i + 19) + (18. z) = yA{~q. h2.
A Differentiation yields » = ^^w (18. i.{z2 .103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials.114a) . . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.102).*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.qz+)2 .z2 ) N t.109) We observe some properties of the function Aq{z) given by Eq. <4q+2 — z — z+ Aq+2iAq+2j (18. for instance by componendo z z et dividendo. as a linear combination of terms Aq+2i. Thus it is natural to explore analogous steps.qz+) [2z2 . z • .4zz+q+ z\{q2 + 1)].106).18.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 . (18. ± 1 . 92 ZZJf (18. .112) From these we obtain. such as periodic potentials. .108) We wish to rewrite this expression as a sum y coefficientiv4. The first such step would be to reexpress the right hand side of Eq.Aq 1 + 2^± 2 Aq+2 (^M^f (18.2 z\) • Aq$4 £Aq + Aq—4. (18. J = 0. We know the first derivative of Aq from Eq. ± 2 .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .113) 2 .3 The Double Well Potential 409 Example 18.+ 2 ^ ± l + 2^±i + .(l + + 2  z y* .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . (18.103): . however.e. Solution: The structure of the solution (18. .110) Aq+2+Aq2 9+2 A. + Aq+2 + Aq Aq+2 . (18. .101a) with A replaced by A . .105) and (18. we reexpress A'q in terms of functions of z multiplied by Aq. (18. (18. = Aq+2i+2jAq..z\) + 2z(z .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. First.
q) = 0.q + 2)Aq+2 (18.112). (18. we obtain 5 K = ( ^ f ) [ ( 9 . leaves uncompensated on the right hand side of Eq. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.2\2 (z* .410 and from this = Hence with Eq.l ) 2 A . where the lowest coefficients have been determined above as (9. (18.aAg+S 12+ 16 (18.q2)Aq2 + (q. q + 4)Aq+4 + (q.115) and (18. (18.4Aq+2 + Aq+A Inserting (18. i. . cf.108).2A g + A g _ 4 ] . (q. (q.9=F4) = (g=Fl)(?=F3).e.+6 + • • • ] .<? +{q. In particular the dominant approximation of A is obtained by setting (q. QA?+4 _ 10 A + 6 .114b) .4 4+* + . Anharmonic Oscillator Potentials A 1 .116) [Ag4 .4 ( g . Since VqAq = 0.116) into Eq.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18. 1_4^±^+8^±±12^±^ +. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.q)Aq + (q. •••].q + 2) = 4{q + l ) 2 . ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.4A ? _2 + 6Aq . Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+. (18.2Aq + Aq4] 1. .117) 4(q + l)2Aq+2 Here the first approximation of A.l ) ( g .4)A„_ 4 + (q.4]j.118) 2s/2c ~^j[(?.q + 6)A<.33) and (17. Eqs.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.(0) V2 ' 2c K + h4+Ao .3 ) A .119) It is now clear how the calculation of higher order contributions proceeds in our standard way. (18.34).112) CHAPTER 18. _ 4 . A(°~) = Aq.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .115) (z*z%) 2 i2 (z 2 . (17.101a) the contribution R . (18.
121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.0)=0. z)\ = 2^A^ h2 '>z)± VA(Q.g) 4z+ {q q)+ 2r (9. h2 2y^ Age? h4 (18. i.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.0) = y^(q.yA(z) derived above are valid around z — 0.0)=0. (g.e. y'_(q.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.g) 8z+ (g.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.94) 8z+ (9.105). (i) _ / V2c\ [ ( g . A' (0) = q/z\) y+(q. Terms jj. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. g) 8z 4 .28.{^ h2.124) .4 ) ( 0 ) ( g . \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A. 9 + 2) Aq+2Az+ (9. of course. 9 + 4) ^+48z+ (18.g2)(g2.0) = ^ .h2.123) ^[yA(q. i.(0).g ( 1 7 g 2 + 19).h2.101a) the contribution Rq . h2. (18. thus yielding the next approximation of A as given in Eq. g . which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .h2.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. where Rq is the sum of terms left uncompensated by AW.z)±yA(q. in the domains away from the minima. g + 4)(g + 4.g + 2)(g + 2.g) Az.122) The solutions yA(z). (18. (18. *)] (18.e.h2.120) The sum A = A^ + A^1) then represents a solution to that order provided the sum of terms in Aq in Rq and Rq is set equal to zero. h2.h2. for i = 0.2 ) (0) (18. y'+(q.94)(g4. g .
Anharmonic Oscillator Potentials Our second pair of solutions. (18.91). 18. we > . We see already from Eqs.126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.2A y(w±).99) for z — z±. + 0(hl2). Since w±(—z) = —h±(z + z±).Evaluating the exponential factor contained in VA{Z) of Eq.93) and setting w± = h±(z — z±). Inserting (18. yB(z). (18. 2(9+) .126b) These are solutions again around a minimum and with yB(z). This means. (18.95b) that the solution there is of parabolic cylinder type. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. (18. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2). Moreover.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2).yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). the equation is — with differential operator T>q as defined by Eq.412 CHAPTER 18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. but w±(z±) = 0. (18. (18.^ % [i(?l)]!24(«.3.125b) It is clear that correspondingly we have solutions yc(z). we have w±(—z±) = —2h±z±.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).1 ) D (18.125a) = 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.23) with appropriate change of parameters to those of the present case. Bq[w±{ ^ .yB(z) is obtained around a minimum of the potential. and another VB(Z) ± 25^ch3wl + c2wi .We draw attention to two additional points.95a) and (18.
the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4.• • " .3 The Double Well Potential (cf. Eqs. (18. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . Here z± = ±h?/2c are the positions of the minima of the potential.18. and comparing with Eq. .^/w + ± 1 . (18.ZI and find that these are given by z+ + 0(l/h) for finite q (cf.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).frW.97b) ~ exp .129) . 24(9D[i(gi)]! hj (18.• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . (18. 2 An+z+e 4w+ (18. ein+z+e in+(z 22+(<?+i) . .172) obtained later. .146). .164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. (18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.z±)dz .« + ) 2 (9+1) e (18.127) Recalling that around arg w± ~ 0..147)).128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.i*l(«^ 4'"+^+ 4"'+v ( « . a= ^ . (18. (18. We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. Later we calculate the coordinates of turning points ZQ. . Allowing z to approach z+ in the solution IJA(Z). 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. Thus for h very large these turning points are very close to the minimum at z+.172)) we add here comments on this approximation.
The involvement of these WKB solutions leads to problems. Furry [102] . even above the turning points. We have to impose boundary conditions at the minima and at the origin. we naturally expect the wave function there to be similar to that of the harmonic oscillator. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.3. Since it is more probable to find a particle in the region of a minimum than elsewhere. 18. since. D. but it is clear that none of the above solutions can be used at the top of the central barrier.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . H. Bhattacharya [23]. S. they assume large quantum numbers. Rau [22]. de Deus [66] and W.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. The pair of solutions of type B is defined around argz = 0. J.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. K.(18. and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima. The solutions in terms of parabolic cylinder functions have a wide range of validity. g .7r and the solutions of type C around argz = ±7r/2. R. K. Bhattacharya and A. Various investigations^ therefore struggle to overcome this to a good approximation. S. 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. and this means at both minima.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. Thus 1 E .z+)]fr+V ' (18. basically. The next aspect to be considered is that of boundary conditions. P.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^.414 and CHAPTER 18.
135) Hence the conditions (18. (18.133) y+(z±)^0. as indicated there. (18. and y+(z±) = 0.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. The odd wave function then exhibits a correspondingly opposite behaviour.134) y'+{z±) ± 0. ^(*±)#0 (18.132) Fig. 18. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. We have y±(z±) = ^VA{z)±yA{z)]z^. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.5. 18. 18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. as indicated in Fig.18.132). an even wave function can also pass through zero at these points. y{z±) £ 0.136) .5 Behaviour of fundamental wave functions. y'_(z±) = 0. y(z±)=0.5. as indicated in Fig. However.Zii yB{z±) a ± a yciz±) (18.
9.416 CHAPTER 18.138) Using formulae derived in Example 18.136) as  M .141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i. (18.. (18. . Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. (1.«*{=(.„*{!<.. _ 8)} s .142) (<Z9o)J(l)^°+1)(l)1/2 .54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula.5.m Using Eq. .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 . Thus sin {^ +1) } = ^^>^30fe^i.136) the dominant approximations we obtain (cf.2 we can rewrite the second of (18. 7 .. Eqs. 1 1 . also (18. . (18. . (18. +1)} = 4 m.
we must also demand this behaviour here along with a nonvanishing Wronskian.z\.18. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. Since the type A solutions are not valid at the minima. 18.3 The Double Well Potential Thus altogether we obtain 417 ^ • .144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. V(z) — • z E=V Fig.» ^ B ^ ' .6 Turning points z$. y+(0)^0.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin. (18.5 (after determination of the turning points) we rederive this relation using the WKB solutions from above the turning points matched (linearly) to their counterparts below the turning points and then evaluated at the minimum. We deduce . y'_(0)^0. 18. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. y'+(0) = 0.1 " <813 1 4)  In Example 18.3. Hence we have to impose at z — 0 the conditions j/_(0) = 0. We do this with the help of WKB solutions.
(B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18.\qh\ + \h%U{z) (18.144).145) \qh\ + \zW Using z+ = h2 /2c. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. As a consequence.^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. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18.e. i. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. In the domain 0 < z < ZQ. to the left of ZQ where V > E. ZQ) means "to the left of 20" . Anharmonic Oscillator Potentials from Eq.148) ("f * "The superscript (I. one finds that . h8 25c2 (18. (18. + £\h%U(z)=Q. We also note that the height of the potential at the turning points is h8 qh V(z) 20.418 CHAPTER 18.e.147) Since the minima are at z± — h2 /2c with /i 2 very large.
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) . 20 dz \qh\ + \h%U{z) .3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.149) In order to be able to extend the even and odd solutions to z = 0. . (18.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 >'^. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1. we have to match yA(z).18.^ e therefore have to consider the exponential factors occurring in (18.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .^ j O e .* 4+ .(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.148) and (18. )+^*+)2lr} rnD h\z'2)}z Q )\ = .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z . (18.^ + ) 2 + igln2(zz+).i (2q ln (18.
In a corresponding manner — i.125a) and (18. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18.152b) [\(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. using Stirling's formula (and not the inversion relation) — we have —(1. Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. of course. l/2 ~h\U{z \(zz+?h\ (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. We see therefore.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+.154) . 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. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. However. This is the aspect investigated by Furry [102]. t h a t the Stirling approximation is amazingly good even for small values of the argument). Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.125a). the results necessarily require adjustment or normalization there in the q—dependence.e.ZQ) .126a).
158) Applying the boundary conditions (18. 27T1/2 9/4 1VA{Z).155) (2z.e.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.104). Returning to the even and odd solutions (18.159) 7 i(«i) r(9^3) ^2~ j(9l) .154) with the typeA solutions (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .157) Since the factorials [^(q — 1)]!.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum.143) T^^*)) 27T (18.e. we obtain in leading order (i. (18. as the results also seem to support.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. . 7 of t h e matching relations y^NKB\z) i. Comparing for z —• z+ the W K B solutions (18.157) is used in Example 18.n('>2o) 3/WKB(Z) =1VA{Z)I (18. (18. However.K?+1).e. The relation (18. (18. linearly matched) W K B solutions.156b) Using again the duplication formula^ the ratio of these constants becomes 1 7 (h2+)i/2(2z+)i [faW e 2n+z+ — Aft 2 72 (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.18. and is shown to reproduce correctly the result (18.153).103) and (18. (18.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. \hl.5 for the calculation of the tunneling deviation q — go by using the usual (i. .99). this result is somewhat imprecise.
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 (18.148).164) / 2 (0) z ca [(1 + k2)E(k) . n + uY 2c „ ^6 8V2c2 (18. 3G2 12V2c ^See P. D..05.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.k'/2. (18. .161) we can rewrite the integral as r I2{z) = = dz^{a2 .163b) The integral hi?) ..162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18.. p.z2)(b2  z2).149) near z = 0. Byrd and M. 213. p. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\.422 CHAPTER 18.163a) (18. Friedman [40].19. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18. (18. . F. G2 K [i 4.iV2 ! i . formulae 220.t The elliptic modulus k (with k' = 1 — k2) and an expression u appearing in the integral are defined by b2 1 U kz = ^ = l + u so t h a t = u=S^q hi = GV2q. 60 and 361.
± ' 3G 2 ' T ( ^ (18..1 " e " O T (18 169)  Y^V / 4 tfo.3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . [167].e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref.( ft 7. .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 . Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 .1. It then follows that (again in each case in leading order) '2_hl)9/4 ^ ^ * > l * = ° ^ 23g/S(_g:)t/4(!)g/4eg/4 "1^Using Stirling's formula we can write this e ( 18 . where the result is shown to be (with q ~ q0 = 2 n + l .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n . . /2n + l 2 ± — . n = 0. n. ^ ^ .) = 2 \.165) in agreement with Ref. which implies small G2. ± .2. [4] misses an n—dependent power of 2 in the result.. (G\ .18. 1.1)]' Gfo..3 )l ! .§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. . The nontrivial expansions are derived in Example 5..
(18.3 concerning the exponentials.6 Eigenvalues and level splitting We now insert the replacement (18.3.. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.1. we obtain for E(q0.173).424 CHAPTER 18.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.. 2 . h2) the split expressions 29o+1/i2(—W2 h6 (18.i ) ] ! e ~ s A ' .»W.157). Inserting the expressions for h+ and z+ in terms of h. (18. we can therefore impose the boundary conditions at z = 0 by making in Eq. 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). Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec. . 18.174) . (18. and obtain ( * .143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. (18. .172) into Eq. (i*™) We obtain the energy E(q. . ..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.175) ~E0(q0. (18. (18..n = 0.h2) and hence the splitting of asymptotically degenerate energy levels. with the help of Eq.^ W ^ / 4 t o .143) with qo — 2n + l.3. 18.h2) + (qq0)^=.106). (18. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18.
V2^\ \n+ i . which evidently supplies some correction terms.. i.174) and (18. Combining Eqs.3 The Double Well Potential where £0(<7o. i n U n / ! Thus AE(q0. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 .1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.fr2) is given by Eq.. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.h2)E+(qQ. the following expression for large q ~ 2n + 1.143) with (18.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .e.q2 7. i. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.106).2 = 2 . (18. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB. .18. 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 / » : = \ .^ 2^ h* .e.e. T \ *n ! 1 ' (18..e.2.157).n = 0. 2h4' i.e.1.. 1 u2 c 2 (3 9 2 + l) ^ « b . 1 7 425 2 . The result (18.178) "In S. the pure WKB result of Banerjee and Bhatnagar [14] (i. (18. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. for finite h2.7 =( e N n+1 .h2) = E_(q0. K. the level splitting.
2X2 V(z) = \z2tL) . h2)\mo=i = . and A there with c 2 / 2 here. 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. we identity the parameter k there with ft. Bhattacharya [23]. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. This. (66). however.2 / 7.180) h8 E0(qo. as is also explained by Bhattacharya [23]. K. (18.426 CHAPTER 18.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). 2/z4 and 2c2 respectively (see Eq. H. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. ** Comparing the present work with that of S. the Furry factor corrected WKB result follows automatically.4/4 here. h4 and c2 replaced by 2E. h = 1.178). that if this symmetry is taken into account from the very beginning.3)). A>0. we would have obtained the result with E. Furry [102].6 \9o/2 e~h&l^\ with (m 0 = l) (18. We see therefore. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. (18. (18.181) "W. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. 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. and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18. Then the splitting A ^ W B of Eq.h2 <> — h2. Eq. . The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. (18.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq.^ 2 + ^h2 If in addition the potential is written in the form A / ." Of course.176) together with Eq.6) to yield an improvement of WKB results for small values of n.
Banerjee and S. P. D. The comparison with Eq. Patrasciociu [110]. Jentschura [293]. See R.18. (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.88) h4 = 2/x2. m 2 <• p 2 / 2 .34) and (E. 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.§ 18. M.11). D. Gildener and A.* A similar correspondence is also found in the case of the cosine potential. (2. (18. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. J. W. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. Bhatnagar [14]. these are presumably not of much interest in physics (for reasons explained in Chapter 20). Liang and H. formula (4. Friedberg and T. Equation (18. J.—Q. H. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. Eqs. Mansour and H. . ^To help the comparison note that in J. MiillerKirsten [186]. J.c2 = A/2. K. Lee [98].7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. first paper.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods. however. so that by comparison with Eq.^ *See E. §J. ZinnJustin and U.3 The Double Well Potential 427 a form frequently used in field theoretic applications.3. ZinnJustin and U.182) implies AlE(l.W. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. We considered above only the symmetric twominima potential. M. D.181) therefore implies the correspondence rj1 <> A/2.15).h>) * 2V2^^^y/2e^3/3A.^ Thus for the case of the ground state Eq. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. (18. Jentschura [293]. and g2 is given as g2 = r]2/m3.
as some relevant references with their view we cite papers of Harrell and Simon [129]. though incomplete list of references in this direction has been given by Garg [105] beginning with the wellknown though nonexplicit (and hence not really useful) ground state formula in the book of Landau and Lifshitz [156]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion.428 CHAPTER 18. As references in this direction. P. Ashbaugh and Harrell [12] and Harrell [130]. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. Pradhan [182]. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. 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. Perturbation theoretical aspects are "See e. The double well potential. There is no reason to view the anharmonic oscillator differently. we cite papers of the Uppsala group of Froman and Froman [100]. Mahapatra. since — as we shall see in Chapter 20 — these nonperturbative contributions are directly related to the largeorder behaviour of the perturbation expansion and determine this as asymptotic. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. like those for anharmonic oscillators. can be found in an article by Coleman [54]. mostly in connection with instantons. has also been the subject of numerous numerical studies. Martin. B. 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].g. B. though not exclusively numerical. [19] — that the expansions considered above are asymptotic. as he discusses. 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.[131]. [132] and Loeffel. 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 both symmetric and asymmetric form. In fact. A reasonable. N. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. Very illuminating discussions of double wells and periodic potentials. Santi and N. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. . Simon and Wightman [178]. for instance.
183) S o l u t i o n : We have k2=1^. Example 18. Thus k' and hence k' = v ^ ( l . Saxena. 1+ u : Gy/2q.187) We now reexpress various quantities in terms of u. which is assumed to be small. fc ~ 1 .190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18.. 2 a./4 E{k) = 1 + • l n In (18.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' .189) =2u 2u2 (18.188) T h e following e x p r e s s i o n a p p e a r s f r e q u e n t l y in t h e e x p a n s i o n s of elliptic i n t e g r a l s . q ~ 2n + 1.2G^2q + 4G2q .3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. X +2 + ' ^ (18. Srivastava and Varma [24] and Bhatacharya and Rau [23]. We have 4 ~k' 2u(l . (18. Cooper and Strottman [221].u)1'2 = V2u{ 1 .185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. T h e r e f o r e it is convenient t o deal w i t h this here. k! (18. + (G 3G2 3^ + 2lnU 2 + i(2n + l)ln(f)+I(2n ~~ 3G 2 ' 2 l)ln(^±i q 2 1 ^ln q/ 4 2n + l (18. [122] as 13 1. Bender.192) 'F 12 16 .186) (18. 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.18. Datta. We obtain the expansions from Ref.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .\G\ + • and 1k2 =2Gv/2q4.G2q + .5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. Biswas.^ + • (18. (18.f + • • •) 2u\ . Gutschick.
) J 4 (18.u 3 ( u .3u(6« . (18.2{«(1 .3 « .3 u ( l .4.13)} + 12u 3 (u .6w2 + 13u} + 12u 3 (« .195) we obtain E(k) 1 + ln uK(k) .1)(1 .13)} + ~u3(u .u)} u{u(l .3 .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .«) + 2} + V [ .e.u) + 2}] 1 H 1 o u(u .193) 2u(l .u) + 2} + i u ( u .430 and CHAPTER 18.196) Now consider the last line here without the factor 2.{ M 2 ( 1 .a = (l«)(jl 2u/ 2 ) .1)(1 . (18.194) ln '^J + 2 In i/2u )+H l + 2u(l .u ( l .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3 u ..3).u){24 . (18.« 2 + 3u] In [ .{l u ( l u) / W 2 + 2} .1)(1 .13)} + ±u3(u .u)} .3u(6u .3) 1 + In ^ .u){24 .u ) (^){« + 5^ 1 )} + 5.u)(u .^ L ) H[(i _ U ){4 + 3u(l .18u 2 + 39M} + 12u 3 (u .3u(6u .u){4 + 3u(l .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .u) 2 {24 .— ) .1)(1 .3u(6u .3) .6 in the denominator and in the last step pick out the lowest order terms in u (i. Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + . .3) .195) With (18.3) = = ~ .197) .3 y .u) 2 {8 .u 13 \ 1 ln(± k> fc'2 +  (18. up to and including u2): u(u . 8 + 9u 2 . 1 3 « 2 = .u){24 .u) + 2u} + i u 2 [ 2 + (1 .« ) ^ { 2 4 . 8 .u)} + — u ( u . 8 + 6 u 2 .194) and (18.u ( l .13)}.
(18. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .it)} . V^LJ 4 2 16 ^ . we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .2 l_ln(1 JL\ V {2.196).«){4 + 3u(l . we have.195) with (18. (18.3).4u .197) we now obtain 4 1 • E(fc) .zo) HWKB normalized _ ( h* y^eM. Eqs.uK(fe) ~ 1 + In I — = Mu2(3u. Inserting here from the above expansions the contributions up to and including those of order u2. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . l2 = ^ ( l + uf/2[E(k)uK(k)}.18. 2 .128) and (18.] in (18.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs. ' 8G 2 q = 4 2 q ln f 26/2 \ o .198) 9 «. where w±(z) — h±(z — z±).e.3 The Double Well Potential Now consider the bracket [. (18. i.148) (for particle mass 1/2 and in dominant order) is given by H.3) .200) Remembering that u = Gy/2q.u H I .199) We now return t o the expression on the left of Eq.156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf. / to + 3GH~ \ 8 . .e.2 (18. i..4u . [(1 . (18...2 /o 2 431 (18. + iu^Zu2)^~u+ — +0(u3) l .196) and (18.6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z). From (18.173) t o be evaluated.128) and (18.^HfHGH Example 18.2{w(l u) + 2}] = u(3u2 . VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .
35)).7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. (26. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. Oberhettinger [181].146) and (18.hA/2 = h^/4 this implies .143). Example 18. i.*o)/ ^WWKBOO (18. pp. 82.3)]!23<«D [5(91)]' >A.f hi\ VWKB. W. We start from Eq.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+.e. (18. 93. 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 .g.l)]![i(0 .147). (18. Magnus and F. are given by Eqs. Then ((r.e. (18. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. Eq.zo). 1 7(l./ . i. normalized = We see that the normalization constant is independent of a quantum number. Solution: The turning points at zo and z\ on either side of the minimum at 24. 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 [ .Aw) = s/2i 2("!) V (41) and [(9 . ° \Edzy/\EV(z) V(z)V4 VWKB.432 CHAPTER 18. (18.** oo / dwD\.204) . and obtained as Eq. oo .158). y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. by using the periodic WKB solutions below the turning points. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.
(18. A.g.3 The Double Well Potential We also note that d dz 433 y±{z) >2+ 1 F = 7 sln h%U(z) 1/4 f X —2 cos 7 2 2^ + 1 \ 1 / 2 71 .. this implies sin cos sin cos Thus e.18.132) and (18. JV = 1. \h\U{z) 1/2 .g.( . 6. 1/2 .210) Choosing the point z to be z+.207) (18. Sec.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 (18.2.^r + + .206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18. Then at any point z € (zo.e. approach the minimum also from the right.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .2. i.e. 7T\ + ^ 4 7r\ dz .3. of course..e.. We could. from z\. i.r + ^=co S [r. from ZQ. z{) we expect tt (18.211) 7 z l .6.212) tt See e. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1...i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds.. Messiah [195].J^Wj +(18.iK h\U{z) + / 4 (18.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left. . i.205) We now apply the boundary conditions (18.
e. (18. the level splitting is given by AE(q0. where u> is the oscillator frequency of the wells.27 9 .434 CHAPTER 18.213) The integral on the left can be approximated by 1. Banerjee and S. i. Landau and E. 7r"y in agreement with Eq.215) we can insert for 7 / 7 the expression given by Eqs. and cot< (q + 1) '!}» for q = 50 = 1. The prefactor 2\/2h2/n is. they become c o t { ( .215) which results from the factor of 2 in front of the sine in the WKB formula (18. (18. in fact (with our use of qo). (18. S.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18.146)). We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). M.Zi / r. (18. In Eq. D.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 .214) {«+"i} 0 for q = go = 3. K. (18..216) Inserting this into Eq.210) assume a form as in our perturbation theory.5. P. + l ) j } ^ 1_ 27' (18.90 — =F— for q0 = 1. 7 . (18. K. 1 1 . and L. . 2IO/K. . (18. 5.3.2 . .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 (18.215) we can expand the left hand sides about these points and thus obtain .149)..143). W "Cf.215).209) and (18.217) 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.205). We now see that Eqs. ' /^i . 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.148) and (18. Bhattacharya [23]. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. Bhatnagar [14]. Lifshitz [156]._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . The formula thus agrees with that in the literature. 9. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). .
Case [45].* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. MiillerKirsten [9]. W. in an early investigation Case^ showed that potentials of the form r~~n. 435 . Nonetheless it was thought that a certain formal analogy could be seen if the field theory is supplied with Euclidean spacetime concepts at the expense of sacrificing the interpretation of the interaction in terms of particle exchange. H. 'For a review from this perspective with numerous references see H. M. Morse and H. J. M.1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state. W. n > 2. Aly. 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. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. are not as troublesome as one might expect. also in P. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. Feshbach [198]. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Giittinger and H. the investigation of singular potentials in nonrelativistic quantum theory was motivated by a desire to obtain a better understanding of the (then presumed) singular nature of the nonrenormalizable weak quantum field theory interaction. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful.Chapter 19 Singular Potentials 19.* However. However. f K .
Liang and Yunbo Zhang [189]. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. W. In fact. Singular potentials arise in various other contexts. for instance. . This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. Gubser and A. MiillerKirsten and Jianzu Zhang [226]. Tamaryan. Callan and J. Hashimoto [124]. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. J. Park. H. (19. R. S. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation.2) § C .e. J. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane.2 19. Recently. Maldacena [41]. K. N. H. some also permitting further research. S. It is therefore natural that we study this case here in detail. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. J. MiillerKirsten. that with the hyperbolic cosine replacing the trigonometric cosine).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. S. § 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).436 CHAPTER 19. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases.Q. D. 19. Manvelyan. In the concluding section we refer to additional related literature and applications. G. W. which are always worth studying in view of the insight they provide into a typical case and the didactic value they possess for this reason. Singular Potentials theories in mind.2. D referring to Dirichlet boundary conditions) in 10dimensional string theory.
Q. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. It is clear that we draw analogies to our earlier considerations of the periodic potential. J.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. W. This is rarely possible and therefore this case deserves particular attention.( I + 0. and so =4 5 and n ez = r = ~ = e^4fir. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. Thus we can expect to find solutions very similar to those of the periodic case (with — for instance — cos replaced by cosh) and with the parameter v given by the expansions we obtained previously. h2 = ikg. . 8. In the following we study the solutions of Eq. R.28) below). 19. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. (19. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. The following substitutions are advantageous: y = r1/2cf>. using the method of Dingle and Miiller of Sec. MiillerKirsten.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation.3 below) g2 is negative and hence h? is real for E > 0.g. (19. (19. We observe that with the replacements p2 z^iz. h = ei^y/k^.19. and as we have in mind in Sec. J.7 that we employed also in previous chapters.mo = 1/2 and h = c = 1. Manvelyan.3a) In the case of the attractive potential (as in the string theory context referred to above. \[\ in the correspondence is nonintegral. It may be noticed that since I is an integer. Eq.4) first for small values of h2 and then for large values of h2. r = 7 e z . ig h \j g (19. This has advantages compared with higher dimensional cases. (19. Liang and Yunbo Zhang [189]. H. The radial Schrodinger equation then assumes the form 2h2 cosh 2z .3b) 7 = % j.
1.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation.2.i l .5) so that Eq.Nv(w). (19. VahediFaridi [10].4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .11) We follow here largely H.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. 8. W. — Zv+i). J.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. —Zu — (Z„_i + Zu+i).e. .438 19. (19. First we make the additional substitution w = 2/icosh z. w 2 . Proceeding along the lines of the perturbation method of Sec.2 CHAPTER 19. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. H.2 A / i 2 . MullerKirsten and N.I . where these are the Bessel function of the first kind. Aly. (19. and H„ (w). i. (19. The zerothorder approximation = Zv[w) (19. H. (19..7. dwA w dw { wA ) (lg.Hv (w).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.6) terms amounting to 1 „ 2 d2Zu A V (19. (19. Eq.9) leaves uncompensated on the right hand side of Eq.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. The solutions Zv(w) are written Jv(w). where £>„ : = —_ + i _ + J l . 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.
14) (19. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z. it is more convenient to use these relations in order to rewrite Eq.v + 2)*{u + 2.u + 2)*Z„ +2 ]. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ .12) (19. of course. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. = §.10) is now particularly simple: i?i°) = h2[(v.16) ( I / .i/) = 2A.19.!/+ 2)*4°J 2 ].2. I / ± 2 ) = 1. The terms (19. and this means in (19. where the starred coefficients are defined by (19. v .6). v + 2)*(v + 2. where We observe that DUZV = Q.i/4)*Z 1 / _ 4 +(u. However. wz The expression (19. v2)*{y2.13) (19.15) (19.13).17) Thus a term fiGu+a on the right hand side of Eq.u (19. z . v .13) therefore lead to the first order contribution jfU = h2[{v. (19.18) Now </>(°) = Zv left uncompensated Rv . (19. v + 2)*Z„ +2 (19.20) 2)*ZV_2 + {y..21) +(i/.2 ) * ( z / . We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). . . (i/.u + 4fZu+A}.10) in terms of functions G„ defined by Gv+a = —nZ v+a .2)*Zi.10) as a linear combination of various Zv. when a or 2v + a = 0. (19. v + 2)Gv+2].2)G„_ 2 + (y._2 + {v.2 ) * i ^ 2 + (!/. v)Gv + (y.
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.2. Ri.v + 2y{v + 2. i/ ± 2)* (i/ ± 2. Evaluating t h e first few terms. .455a + 1291a .v) + hA[{v. . .v)] (19. p4(±2) = (i/.4)(»/2 . v + 2j) (19.1169)/i 12 . v — 2)*(v — 2. i/ + 2j).2 + 2 9 ) ^ 64(^ 2 .24) P2i2(2j + 2){v + 2j + 2.v) + {v. p 0 (0) = 1. v + 2)*{v + 2. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j.g. subject to the boundary conditions P2i(2j) = 0 for  j  > i. we obtain 9 h4 = a 2(al) 3 2 (13a .2)(v + 2j . in Eq. (19. i/)] + • • • . we obtain t h e expansion .440 CHAPTER 19. (19. v ( 4 5 a . P2*o(0) = 0. . .4 + 58. .23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . and p 0 ( 2 j ^ 0) = 0. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . i/ ± 2)*.v2)*{v2. (19.25)/i 8 32(al)3(a4) . i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.26) iy — 2.17)). (I/. etc. v) +(i/.I/±4)*. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9..v + 2)*(i/ + 2. .i/±2)*.j^0 (19. we obtain 0 = h2{v.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . j=i.22) where p2(±2) p4(±4) = = (i/. Adding these terms and setting the coefficient of Gv equal t o zero. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v .I/±2)*(I/±2.
3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. (19.An(u + n). We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . Thus v is real for small values of  h21.2.cosh 2z)4>v = = 2h2A(/>„ . (19.4).^ + e±(v~2)z. £>„ = — . apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . (19. 19.29) sinh vz or e±uz. v .32) (19. we can rewrite the latter as j ~ .i . The solutions 4> of the modified Mathieu equation are now completely determined.30) £>„&.v + 2)<t>v+2]. 2 .7) into Eq.h2[(j)u+2 + 4>v_2] h2 [{u. = 0. Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv . (19. Substituting (19. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z.19. so that J2 (19. v)(f>v + {v.cosh 2z)4>. Thus to O(0) in h we have (jr ' = 4>v — cosh vz.33) = = cosh(^ + 2)z + cosh(^ — 2)z. . and is therefore real for both cases g2 > 0 and g2 < 0.2)<t>„2 + (v.v2(j> = 2/i 2 (A .31) It follows that Dv+2n4>v+2n = 0. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. ±2. (19.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.34) .
F. 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. Meixner and F. 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.2.h).I/) = 2 A . . > These solutions correspond in the periodic case respectively to the solutions and me^.J/±2) = . v ± 2) etc. Arscott [11].4 Notation and properties of solutions We now introduce standard notation as in established literature. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. In order to avoid confusion arising from the use of different equations. 8. CHAPTER 19. The solutions in terms of cylindrical functions are written *J. Schafke [193]. 19.442 where (I/. 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. (19. M.7 — we refer to Meixner and Schafke [193]. (19.h2i *=1 J2 j=i. W. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. we prefer to adhere to one equation with different solutions. Singular Potentials (I/. The use of the symbols (v.h) = <(>„ +Y.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.37) where p M (±2) = (I/.h).J7t0 p2i(2j)<f>u+2j.h).I/±2)*.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. > (f> with </v = sinh vz — Seu(z. • (19.1 . etc.38) (f> with (\>v = exp(vz) — Meu(z.
h) are therefore proportional to each other as a comparison of Eqs. 'See e. As emphasized in Chapter 8.MP(z. 17. as we remarked earlier. (19. W. i. For the relation of Hankel functions used below.(2hcosh z).45) T Below we frequently write the normalized solution as in J. see this reference p. (19. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. 443 (19. (19. Meixner and F.g. — oo cv2r{h2)e^+2r>.h2) oo = Y. 4>v = Hv (2h cosh 2) * M^\z. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z.h).39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. h). (19.h).h). h). Mi. Also from the expansion of Jv{z) in rising powers of z. h) = exp(inwn)MP(z.40) Cev{z.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z).h) = exp(vz) + Y.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. which. i.e.41) i=l We see immediately that Meu(z + niri.19.h)±Sey{z. ' (z.44) implies.e.40) does not. .h). whereas the — as yet — unnormalized solution (19.43) and hence similarly M^\z + inir. (19. p.h) and oo i M^\z. {2h cosh z) • MJ?\z.42) The solutions Meu(z. h). 4>v = Hi. h) = exp{vniri)Me„(z. <f)v = N„(2hcosh z) MJ?\z. W. 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) . Me„(z.44) (19. Oberhettinger [181]. (19. h) = av{h)M^ (z.42). 16. Schafke [193]. this normalized solution possesses contributions exp(i/z) in higher order terms. Magnus and F. as Mev{z.
r=—oo (19. ^ . i. oiv{h) = (19. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. Mi 3 . we have the following for a change from v to — v. 2M™ = eiuvMi3) + e~ivlrM^\ .8. p. The functions Me±v{z.e. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. Meixner and Schafke [193]. the first relation is obtained as follows: 2M<£1 = M^l+M^l. MJ>>(0. i.h) on the other hand can be shown (cf. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.43).46) H^z) = e^H^(z).47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z.49) Inserting this expansion into the modified Mathieu equation (19. These points are important in our derivation of the S'matrix below. . 130) to converge uniformly for all finite complex values of z. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).h) can be shown (cf. p.50) E. As mentioned earlier.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. Singular Potentials where clearly MeJQ.h) Using further properties like (19.exp(TM"r)MW(. h) = M™(z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.g. Meixner and Schafke [193].444 CHAPTER 19. s (19. h).h2) = ] T cv2r(h2)e{y+2r)z.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. For later essential requirements (i. H™(z) = e^HfXz).e.48) The series expansions of the associated Mathieu functions M„ (z.h2): oo Mev(z. h) . (19.h) .e.
2 [ A . 117. Eq.* For 'Actually.19. . 2r (19>51) 2r For ease of reading we give the steps in rewriting this.51) we have 2r (19.^ p 2 2 ^ or c 2r+2 2r2 1 (19. p.(v + 2r)2} / z . see J. Meixner and F. (39).^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . Schafke [193]. Alternatively taking the inverse of Eq.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . eg = c~v = 1. (19.54) 2 C2r 4 ^ h. .52) in agreement with Meixner and Schafke [193]. \) /.53) /i2[A(^ + 2 r ) ] .T22r[\ A .^ + 2r + 2 ) 2 ] .^ C 2r+2 (19.[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. p. \ . W./ i 2 C ^ . + 2 r ) 2 ] . h~2[\ . Thus now c 2r . 122. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(.(/^.
1 for the evaluation of a typical case): hb + 0(hw). (19.[v^. H. and also R. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .Q .55). H.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. + i).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. Aly.5 D e r i v a t i o n of t h e S . MiillerKirsten. M. . Manvelyan.^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). ._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(.446 CHAPTER 19..l ) 19.+0^"> < 19 . Here we insert for A the expansion (19. Solution: We put r = 1 in Eq.+4)!ll_.. J.C + l) 3 (i/ . Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. 128(i/+ l) 2 (i/ + 2 ) ( ! / . W.27) and truncate the continued fraction after the second step. t For this purpose *We do this in the manner of the original derivation of R. but follow our earlier reference H. VahediFaridi [10] (this paper contains several misprints which we correct here). MiillerKirsten and N. H. + 0(/. J. (19.( " + 2)2] ~ h2 h. Spector [257].55 ' Example 19. J .2. Then di 2 1 t" + 2(^1) + • • • .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. W. Liang and Yunbo Zhang [189].52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq.
I / + 2 2 l\TT +E ^ i\ E j=i.j7L0 P2i(2j)exp i[v + 2j + 2 2 . p.56) where by Eqs.57) ^ 1 . In the case of the repulsive singular potential here.19. (19..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 .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. we have W r = ip + ut — 7r/4' . Oberhettinger [181]. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. W . . \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const.59) which tends to zero with r. Magnus and F. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19. and considering the propagation of this wavefront.3a) and (19. (19. 22. Then y reg = = r rV2M®(z. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. and then the continuation of this to infinity.h) 1/2 w) (19.
^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).j¥=o Ki(2j)exp(Tmj) (19. (19.3a) and (19. these solutions are seen to have the desired asymptotic behaviour for r —• oo: » 1/2 y± e±ikr e x p ml 1 nk ) 1 + ^2h2i i=\ Yl —i. Singular Potentials so that when t — oo : r — 0. 17.#.V + § 9 2 > 4hh* .62) Also with Eq. (19. 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.h) = exp(Mr^)il4 4 ) (z.61) In fact. Now from Eqs. We require therefore > > the continuation of Mi.A1"—2 z '+"2 A~~ n or rA . '2"4 . We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. 4%r g >0 4r^r ' rg See W. p. h) through the entire range of !&{z).448 CHAPTER 19. Magnus and F.j^Q p2./i)./l ~ 2 ~ 2z + " 4 > 0.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.3b) ln • 7T Q"*4' r fg ° Vk (19. AV . but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.h) can be shown (cf. 178) to be convergent for  cosh z\ > 1.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1.60) Using the above asymptotic expressions for the Hankel functions. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. 2h and the square of this expression implies 2 2 ~ ' A. . Meixner and Schafke [193].A14A. '(z.(2j)<gH • (19. p. Oberhettinger [181]. The series M^ (z.
rl)(r2 .2 The Potential 1/r4 — Case of Small h2 Hence (r2 . Thus we choose . and in order to maintain this symmetry. . (r — r+)(r + r+)(r — r_)(r + r_) > 0.19.38) or (19. A suitable set is the pair of fundamental solutions Me±v defined by (19. 19.49). i Im z z = In r/r0+i7t/4 . (19.1. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. Originally we chose the sign of m / 4 as in Eqs.3b) and (19. 449 This condition. the domain r_ < r < r+ — as illustrated in Fig.63) Thus there is a gap between the two regions of validity — i. i. is satisfied only for r < r_ < r__ or r_ < r + < r. These solutions converge uniformly for all finite complex values of z (cf.62).1 The domains of solutions.e.3a). r 7r for 0 < r < ro.r2_) > 0 with r£ = rg(2 ± >/3). (19. Meixner and Schafke [193]. p. 130).ijt/4 Rez .ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. which has to be bridged by using another set of solutions. (19. we have to assign different signs to the imaginary part of z in the two distant regions. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. 19.e. This interchange in the > solution My interchanges y reg with y+. Since the real part of z changes sign at r = ro.
131). these relations can be reexpressed for one and the same point z = —iir/A: rl/2[aMe„(z) rl/2[a'Meu(z) + + /3Mev(z)]z=_in/A p'Me^(z)]z=_i7r/4. Thus. as just explained.v{z)\ . (3 = a . (19. Hence we match the right hand sides of Eqs. at this junction we require also the imaginary part of z to change sign.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.e. At r = ro the real part of z vanishes and then switches from negative to positive.z=—in/4.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.l/2Me_i/(T. i.B^0.(2)]z=w/4 + 0Me. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. to r^2[AM^ + BM^\ A.^)] dr z=in/4 dr (19.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. (19. Singular Potentials Then starting from the region r ~ 0. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.66) a = (3 .450 CHAPTER 19. h) (as one can check or look up in Meixner and Schafke [193]. .v(z)\ —m/4) + ^[r^Me. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. p. h) = Me^u(—z. i.[ r l/2 M. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.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_.
= aMev(z) + 0Mev(z). and the second of Eqs.3a) and (19. dr BM^(r)]. (19.68) From Eqs.69)) cancel out in view of the nonderivative relations. for instance. Thus we obtain from Eqs.19.g]:=f(z)*&g(z)WZ) dz dz Thus.71) We now determine the coefficients a. (19. (19. (19.64) the relations M&Hz) jz[M^\z)] and from Eqs. (19. we use the notation W\f.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19. (19.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . Then.70) by Me'_u (the prime meaning derivative). (3.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. and we are left with relations expressible only in terms of z.69) Thus if we replace in the derivative relations of Eqs. A and B. (19.70) by MeU and subtracting . the nonderivative parts (from the first term on the right of Eq.65). (19. (19. multiplying the first of Eqs.3b) we infer that (z = l n r + const. (19. with the replacements of Eq. (19. ^ ] = ^ + ^ ^ .68) the derivatives with respect to r by this relation. (19.65) and (19. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.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.
. we obtain the first of the following two equations. the leading terms of the respective cylindrical functions. 7T (19. M^4)] + W[MJ>3).Mev] = = = av. av{h) = Me.M^] W[Mev. 7T [1.M^} W[Me„.46). h).^ 47 9? MJ>4)] MeV] [3. p. Meixner and Schafke [193]. (19. W[MJ?\ M^] W[Mev.^ ^ a „ Q _ . .71) we obtain in a similar way A B = = W\MP\ Me. .Mevy W[M?\Mev] W[Mev.47) we obtain (av defined by Eq. (19.46)) W[Mev..73) We now use Eq.] W[Mev. TV W[Mev. h)/M^\o. B = J 2 L _ e . and the second is obtained similarly: W[M?\Mev\ W[Mev.75) With these relations and Eq. using Eqs. W[M^\Mlf)]W[Meu.e.4] = . Me.58) and considering large values of \z\. i. (19.j] given in Meixner and Schafke [193] (p.3] = [1. we use the following circuit relation which can be derived like Eq.g. Singular Potentials the second resulting equation from the first resulting equation.74) Moreover. (19. we obtain immediately W[Z.Me^} (19.e. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.M^] J™aV. 170/171). (19.4] = .* « * r r z H .(0.u.A] = W[Mi3\z). i.M^]= IT e~iv*av. and Wronskians W[M^\MIJ)] = [i.57) and (19.u}W[Me.48) (or cf. (19. or obtainable by substituting e.76) . 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.77) av I 2ism VK \ ctv av ^Thus.\ \ ( av 2iwKav ^ v . = IT <*„.452 CHAPTER 19. 7T W[Mev. (19. (19..Mev]' P l ' From Eqs.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.
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. (19.h) + BM^\z. (19.h)./) = (l)1/2R2^2i^)ei^)U^.(l)le^}Pi(cos 9).78) we obtain Ay reg « _ .l) ' We note already here that with the substitution R = ei7r7._ [ / ( f c .(ite™)" 1 { 8 J ' f(k. Z ) e ^ ] .g.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).f c . eii/7r/2 (19. Z ) e . Then f' . B=tfD'"". (19.85a) .79) where the second expression follows from Eq./ ( . (19. (19.83) The Smatrix element is therefore given by ifc*"* .46) together with the relation Mev(z. (19.19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). 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 . 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. as we recall from earlier considerations (e.h) = Me_u(z.80) 2^fc (1981) where \7r/ i t sin VK /(*. in Chapter 16). f c r .
9?. we can write this (multiplied > by 2ismvTT. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u).Smatrix in the form s i n 7T7 Si = CHAPTER 19. V(r)=g2/r Fig. (19. R 2i sin TXV. Thus in terms of the (3) variable z. h) (see Eqs. Fig. R — — = 2i sin ivy.2 The repulsive potential and the various waves.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.2.58)) 1/2 2r e~ 2hir cosh z. We can rewrite Eq.454 we can rewrite the .86) .58)) in the limit r — 0. 19. and with the left hand side following from Eqs. we can define respectively as amplitudes of the incident wave. (19. (19. (19.56).81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. cf. 19. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19.56) to (19.z—>oo 2r \l/2 i(is+l/2)ir/2 2hir cosh z r> \ „2ih cosh z RU + i [Re —2i/icosh z R Thus. and recalling that yreg is proportional to a function Mc> {z.
e. (105) to (108).88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. therefore we cite it from Meixner and Schafke [193]. if R becomes a pure phase factor and eil/7r a real exponential. i.** 19. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. (19.e. H. Jv+2r. l=—oo 1 1 lite*"* . Manvelyan. Hashimoto [124] and R.89) . A derivation is beyond the scope of our objectives here. in which v is complex. M^l)(0.19. Gubser and A. (19. Meixner and Schafke [193] therefore developed an expansion in terms of products of Bessel functions which is more useful in practice owing to its rapid convergence.( i J e ^ ) " 1 ! 2 = R2 + ^ . J.2COS2VTT = (R." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .28)) and R = ey real.e.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). Eq. S. (19. / i ) = Y.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. By inserting in Eq.Q. The expansion is: 4 r ( / i 2 ) M S ( z .h) The function Mo (0. i. 178). but in the 10dimensional string theory context.79). Eqs. Liang and Yunbo Zhang [189]. (19.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . The latter is what happens in the S'wave case of the attractive potential.±)2 + (2sinj/7r) 2 . and then exploit it for the evaluation of the quantity R.m a t r i x Our next task is to evaluate the expression (19. i.. This implies basically the evaluation of the expression R of Eq. (p. J.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. however not here. W. this expansion is given as 1 oo MJ> Hz. " S e e S. MullerKirsten.e. 180). (^)l4lu{h2)Jir(hez)J±u+i+r(hez).84) of the ^matrix.In Meixner and Schafke [193] (p.h) is (cf. i.2. unity minus reflection probability = transmission probability.e™*/R\2 ' ' . (19.88) where the factor Me„(0. (19.6 E v a l u a t i o n of t h e S . unitarity is preserved. h) serves the use of the same coefficients as in the other expansions.
r = 2 as a check).49). (19. (19.3 9 ^ .6 2 ) m 6 (^_i)3(z..456 so that in particular for z = 0 oo CHAPTER 19.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. Eq.28)). by allocating different values to r. we obtain M&(0.91) Inserting here the coefficients given in (19. (19.g.49) from which we obtain for z = 0: Me„(0. e.55). (19.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. u»^. As a matter of introduction we recall the definition of the coefficients with Eq.g. Here in the case of the repulsive potential. Next we evaluate M^1}(0. setting r = 0 in Eq.90) and choose r = 0 (one can choose e. (19. i. one obtains *. h) in many different ways. Singular Potentials 4r (h )M$(0. ^ (19.90). /=—oo (19. 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./>) = Co> 2 ) £ 7 " = — OO j ± > .90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq.e. /n ^ .h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h). + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ .2_4)(l/2_9)iv2y) + <AA. h2 is complex but v2 is real (cf.94) ./i) with the help of Eq. r = 0 and 2. (19. The formula (19.
2 _ ^ 4 ( ^ 2 _ 4)2 (19.93) and the coefficient expansions (19.96) With this explicit expansion we can evaluate the S'matrix.] 2 \v i + {v2 .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).32i/ . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .IV 2 2 2(v =F 3i/ .111) / / i 5 4 3 1+ 4i/ 2 + +' (z./I) h u .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy..19. From expansion (19. one obtains M£!(O. + # "+" ( I / . 4(z/ T H ^ 3 .l) 5 (19..97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves. We observe that this approximation involves v and h4 and hence is real.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .2^ 2 ± 59^ .12i/2 + 64u .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 . In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.7) F " (i/±l)2(^^l)(i/24)V2. _2_(h\2 .4)(i/ .23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ ..1) 2u(4u + 15i/ .!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .9) V2 1 fh ±v 1+ 2 + • (19.1 ) 2 ( I / + 1 ) ( V 2 . .
and hence allow for the general case of complex parameters v (the Floquet exponent in the case of the periodic Mathieu equation) although here in the case of small \h2\ we know from Eq.9 / . Eq.101) where / is complex and g is real. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. (e iun e~iv*^/R*2)' (19.458 19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.98) (cf.104) We now consider the case of a — 0 implying that g of Eq.84): Si ST 1 (11/R2)(11/R*2) . 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.101) is > zero.2. This is the case of real . In this case R — R* and so 1/i?2 ~ Ofji4). + ( i + s w (E 4 . _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2). (19. and sin 2TT/3 = ft/.28) that v is real. Then cos 2TT/? = 1 . (19. (19.102) (19. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. (19.eiun/R2)(eiu*n ~ (19.^ i 4 2+^ # 1 / •.R*2 ^ * #2 \ 11 tf2 R* (19.103) Then SiSt (1+5) i .100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. (19.7 CHAPTER 19.99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19.28)).
61n/i} + ~9~ 0(h8) (19.106) and evidently violates the unitarity of S. At = 2i[l + 0(h2)}. "This is the result effectively contained in R.6C .3b)) h2 is real for E = k2 > 0.107b) = Ah2[l + 0(h4)}. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. Hashimoto [124]. (19. (19.l ) 2 2 \) +0{h«) (19. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 . W. .2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.61n2 . Liang and Yunbo Zhang [189] (there Eq.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. H. With t h e help of Eq. 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. i. MullerKirsten. Hence with Eq.102) we obtain Sis.1)\} 2 h \2 iv 1 + Au {y . We can also evaluate now the amplitudes (19. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). Here in t h e case of the attractive potential (as remarked after Eq.107c) Here h2 is actually Vh4.Q. Magnus and F.3. Gubser and A. = 1 .28).577. p.e. S. (19. 1 29/ ^ (1 .* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .SiSf » 4 sin7rz/ R (19. f W . ( I l l ) ) in agreement with a result in S.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].107a) In this result* we now have to insert the expansion of the parameter v given by Eq.105) sin7r/3 (19. J.^ ) ~ 2 2 459 1+ 9 / • & ( ! . (19. Oberhettinger [181].19.97) this can be written A 4TT2 [v\{v . J. Manvelyan.
K. Wannier [278].3. Singular Potentials 19. i.1. S.$ q(l. MiillerKirsten and Jianzu Zhang [226]. completely different solutions. (19. we again make use of the replacement (17. W.109) ^We follow here mainly D.460 CHAPTER 19. This is therefore a highly interesting case which found its complete solution only recently.e. Park. H. 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. where the part of interest here is based on H.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. .4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . Aly.3 19. J.a }tp = 0. Tamaryan. We are again concerned with the associated Mathieu equation given by Eq. H. 19.4 Fig.26) and hence set a1 = 2hl + 2hq + A{q. VahediFaridi [8] and G. h) for large h and upwards I = 0. 2 2 a = [l + 2 (19. J.108) For the large values of h2 that we wish to consider here.h) 10.3 The function q(l. Miiller and N. W. 2.3.H. H. N.h) (19.
i. however. y± oc A{z)e2h sm * ± A(z)e2n sm z . The replacement (19.yJ\ — 1 and cos TTV . and a is its normalization constant. to which we have to return in this subsection. y(0) = 0. See also Eq.21b).3.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq.21b). Therefore the second term on the right is nonzero. (17.111) we have W[y+. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. which therefore cancels out or can be taken to be 1. 19.63). (19. i. We also found the boundary condition (17.§ a W[y+. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.y} where VF[y+.19. (19. (19. With the boundary conditions y+(0) = l. (17. we defined the Floquet exponent v and observed with Eq.110) a where y+(7r) is the solution evaluated at z = TV.5) that this is given by the relation cos iris = ^ M . We begin.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. (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.h*y_(l.3. .113) 'We are considering nonintegral values of v. which is even about z = 0.l + 2y+(l. J4(0)=0.e. with an essential mathematical step.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case.h*y (19. 19. The behaviour of q as a function of h for some values of / is shown in Fig.e. y'_(0) = 1.
(17.e.39 2 14 /i 2 + (19.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19.462 CHAPTER 19. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. in the second expression 2h is obtained in addition from differentiation of the exponential factor). Dingle and H. (19. B.112) from evaluation of Eqs. B. and hence with Eqs. 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 .32). J.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. Miiller [73]. (19.A(z)e2h sin 2 sin z sin z (19. (17. (17.114) at z = 7r/2. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . 1 1 Or see R. Hence we require (obtained from Eq. .43) we see that z = 7r/2 implies W(TT/2) = 0.46) 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 . We now obtain the expressions needed in Eq.118a) "We emphasize again: For integral values of v the normalization constants are different.114) a a Our first step is therefore the determination of the normalization constants No. which is \/2 at z = 0. J. W. K [ (z)}2hsinz (19. Miiller [73].117) a a Referring back to Eq. W. (19. (19. This paper contains the normalization constants of large/i 2 periodic Mathieu functions.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). i. Dingle and H. See R.
8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.! ) ] ! [ . a in Eq.62a) and (17.1Q11QM cos irv + 1 ~ •==— e .[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . 210.62b) (see also the explicit calculation in Example 17. (19. (19.\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].87 2 14 /i 2 [ i ( g .h . With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q . p. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. (19. (17. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.19. (19. .! ) ] ! [ . Tamaryan.118b) The factors a.117) are known from the matching of solutions in Eqs.i ( g +3)]! 1 24/i {2hl>2).3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .2g + 3) 27/i ( 3 g 2 .3).119a) This formula — derived and rediscovered in Park.
17. h. z) are TP(q.122) that given one solution ip(q. . h(q. V— 2in (19. h. for h — oo).h. —h.464 CHAPTER 19.h).z)exp[±2hismh z\.h.2. of course.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i.4.z). h..e.h.122) Correspondingly the various solutions ip(q.z) or as tp(q.? s m h z) Vcosh z (19. h. i = 1. z) can then be written as .124) Since this function differs from a solution ip by a factor k(q.z). —z). —h. . tp(q.z) = A(q. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19. Vcosh z \ 1 . z).3. With the solutions as they stand.108) and set ip{q.3. the expression (19.120) to (19.h) =?==. —z).109) into the equation (19. We define these solutions in terms of the function Ke(q.109) remaining unchanged. h. „ 1 A „ —A cosh z——H .z) i/>(q.3. dA 1.h.2 we insert the expression (19.h. Then straightforward integration yields > inq 4 „.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)iP(q. z) = ip(q.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. z) = ip(—q. it is still a solution but not with the symmetry property ip(q. each with a specific asymptotic behaviour. we can obtain the linearly independent one either as ip(—q. h. .120) The resulting equation for the function A(q.h.3 Following the procedure of Sec. In the following we require solutions H e ^ ( z ) . 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . (19. (19.123) We again make the important observation by looking at Eqs. h.
av{h2) = ^(0J ^ Mi i .<5±1) (19 125)  (the expression on the far right in leading order for h2 large). : 7r/2. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. * ** = > rojKe( ".^ s i n h 7?)]g.e. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. . In order to be able to obtain the Smatrix. (44) of W a n n i e r .h2) = OvMPfrh2). as we saw earlier.45). we h a v e 77 = 0—> 6 = 0. achieved with the help of the Floquet solutions Me±v(z.fi2) can be reexpressed in terms of Hankel functions./h). [ \ . 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]. 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. (0.e. in performing this cycle of replacements the function picks up a factor. i.127) sin z T /'1 M TT\ — <>{ 4 .h2). 2j<. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . (19.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . i.19. >• * F $ 0 * e". (19.125) 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 . Me„(z. This is. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. IT a n d 7 = 00 — * 7 > : 2fcsinh y + i<j7T h Aw 26/? T h e factor e " r ( ? + 1 ) / 2 o n t h e r i g h t of E q . Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. .3 The Potential 1/r4 — Case of Large h2 465 Instead. 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 ./i 2 ) M (19. i.WJ " C o n v e r t i n g t o n o t a t i o n of W a n n i e r [278].e.126) For large values of the argument 2/icosh 2. We observed that these satisfy the relation (19. (z. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. the Bessel functions contained in the expansion of the associated Mathieu function M{.
(2) (19. 3te»0.vr .q.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19./i) Be^(z.q.130b).h). (19.g.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. z): Re(2\z.h). (19.g.q.h) (3) = = = = Ke(q.eW(z. In this way we obtain in the domain . He (z. exp[—ikr — in/4\ kr Re{2)(z. h.h. 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./i). 3te»0.q.3.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q. (19.130a). (19.z).h) = e(z) exp ihez .q.q.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO. KeW(z.q.h) He (z. HeW(z.130a) Re^(z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .466 CHAPTER 19.4 T h e c o n n e c t i o n formulas We now return to Eq.130b) r_0 rll2exp\g/r (ig) ' 19. (19.q.h) Be^(z.
g . . (19. we see that the solutions (compare with Eqs. z) are proportional there.134) Considering now the first of relations (19. we obtain in the domain tfcz <C 0 the relations: Me.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z.q.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. Eqs.i ^ / 2 H e ( l ) (z> ?) (19.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z. and He(*> exponentially decreasing. / i ) + sin7r(7^)He(4)(2.h2) a. q.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19./i) — sin 7T7 Re^(z.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z.129)) He (2) (z.h) is asymptotically small (cf.135) sin irv 'This means.h).li 2 ) Oil. (19. h.h2) Me„(z. (19. (19.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . q.q./i).130b)).131) where the second relation was obtained by changing the sign of v in the first. these are exponentially increasing there. Changing the sign of z. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e . e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e . (19. h) = Ke(q.19.130a). h) — sin 7T7 He^(z. q.h) sin irv = sin7r(7 + ^)He (3) (2.132) into Eqs. z) and He (3) (z. (19.133) In a similar way one obtains the connection formulas — sin nv HeW(z.132) These relations are now valid over the entire range of z.(z. — sin 777 H e ( 3 ) ( 2 .g.g. Substituting the Eqs. h.(z. (19. t With Eq. h) = sin 7r(7 + v) He (1) (z. h) = Ke(q. q. (19./i).134) in the region where the function He^(z.g.q./i 2 ) = Me_j. and thus can be matched in this region.
Squaring Eq.(19.5 D e r i v a t i o n of t h e ^ . H.139) ''Equation (19.( . exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. Singular Potentials The factor on the left. (19. Wannier did not have the relation (19.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.135). of course.m a t r i x From Eq. sin TXJ. (60) in the work of G. (19. Wannier [278].l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr .135) is the equation corresponding to Eq.119b).119a) or (19. .l ) (l„—ikr e From this we can deduce that Si (19. (19. cos TXV + sin TXV. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . q. (19.85b) obtained earlier. i.e.Sin TXV.135).1 . The quantity 7 is now to be determined from Eq.3. (76): "It is not likely at this stage that an analytic relation will ever be found connecting (what we call) v and 7 to (what we call) (I + 1/2) 2 and h2". where 8i is the phase shift.468 CHAPTER 19.119b) as „4/i 1 + COS TXV ~ rrl ( g . The latter is defined by the following larger behaviour of the solution chosen at r = 0.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. is only the dominant term for large h2} 19. (19. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . which in our case is the solution He^4^. (19.119a) and (19. and therefore remarks after his Eq.e'i7ri 1.137) We see that this expression agrees with that of Eq.133)) — sin TXV He^4' (z.133) we can now deduce the 5matrix Si — e2t51. Thus here the S'matrix is defined by (using Eq.
the real part of v must be an integer. This is different from the behaviour for small values of \h2\. 9/2 (91) with the approximation q ~ h obtainable from Eq. (19.4 The absorptivity of the S wave (attractive case).e. i.139) could actually be neglected) implying  cos m>\ = cosh nvi.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^.e. 19. imaginary part uj (so that 1 on the left hand side of Eq.141) COS TTU irvj — i s i n WR s i n h TTUJ.e. 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. T h i s is r e a l for UR a n integer.28) shows. (19.109). (19. we can approximate the equation for q ~ h (i. h) of the Zth partial wave.8h COS From Eq. (19. A{l. i. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1.§ Since the right hand side grows exponentially with increasing h the Floquet exponent must have a large 1=0 1 large h approximation small h approximation Fig.19.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.137) together with (19. for the attractive potential. since cos TVU is large. .140) From this we obtain the absorptivity A(l.138) we obtain. as a comparison with Eq.
in fact. Handelsman. A. The diminishing fluctuations of A(l. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . in further contexts beyond those already mentioned. 19.e.139) 2Tr(16h)q /32\ f e Thus with Eqs. B. . YuanBen [285]. 11 J. Challifour and R. 19. Partial or other aspects of the weak coupling (i. and. (19. F." as well as other aspects. Paliov and S. (19. Fubini and G. Cvetic. and G. N. Pao and J.142) we can see the behaviour of the S'wave absorptivity as a function of h. C.4 . Jones [78]. Esposito [88]. Bertocchi. Furlan [20]. Tiktopoulos and S. Treiman [271]. 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. In essence. "G. 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.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently.k) in the approach to unity are too small to become evident here. VazquezPoritz [61]. Apparently this is one of the very rare cases which permits such complete treatment. **R. see also M. Masson [191]. H. I r a n [61] and M. Rosendorf [225]. D. **A. S.4 (there are tiny fluctuations in the rapid approach to 1).P. This is sketched schematically in Fig. one may expect corresponding results for other singular power potentials. Shabad [248]. E. Apart from the papers already referred to. Cvetic. Lew [128]. Dombey and R. Pope and T. Y. A. Lii. A. W Some of these sources can be helpful in further investigations. such as phase shifts. S . L. N.470 CHAPTER 19.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. Lii and J. Eden [46]. J. small h2) case have been considered by some other authors. Limic [177].107b) and (19. D. H.** Finally we should mention that the potential r~ 4 together with the associated Mathieu equation have also been studied in interesting contexts of string theory.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . H.
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. *R. f R . M. Wu [18].Chapter 20 Large Order Behaviour of Perturbation Expansions 20. T. B. the subject is much older and had been explored earlier in great detail in particular by Dingle^ with the subsequent investigation of the behaviour of the late terms of asymptotic expansions of the eigenvalues of the Mathieu equation and others. Thus in the following we shall not only obtain the large order behaviour of the coefficients of the eigenvalue expansion but also that of the coefficients of the wave functions as well as the connection between these. also convergent expansions are known and a lot of literature exists on the equation.* However. This is indeed a remarkable connection. We concentrate in this chapter on the large order behaviour of asymptotic *C. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. B. 471 . and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. Dingle and H. Muller [74]. the asymptotic solutions in different domains all have the same coefficients. W. [19]. Dingle [71]. We shall also see explicitly.* In fact. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. Bender and T. e. J.g.
r for p = 4. Fig.2 and below.1 The function ur = pr jr\ vs.472 CHAPTER 20.7. 5. This approach leads . It is a feature of a physicist's approach to a problem that he employs a method of approximation which is such that the first term of the corresponding expansion yields a rough value of the desired answer. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. Presumably this has never been done so far. r for p = 4.2.2. 10 12 14 16 18 20 Fig.6. as evident from Example 17.1 and cases in Sec.6.2 The function ur = r\/pr vs. 20. Since the method is also applicable to convergent expansions. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. 8. 20. 19. 8.8.
In some cases the answer is affirmative.20. (20. 20.2) or. .1 and 20. and the expansion is said to be Borel summable. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. the function E(g2) is given only approximately by Yln=o En92n.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E.1 Introductory Remarks 473 automatically to asymptotic series. equivalently for g = l/h. The behaviour of such terms is illustrated in Figs. in the power expansion of the exponential function). The traditional method of using an asymptotic expansion is to truncate the series at the least term. i. for instance.2.e. and the integral E(g2) is called the Borel sum. 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. this term represents the size of the error. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). It can be seen from these. even if each En is known.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. For an authoritative discussion of these aspects we refer to Dingle [70]. 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.J2 2n h E (20. The formal perturbation expansion of a quantity E(g ) . T V lim h2N 2 n EQ?) . i. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20.3) n=0 Thus for any given value of g . as we discussed earlier in Chapter 8.e. 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.
We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . . 1 we obtain oo F(z)£(*)" = — . In this ideal case the integral of Eq. (20. Thus if CO F(z) = aiS(z) + J2 anzn.1)? Asymptotic expansions originate from integrating a series (i. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . . Large Order Behaviour of Perturbation Expansions Eq. In fact. (20. (20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. . n=0 (20. Q. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]).e.474 CHAPTER 20.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. (and E0 = 0). In any case. 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. . .n _ En+\ n! En+1 ^ (!)"«!. it is important to know the large order behaviour of the asymptotic series so that the question of its (exact or only approximate) Borel summability can be decided. . many asymptotic series do not even alternate in sign (as in the case of the series in Eq. n=0 (20.4). (20.5) then with the integral representation of the gamma function F(z) = (z — 1)!. El .6)). In general this is not so easy.
S. we have a quantum mechanical problem where probability can leak away.. 20. ? When ?s E ^ 0. (20. and E represents an eigenenergy..10) r as shown in Fig.2. M. 20.N = l.e. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).3 The potential V(r) gr»...2. e. Popov and V. The potential (20..20. or can we determine Ar independently? Fig. 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.9) The question is therefore: How can we determine 9 E(h).1 Introductory Remarks 475 except for possible subtractions. for < ^ 0. $s E can be calculated by the WKB procedure. (20... This is the case if the real potential has a hump of some kind. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. if the potential is V(r) = +grN.. .10) is considered in detail in this work. Weinberg [228].. (20.11) V. i.3 and g < 0.g. § where 7(ff) = To exp  . N = l./ \p\dr\.
L. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq.76) (see also the comment on the comparison with the Kepler problem before Eq.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. (20. (20. 20.73) in the W K B or semiclassical approximation. S.113)). However. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. Popov and V. Here n is the principal quantum number of the Coulomb problem (see Eq. Equation (20. Large Order Behaviour of Perturbation Expansions where the frequency of particle bounces against the potential barrier is 70 ~ ^classical/27r and w c i assica i = 2ir/Tkepier oc 1/n 3 . (14. Eletsky.11).^ The logarithmic factor in Er is a novel feature of this case. Solution: The solution can be found in a paper of Eletsky. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available.134)).9) for Ar. (11. (11. Thus in the following we apply these methods to our typical examples. . 8.7 with its focus on the structure of coefficients of perturbation expansions was seen to permit even the formulation of the recurrence relation of its coefficients. the BreitWigner formula (10.1: Application to the Yukawa Potential Evaluate 9 £ = —7/2 for the Yukawa potential V(r) = g— r / \ r and insert the result into Eq. of course. V. as was (and still is) not the case for t h e equation 1 V . Weinberg [83].1.476 CHAPTER 20. Popov and Weinberg. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential.11) is. as we expect on the basis of Eq. the perturbation method of Sec. Example 20. M. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. the cosine potential being t h e most completely investigated will also be seen in the following. An interesting application is provided by Example 20.
20.13) represents effectively a partial difference equation.13) In the special case j = i the boundary conditions stated after Eq. Eqs.26) and (17.j) of functions ipq+4j.34)) that the coefficients Pi(q. Eqs. (cf.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential. J. 3)? . h). arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.55)) X = E(q.2 A = Mx + ~M2 + 7^WM3 + •• • . or equivalently of the quantity A. (17.14) Thus these coefficients can be obtained once the coefficients Pi(q.e. j .l ) +2[(q + Aj)2 + l + A]Pl. (20. to avoid discussion of difference equations here) 2 l{q j) ^ (16)«(»l)!(2i)«/° / _ 2q 2q + 3\ f i \ ' [\{qmwm\ sr ){j)> (2(U5a) R. % = Aq. (17. i. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. (17.3 ) P i _ 1 ( g .l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. (20.j + 1).12) (20.M2i+1 for large i. W.j) are known.60).l ) ( g + 4 j . Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. . (17. B. (17.l(q. obey the recurrence relation jPi(q. Dingle and H.Bq.h) = 2h2 + 2hq+j.58) and (17.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.j) a n d M2i can be deduced. Eqs.J) = (g + 4 j . This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. are given by i . The recurrence relation (20.Cq. Miiller [74].58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g . M2i = 2j2 J[Pi(q.
19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.91 l)]![i(g3)]!} 2 { [ .16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.15b) The zth term in the expansion of the eigenvalues A of the Schrodinger equation with cosine potential (Mathieu equation) is therefore given by ith term i\ (8h) il 1 2g2 + 3 2i )[{[\{q 7. Using this relation and the duplication formula lw('\W±^.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i .478 CHAPTER 20.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.18. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .•91 {[l(? + i)]![(? + 3)]!}2 (20. . In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.l (20. (20.*.16a) (20. see also below) .l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula. Hence exact Borel summation is not possible. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods. also written lCj.
20.3. l . in the calculation of the level splitting as a result of tunneling. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. \ 1 \ \ \ \ \ 0. 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.3 Cosine Potential: Complex Eigenvalues 479 20. we emphasize a few points *M. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. i. remains unchanged. we follow here a formulation in the context of our earlier considerations of the Mathieu equation.5 \ \ / \A Fig. .1 Cosine Potential: Complex Eigenvalues The decaying ground state Our intention here is to consider again the cosine potential and hence the Mathieu equation with the same large/?. f H .20. We retain the previous boundary condition at one minimum. but with different boundary conditions. the difference q — q$. Reeve [262]. MiillerKirsten [200]. . but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary. .* However. 20. Previously.2 solutions obtained earlier. A means to achieve this was found by Stone and Reeve. this difference was found to be real. 2 . n = 0 . .e. The potential we consider here then has the shape shown in Fig.4 The cosine potential shifted by TT/2. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. W.3 20. J. when this is g0 = 2n + l .4. At the risk of repeating what we explained earlier. Stone and J. .5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0.
but achieves exactly what we want.e. i. 20.21) <u> + ( i « . Previously this situation was the same at the neighbouring minimum. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. and correspondingly we assume alternately even and odd oscillator eigenfunctions there.4 (around z — 0) to the one on the right (around z ~ ir).20a) As stated.4. We can distort the potential there or alternatively impose a different behaviour there on the wave function. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. Effectively we consider the tunneling from the left well in Fig. 8 the equation can be approximated by d2ip(w) (20. Here we choose t h e second way. and allows its imaginary part to tend to infinity. (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. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 .T K ^ a D_i_{q+1){±iw). 20. Thus one chooses not the solution with real argument at the minimum. Setting w = 2h1/2z. (20. as we found these for this potential. but t h a t with complex argument there.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. X + 2h2 = 2hq+—.Ah2z2 • • • }^{z) = 0. a boundary condition which at the face looks somewhat abstract. we shift the argument by n/2. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. .2h2 cos 2z]ip{z) = 0. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . which means replacing z by z±ir/2. Now we change this situation at the minimum on the right.480 CHAPTER 20.4. 20. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.
(20.20. we naturally constructed the even and odd solutions with these (cf.e. .24) Considering t m O i n Eq. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions.63). whereas here we construct these from oscillatortype solutions. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. Since the large/i2 solutions valid away from the minima were the solutions of types A. such that this deviation becomes imaginary. . 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.4. 3 . whereas here we start off with a solution even or odd about a minimum. by a different method. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. Thus.> 0 for z > ±ioo. 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. 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)}.22). Eq. 17. A derivation is given subsequently. 2. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. however. that at z = 7r.23) .e. we see that ips ~ cos \j2qhz for z ~ 0. 20. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. . i.3 Cosine Potential: Complex Eigenvalues 481 As explained above. 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. Here — again as before — we take the finite height of the barrier to the right of z = 0 into account by taking q only approximately equal to go = 2n + 1. (20. We consider here explicitly only the case of the ground state around z = 0. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. .
Magnus and F. 17.29) Equation (20. (20. (20.482 CHAPTER 20. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20.24) are linearly independent around z = 0 for q ^ qo.g. this can be handled in a few lines. Oberhettinger [181]. 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.27) requires the differentiation of this expression with respect to the index v. tp(v).•w) + 2ir T{u) <"+!)% Dvi(iw). We see this as follows. 92. (20. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq.w) = e T Dv{w) 2TT n») AvViDv^iw). Setting v = 0 in T(—v) = (—v— 1)! gives infinity. p. Oberhettinger [181]. From Tables of Special Functions'. §See e. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function.(. Magnus and F.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. W.h(2h1/2 J £2=0 (20. .we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.27) differ by an angle IT. we know from the solutions of types B and C in Sec. (20. Thus the only nonvanishing contribution comes from differentiation of Y[—v). The arguments of the functions appearing in Eq. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.25) qo) E2 0: (20. However.g.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial.30) 'See e.3. 2. W.28) Extracting Du{—w) we have A. p. and then evaluation at u = 0.
29) v=0 (in + In w)D0(w) + v2ire (m + lnu. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. which are valid around z ~ 7r/2.. i. p. pp.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. (20.31). Magnus and F.e.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . 3 .27) exponentially decreasing contributions. i/^0 483 T(v) It follows that d dv [T(u)\ (20. . W. Reeve [262]. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.)e" w /4 2 l ^Dx(iw) (20.33) = IV2~TT.//n and ij>{y) = C . those of type A.20.g.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. .+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. Oberhettinger [181]. We now go to the WKBtype solutions. dropping from the derivative part of Eq.4. Oberhettinger [181]. r(i/) We now return to Eq.e. . (20. "See e. (20. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. . we obtain 4>s D0(2h^z) D0(2h1/2. 5 7 7 . Stone and J. This result is » not easily generalized to <jo > 1. Differentiation of this equation and subsequently setting v — 0 yields. and "Prom W. For u — 0 one obtains the result (20. Thus calculations for qo ^ 1 are not given by M. including the derivative of DE2ufl(2h1/2z)).29). 92. Magnus and F.
— 721 . 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. that in approaching z = ir. (20. We now return to the WKBlike solution (20. (20. (20.e. y COS c\ J (20. (20. (20. by there replacing z by z + ir/2. smf (20.40). 2 c2e2hehz UJ2 . Eqs.e. i.41) 2TT . x . where we can then match these to the solution ips.. near z = n the boundary condition (20.sm I We now want to match this solution to that of Eq. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). cos z/2 ~ (20. Then V'WKB becomes (with cos z ~ — 1 + z' /2. We observe that in the matching domain (i.37) Comparing Eqs. Recalling that we have shifted the potential by 7r/2.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo.38) in its limiting domain towards 2 = 0.36) V'WKB cie~2hehz' 9h  .29) and (17.e. i. (17.34) and (20. Thus in the present case and with q ~ qo = 1 we obtain „2ft.35) We construct with constants c\. we obtain the required solutions from the earlier ones. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .34).36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20. To this end we set z' = z — ix.39) Now.40) From Magnus and Oberhettinger [181] (p.36) and consider this at the other end of its domain of validity.cos z — 2/icos z ipA^ Tcos  and i n A —.36) cos  .484 CHAPTER 20.
44) = DQ{2h / z) ± (2h7r) / e.45) Comparing this result with tps of Eq.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. i.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.46).1/V)_le^_ Z With this result Eq.43) Comparing now Eqs.43) we obtain. (20. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. with c\ = exp(—2h) from above. (20.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2.20. J Hence Eq. (20.38) can be written TAWKB (20.e. We derive the general result in the next subsection using our own method.^. 1 2 4 1 2 8h ^^2 2 (20.39) and (20. The general result stated by Stone and Reeve [262] is therefore presumably guessed. (20. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l .34) we obtain E2 = i%\ = ±ih22Qe~8h.46) Repeating these calculations for the general case (i. (20. (20. . ^'mliw <2a47) For qo = 1 we recover the result (20.2/. 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 .e.
E curre it \ j. at z = TT/2.5.5.46) we consider a half period of the cosine potential as illustrated in Fig. We have to impose two sets of boundary conditions. 20. Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e 7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P. Considering the original equation.e.e. i. (20. (20.Ah2 ( z + . W.48) These conditions provide the required quantum number qo = 2n + 1. 20.3.5 The cosine potential from — TT/2 to n/2. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. . 20. J. Fig. i. •_(:)*<>.e. Achuthan.486 20.20a). MiillerKirsten and A. this equation may be approximated there as d2ip ~dz^ + E + 2h2 .z. On the other side of the barrier at z = —TT/2 we consider different boundary conditions.()#0. i. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. Eq. At the minimum to the right of the barrier shown in Fig. Wiedemann [4]. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .2 CHAPTER 20. •+(=)* V(f) 0. H. V(z) .*. the result is formula (334) there. and expanding cos 2z around the point z = — TT/2. (20.) + 7T V> = o.
20.r/2 ~ e i(2hq) V 2 .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. i. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum. B and C.z)=tpA(q. (20.h. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.z).e.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. 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.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 . For the continuation we use again the WKB formalism. three pairs of solutions of types named A.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish.21)).h.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. as shown earlier.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. (20.51b) 2i(«1)[i(9l)]'. we demand that ^(*)L<. Proceeding as in our previous cases and substituting the expression (20.21) into the Mathieu equation we obtain. Recapitulating these from Chapter 17. (20.
55) Extending these solutions to the domain around the minimum at z = 7r/2.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. (20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . (20. we obtain our first transcendental equation. ^ W H . (20. we obtain 1>±W = \ a a (20. and the second around z = 7r/2.e.488 CHAPTER 20. 5. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.52b) The first solution is valid asymptotically around z = —vr/2. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. 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. the relation q .48) and proceeding as in Chapter 17.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}.=F2 2\1/2{28h2y/4e~Ah l + O (20. Here the upper sign refers to the even states and the lower to the odd states.52a) = Cq[w(z)}.57) where go = 1) 3.^ i . Finally the third pair of solutions is Cq[w(z)}+0 Cq[w(z)}+0 Cq[w(z)} where Cq[w(z)} = 22(9+1) •((73) 1 h1/2 1 hW> (20.qo . \vw). i. We return to this condition after derivation of the .56) Applying the boundary conditions (20.
and we obtain the proportionality by appropriate expansion. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.49) and we can apply our boundary condition. Some nontrivial manipulations are again required to uncover the generation of important factorials.5 9 ) The WKB solutions to the right of z+ (where V > $IE to the left of the maximum of the barrier) are given by ttw . I^A(Z) overlap parts of the domains of validity of these solutions. Byrd and M. as we shall see with the result of Eq. D. p. i. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A.58) Ignoring the correction A / 8 . We begin therefore with the matching of V'WKB ( Z ) to IPA{Z) and integ rait* V'WKB (Z) to I=f V'AW Hence it is necessary to consider the elliptic (20.20.[ [Ah cos z .01.67). r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. Our procedure now is to match these to exponential WKB solutions. Jz+ ' T In principle the integral could be evaluated exactly — see P.61) [Ah2 cos2 z2hq]1/2dz. . Thereafter we extend the latter across the turning point down to periodic WKB solutions. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (20. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. we see that the turning points at z± are given by 2hq .e. F. (20.2hq] / dz) We know that the solutions I/JA{Z). formula 280. Friedman [40].Ah2 cos2 z ~ 0. e x p ( . C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °.Ah2 cos2 z ip = 0. These then provide complex exponentials like (20.
59) 2h COS Z+  ~ 0.64) .sin2 z = l .62) as follows.62): L: = In ~ and hence L = ln In * f 2(z + f) {q/2hyn + + (* I In + 1/2 1/2 2hj 2 In In 2h 4(1 + sin z) (g/2/i)V2 cos z' 2(z + f) (qphy/^z+Z) !sin2(f + f) ( 9 /2fc) /22sin(§ + f)cos(§ + ^2 ' 1 4 = ln 4tan(f + f (9/2MV2 (20.490 CHAPTER 20.63) We handle the logarithmic terms in Eq.cos2 ( z+^) = (z+ . (20. (20. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.) +0 z+ IT Inserting this into the integral I. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. We have for the following expression contained in (20.» _ i y v . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). we write cos2 z = l .
65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.64) into Eq.50a) and (20. (20. We ignore this here.20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. (20. . on using in the second step Eqs.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 .68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated. (4^2)i/4 [ i( g _3)] ! V/ l (20.50b). we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .+ qj V2 4 A.3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. 1/2 (20. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20. .60) by (4h2 cos2 z)1!4).62).63) and (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I .66) The WKB solution therefore becomes (approximating the denominator of (20. ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  .
T 7 77 r )V2 /4(2 . i.69) l(8M9/4e2^ \2 [ i ( g . (14.49).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.KW4^8 4 .+) + B^) HAh^/\ir.49).2 [ ? ) Imposing the boundary condition (20. i 27r(8^/ 4 2/t /4 ' r(±) . (14.™) .492 CHAPTER 20. (20.U ^K^.2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T . we obtain T _ ( ± ) = 0. We now return to the even and odd wave functions defined by Eq.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20. Large Order Behaviour of Perturbation Expansions again using the Stirling formula.i ( 2 / i g ) 1 / 2 ( z + . where _ 1 + [±) *) + £ (20. 1 2 1 ±2 .l^l !T ^^* T2 7 r ( ^^ /.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i).1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .50)).55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+. Eqs. io(8/*)g/V^i6ft2)V4[±(g3)]! .67) and (20. (20. S1H (2/ KZ ) / (z f 2 (2^)..?/ 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. Thus with IPA(Z) and TPA(Z) taken from Eqs. we obtain l(8/i)g/ 4 e.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .z)}].
72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.72) We observe that with the square root as on the right hand side both sides are complex. Combining these equations by replacing (28/i2)<?/4 in Eq. (20.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .h) + (qq0)( BE — 90 E{qQ.21) expanded about q = qo. for q an integer both sides are imaginary.h) = = E(q0. h) of Eq. In both cases we wrote (cf.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. 20. Eqs. (20.3 ) I = }(»« !7rl/22I(.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . (20. we obtain E(q. i.72).h)+2h(qq0) (16h)i°+le'Sh E(qQ. (20.72) is our second transcendental equation. Equation (20. Ex = 2h2 + 2hq0 + A 1 (<? .qo) El Ah' (20.47). the result (20. (20.e.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.74) This result agrees with that of Stone and Reeve [262]. (20.57) by the left hand side of Eq.h) + i l + O '4{[^ol)]!}2 E(q0.25)) E = A = Ei + E2. who derived it for qo = 1 and guessed the form for general values of qo.3.73) Inserting this expression into A = E(q. Eqs. We now have the two equations which have to be satisfied. in fact. (20.57) and (20. (17.26).20.h)+i%Ego.75) .
* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state.e. (17. The same formula can be derived in the case of the double well potential. Eq. Bogomol'nyi and V. A. MiillerKirsten and A. A = . Achuthan. (20.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. H. i. the discontinuity across its cut from zero to infinity. 20.W [£(901)]! < (20. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. .494 CHAPTER 20.I g  ± l ( 2 „. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj.+  .74b). cf. J. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20.80) *P. its tunneling properties and the large order behaviour.3. B.2 below. Setting ng AEqo:=2iSfEqo.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.9) the coefficients Ai of its perturbation expansion. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. Wiedemann [4] and Example 20. Fateyev [26]. we can recalculate with the help of Eq. 25Eqo=8h[ 2\1/2 . (20. by others. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. W. We had A= or _2ft2 + 2ft.) {I6h)™/2e4h 1 + . 7 8 ) i=l V .4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. f E .
(20.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.14).3 ) P i _ 1 ( c ? . We also observe that these terms do not alternate in sign. j . For convenience we write Eq. J. (20. Dingle [74].l ) ( < ? + 4 j . J).83) One now sets ftH = ( 2 % P * ( 9 .l ) +2[(q + 4j)2 + l + A}Pi_1(q.13) out again: jPiiQj) = (g + 4 j .81) We observe that this result agrees with the result of Eq. W.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . (20.j + 1). The integral is recognized to be of the type of the integral representation of the gamma function. a = J + \<1> (2084) . the integration is seen to give (20. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q.j) + {q + 4j + l)(q + 4j + 3)Pi1(q.1).j) satisfy the recurrence relation (20. *R. But as Dingle remarks.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. Using again n = (go . We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. and the method can be applied in many other cases. (20. MullerKirsten [200].20. f H .. so that an exact Borel summation is not possible.74). 20. there is nothing special about this case. (20. B.19) for the large values of i under consideration here.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. A.a R . Taking dominant terms. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. (q + 4j) 2 ~ 4j(4j + 2g).e. a.89) Replacing here K by a and \x by 1/2. d2WKh dz2 + 1 4 K z zM2n WKtll = 0. (20. i. Now. which are solutions of the differential equation known as Whittaker's equation. 0 4j(4j + 2g) pi_i(a .^(Z).(a) oc .h) = ^2pi(a) with p.4 ) ^ .h) +p(a + l.31.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.90) See M.496 CHAPTER 20.1) + 2 P i _i(a) + Pi_i(a + 1).g. we obtain (2a .z)WaA/2(z) Setting V. 4j(4j + 2g + 4) + / . the recurrence relation (20.h) 2+ 8h a p(a. 1  1 / 2 (4 (20. .86) one can convince oneself that Eq.91) + Wa+lil/2(z) = (1 . g < 4 j . (20. nx pi_i(a + l). Stegun [1]. (20.e. i.87) Taking terms of 0(h~l+1) of this equation. z2 (20.l.h). p(a. we regain Eq. e.85) implies p(a . (20.85) Considering now the generating function of the perturbation coefficients.4. (20. Pi(a) i.l/2l Wa. Abramowitz and I.ll{z)= L . formula 13.e.83) can be written j .1/2(z) (a1)! (20. Whittaker functions WKjfl.27h pi(a) a = 4j(4j + 2 g . (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .1) + 2 (a — 1) a .85).K + ^ ] L + K^)WK.
(« . (20. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. h) is expressed in descending powers of z = 8/i. p. (20. a (20.4 Cosine Potential: Another Method of Calculation we obtain Vai. > > ^See W.88) is known.h) ~/(fe) ' (a1)! . In the present case this means the replacements q — —q. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . ^Observe that if one solution of Whittaker's equation (20.94) Hence to insure that p(a. . Magnus and F .86) have to be chosen such that p(a. v .i/2(z) + Va+lA/2(z) = 497 2+ Va. i. Oberhettinger [181]. (20.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. such that p(a.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.87).92) Comparing Eq. h) is expressed in descending powers of h. we have to remove a factor (recalling a = j + (7/2 of Eq. > (20.i/2(z).92) with Eq.(« 1) 2 ]. g(h)(20.84)) {8h)q/2( Ah for j < 0.\)V . We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20.g(h) are functions of h which in view of Eq.93) (a1)! ' where f(h).(« . Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). h —> —h.e..j — —j. (20. Dingle discovered the solutions^ p(a.20. 89.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo.h) = \~\Pi(&) with pi oc h l.
. (20.83) and is given by 1 1 . n = 0. h\ oc (8h)q/2e~4h ( _ .95b) Then for j > 0 we obtain from Eq. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.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.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. W. Comparing > > > this result now with Eq. we obtain the result of Dingle: (20.498 CHAPTER 20.95a) Inserting (20. Muller [74].96) and a similar expression for j < 0 with /i — —/i. (20. . g — —q. Correspondingly we set for j > 0 / 1 A pla = j + q. Dingle and H. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . ^ . This expression can be obtained from the recurrence relation (20.94) here we have a series in descending powers of h. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2..l ) ] ! 11 R.1/2(8^) ( . B.+ i f or 3 < 0 (20. _ i s.j _ g / 2  1 / 2 ( . 2. (20. J.). j — —j.8 / t ) for j > 0.. i).1.
p.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. . l{Q.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.97d) Then the coefficient M<n of the eigenvalue expansion.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. E* J'=I i+M 2 [2i + 9 .98) We evaluate the sum by approximating this to a formula given in the literature ** r \ f R\ __{r + Rl)\ J (rl)\(Rl)\' E j=0 i.97c) We assume that in a leading approximation we can cancel the factors (i + \q).3)]!2 2 '+^.157(4).l ] ! (20. W. is M2l = 2j2j[Pi(<l.14). Thus ° 9 (i + \q) 2«(i+Igl)! p . note the misprint there: (r — s) ought to be (r — s)\. A special case of the formula can be found in I. (20. J. Ryzhik [122]. (20.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . M. S.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.{j + \<l). B. Dingle and H. i. formula 0. formula (30).4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. 5. Miiller [74].100) **R.j) (i + lg)[2i + i ( g .e.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq. (20.+i?_l]!}2 (20.0) = 1 as desired (although not really enforcible for large values of i).20. of Eq.l ] ! {[.e.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. Gradshteyn and I.
replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.e. (18. Eq. 20.1.e.l. (18. [19]..86) and set (cf. .o i • . Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20.3 ) ] ! _ 1 Hence M or. i\ M i large 1 .i ) ] ! [ ^ +  ( g .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.86)..5 20.34) that £ is an expansion in ascending powers of h 6 /c 2 . [^ + i ( g . i.500 CHAPTER 20. i.2.Am i . n = 0. i.e. since \{q — l) + \{q — 3) = \q.. With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator.. Eq.' 1 i • (20.5.. We refer back to Eq.101) This formula allows us to cancel all factorials [i.102) Thus the result is M m 1 2«Fi + g .105 ) the latter because we see from the eigenvalue expansion (18. the result of Bender and Wu [18].17).e.]\.e. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity. (18.4)) £ = ^ = ^+ w and y= ^ > ( 20 . i.l l ! with the large i behaviour in agreement with the result (20. i.81) for q — qo = 2 n + l ..
l ) r .2 The double well It is clear that one can now proceed and apply a nonselfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue. i.77) can be verified.109) or. the discontinuity across its cut. (20.106) With £= E r=0 Mr (20.(20.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.86). and then the relation (20. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.e. in terms of K where go = 2ftT + 1. (with E = h2£/2) my') = 290+1 (y')9o/2 e 2/'/3 (27r)V2[i(go_i)]!' and integrating with the help of the definition of the factorial or gamma function.5. .4) of Bender and Wu [19].107) we obtain (i)r r 71" JO yirZ£{y')dy'.108) Proceeding in this way.20.e. 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.110) This is the result in agreement with the large order formula (4. We leave this to the following Example. (18. and hence inserting here the imaginary part from Eq. The result establishes the series unequivocally as an asymptotic series. This can then be compared with the level splitting we obtained for this potential. in fact as one with Borel summability in view of the factor ( . h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . 20. i.
It was later explicitly established for the level splittings and imaginary parts of arbitrary states in the cases of the cosine potential and the double well potential. Cizek. W. 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. J. H. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). J. Gram. Achuthan. R. t p . Paldus and H. MiillerKirsten [164]. J.Q . J. J. V. . t The relation (20.502 CHAPTER 20. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. Wiedemann [4]. it can be used to obtain the other. A relation similar t o (20. Cizek. S. *R. Eqs. J. M. K. Liang and H. Wiedemann [4]. *Some of the literature in this field following the work of C. E.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161].77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . [19] can be traced back from articles in the Proc. Grecchi. S. Silverstone [52]. Silverstone [62]. Its usefulness has also been demonstrated there. in t h a t when the quantity on one side is known. M. as in later chapters. " P .77) and its possible generality has been referred to by Brezin and ZinnJustin [36].6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. MiillerKirsten and A. W. MiillerKirsten and A. Damburg.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26].2: Late term behaviour of double well eigenvalue expansion Derive for the case of the double well potential — by application of a nonselfadjoint boundary condition (as in the case of the cosine potential) — the imaginary part of its eigenvalue and hence the large order behaviour of the asymptotic expansion of the double well eigenvalue. + 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. Large Order Behaviour of Perturbation Expansions Example 20. S. (357) to (371). V. J. Bender and T. J. R. R. . Nakai. Graffi. H. J . J. W. J. J. T.^ 20. M. Grecchi. The relation (20. Wu [18]. Harrell. K. Harris. Propin and H. G. J. E. Damburg. of International Workshop on Perturbation Theory at Large Order (Florida. Harrell. Solution: For the solution we refer to the literature. Achuthan. 1982) [139]. Paldus. Propin.
1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. However.Chapter 21 The P a t h Integral Formalism 21. Probably the most readable introduction. Kleinert [151]. We illustrate the method here by rederiving the Rutherford scattering formula with this method. This formalism is particularly useful for evaluating scattering or transition amplitudes. P. The path integral is. *R. Felsager [91]. The formalism is not so useful for boundstate problems (in this case the initial wave function is. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. For the application of path integrals to the hydrogen atom (i. A standard reference is the book of R. A further wellknown reference is the book of L. in general. Feynman [94]. However. can be found in the book of B. A brief introduction is also contained in R. R. Schulman [245]. 'See in particular H. Feynman [93]. Chapters 2 and 5. 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.e.* 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. Hibbs [95]. which explains also difficulties. S. 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. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. 503 . however. Coulomb problem) we refer to existing literature. ^ Our interest here focusses on quantum mechanics. since in such cases the probability of this system choosing a path far away may be considered to be small. Feynman and A. P. P. unknown). Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method.
tf.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. U. i.ti). Obviously K(xf.2) where K is a Green's function. */ = U + (n + l)e. =J dx(xf.x. we can compute the action S1. 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). (21.x).3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.504 CHAPTER 21. the initial time.tf) = / dxK(xf.t). be ip{x. (21.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). ti).tf\ip) . *In Dirac's notation we write this equation later (xf.1) Let the wave function i/> at time t = ti. (21. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf. we wish to know* ip(xf.tf\x. 2mo dx2 (21.5) Sn = X / fc=l mo (zfe . The Feynman Path Integral Formalism 21.ti\tp).e—»0 Sn. ti) = 5(xf .ti)ip(x.2 P a t h Integrals and Green's Functions The onedimensional Schrodinger equation for a particle of mass mo moving in a potential V(x) is (with mostly h = c = 1) i d2 + V(x) tff(x. (21.ti){x.e.
1 A path in (x.6a) . The usefulness of the method therefore depends on the fact that some paths are more probable than others. The multiple integral involves individual integrals like f dxk exp Xk)2 + {xk Zfci) 2 } with Xk contained in the step up to this point and in the step away from it. 21.t). The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation. as will be shown to be possible below.1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig. of which one is illustrated in Fig.tf). 21. Here and in the following we need one or the other of the following integrals (cf.ti) to its final position at (xf.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. We now consider intervals in position coordinates and in particular the following integral over intermediate position coordinates Xk which corresponds to the sum of exp(iS/h) over the uncountable number of possible paths from Xi — XQ to Xf = xn+i.28)): f dr\ exp ia a) ' (21.21. the Fresnel integral (13.
. We then integrate and finally let /x go to zero. n > 0.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.z f c _i) 2 } } . 2 + 2a.506 CHAPTER 21..0 ^7 (x f c + i . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.s f c _i) + (xk+i .tf.t ~Xk+l> + xk+i 2 ) +{xk . one has j ^ dr1eifa+i^ /£ = .m0 . dxneiSn/n. (21. One therefore removes the troublesome factor by defining (Sn being the sum in Eq.ti) '•= f g m ) (N(€\]n+i dxL. J—oo dxneiS/h — • 0. (21.Zfci) 2 } I +{xk+i x*.(x fc+ i .* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.8) Note t h a t in spite of the n integrations. dx exp z—{a.5)): G(xf.xk_i) dx exp i — { 2 a .xk+x oo —oo oo .6b) —c ia o drjr] e x p—V € (21.37)) dxk exp d(xk . tf > U.Xi.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. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.. 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. ie f me\ 2a\a~) 1/2 (21.mQf Xk L ~2e^ ^ .6c) In the evaluation of these integrals we replace a by a + ifi.xk+i) exp .2 + (x + x f c + 1 . for example.
A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. B. (21. p.24).10) can be looked at as the product of a succession of free particle Green's functions (or propagators).t In the following we demonstrate that the function constructed in this way is.Xi. The consideration of the denumerable number of possible paths connecting the particle's initial position with its final position may be handled in a number of different ways. (21. 183.U) = / JXi=x(ti) = K(xf. (21. and will therefore not be attempted here.e. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. 21. However.5). We shall see later (with Eq. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. Felsager [91].g.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. 11 For discussions see e. 183. It is neither a variable of integration in T){x} —> Ylndxn. (21. i. . B.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. G(xf. as in our discretization in Eq. and so on. (21.ti) For tf > ti we now write* G(xf.tf.26)) that — ignoring the potential — the multiple integral (21. in fact.2). (21. Different constructions of these families of paths usually lead to different measures.e.1. p. however prefixed with a new factor. summation over all possible paths.tf.Xi.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. which all vary from —oo to oo. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. See the square root in front of the exponential in Eq.2 The Path Integral and Green's Functions 507 Eq.g.21. (21. Felsager [91]. we arrive at a path integral without the measure factor N(e) above. These integrations are indicated by the horizontal arrows in Fig.5).ti) for tf > U.tf. § See e. the Green's function K of Eq. nor the limit of integration of some variable Xi. it may be noted that with a different philosophy concerning the summation over paths. i.7) for n — 1) to give a time interval.Xi.
Expanding this in a Taylor series around x. one with the full action S. we obtain the same as before except that there is no factor N(e).tf + e.1 and allowing n to go to infinity. . we have G(xn+2.tf + e. Each of these linear pieces is completely characterized by its endpoints. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig.tf. Then the measure drops out.xi.188. Felsager [91]. pp. so that the piecewise linear paths fill the entire space of paths. 187 . (23.xi. I We employ this method later in Example 23. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.ti. the other with that of the action of the free particle SQ.ti). (21. 21. and one is calculating the Feynman amplitude or kernel relative to that of the free theory.74)). Consider a time t = tf + e for position Xf = xn+2.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.3 in the evaluation of I a specific path integral (see Eqs. the discussion in B.72) to (23.Then we can construct a recurrence relation as follows. we have See. Then we have on the right the factor G{x+rj). We have G(xn+2.U) imo 2 G(xn+i W)Idr. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.ti) = 1 j^p: f°° cte n +iexp } • fm0.Xi. It is for these reasons that — unless required — the measure is frequently ignored.508 CHAPTER 21. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.exp 2e n ieV(xn+2) G(xn+2 + V.Xi. 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. however.
U) + Q(e2 = (lieV(x) + 0(e2)) G{x. we obtain: G(x. d and this means d i—G(x.G(x.ti) = 1 d2 + V(x) G(x.Xi.ti) 2 —l€ — ^2 + V(x)G(x.Xi.tf. tf. U) = e^G + 0(e 2 ) dtf 1 d2G(x.ti) N(e) or G(x.tf.tf + e.tf + e.(21.6c).tf)= dxG(xf.ti) for tf > ti (21. Xi. tf.2 The Path Integral and Green's Functions so that G(x.Xi.tf.Xi. Retaining only terms up to those linear in e.xi. It follows that ip(xf.14) = G(x.Xi.xi. (21.21. U) .tf + e.ti)ip(x.15) We see that G(x.tf''Xi.tf.xi.tf. ti) satisfies the timedependent Schrodinger equation.— p G ( j . 2mo dx2 (21.t Thus + 0(e2 (21.13) We integrate with the help of the integrals (21. . tf + e. tf.xi.tf.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.6a).U + 0(e ).ti) +\if(^G{x.tf.xi.xi.U).xi.tf.xi.Xi.ti) +V(x)G(x.tf.tf.x.ieV(x) + ri< —G(x.ti)\ . U) + .tf.ti) .6b) and (21.U) +••• \ 509 G(x.ti) = j^rr dnexp I ^rj1 .Xi.ti) = G(x.16) .Xi.ie 1 d2 2nV0dx^G{X. tf.Xi.
X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation. The result.3.17) Differentiating K with respect to tf we obtain —K(xf.U)G0. and the Green's function KQ\ S0= Here (cf.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.x.8)) Go(xf. now called So.U) = / T){x} eiS\ [fdt \mx2.16).1)). It follows therefore that {{W ~V+2m~0~^) K{Xh t/.ti) = 5(xj — x). 21. denoted by KQ. K0 = 9(tf .15) is a wave function (cf. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. 21. We have therefore established that the Feynman integral (21.ti) = = 9{tfti)—G + G5(tfti) .U) = G(xf.U)6(tf .10) is indeed to be identified with the timedependent Green's function. Eq.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.20) . (21.x. or K(xf. (21. as this is also called.x. (21. Eq.x.510 CHAPTER 21.tf. 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. In a problem where V is small it is convenient to calculate K with a perturbation method.tf.U.U). (21. as may be verified by inserting this into Eq.Xi.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. The function G which satisfies the timedependent Schrodinger equation (21.tf. Hence our original function K is the same as G. (21. is the Fourier transform of the nonrelativistic propagator.tf.
(21.25) T h e quantity Ko(x.Xi.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.Xi. In Example 21. But t = tf — ti = (n + l)e.22) • \ n/2 .ti) = = G0(x.tf.37) — by explicit solution of the differential equation which the Green's function satisfies..21.ti)9(t). (21./ . oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.i ) 2 .tf.ti)= iX (o.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.3 The Green's Function for Potential and hence Go{xf.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.ti) Ko(xf.1 we show t h a t oo /.24) ( s / .tf.Xi.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.23) We now insert our expression for In into Eq.23) for Go and obtain Go(xf.Xi.z / c .Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.tf. n + i . we have Go(xf.Xi. exp where Xj = xo and Xf = xn+\ and there k = 1).t) K0(x. (7.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2. (21. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21. 2t Go(xf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.Xi. (21.tf.. . Then Go(xf.x 0 )" n + 1 (in agreement with Eq.21) fe=l and t = (n + l)e = tf — U and h = 1. exp 2 (21.tf.8). f c . We obtained the same expression previously — see Eq.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.
(7.25) we can rewrite the Feynman path integral or kernel (21. we have instead of 5*0 and KQ expressions S and K. For n = 1 (see below) h This result can be verified like Eq.x „ + i ) 2 = (x„ + 2 . In our proof of induction we assume the result is correct for n.. i. the latter being a Green's function with interaction like that for the harmonic oscillator. (21. Then oo 2 / (21.1: Evaluation of a path integral Verify Eq.xi)2 = 21 xi ^2o(„.2y(xn+2 . Example 21. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 . If the potential V(x) of Eq. (21.ti. as / JXi f r D{x{t)}eiS°lh .tn)••• K0(xi.22). or else by induction.xo)2 + (x2 . (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .54).22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:.x „ + i ) 2 = ( x n + 2 Xi + Xi . Solution: The result can be obtained by cumbersome integration.t0) (21..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. (21.512 CHAPTER 21.e.26) with n integrations and n + l factors KQ.)!} J — OO . ^ (xi . (21.tn+l•xn.f_^y e.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.6a).5) is carried along.7) by direct integration with ^ . e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .)+(1"+2.x ^ 2 + y2 .x0. Eq. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.an) Then n I 1/2 (n + 1)A" "I 1/2 fX r ioo ^n + dxn+1eiX{^+v2y2Hx"+2x"+1 l ) 2 } J —OO (n + l)A n «ix„+1ea(Sl»22»lI«^I.
0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. A(n + 2) 21.29) Hence with KQ from Eq.e.Xi) + ( x n + 2 Xi)2.6a). We pass over to momentum space with the help of the following two integrals.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.x2rriQ 1/2 toij_. (21. e > 0 .27) we have (with a = t/2m0) .i) H —3/ . ) H — (xn+2Xi) n+2 n+l n+2 y2 . a dp exp [ipx + iap2] = exp — i(21. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A . Using Eq. (21.2 y ( a . This quantity is also described as the nonrelativistic free particle propagator.2. it ..max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21. (21. i.a .21.Q .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.25): Ko{xJ) = = (S) e{t)(^X'2 y J . 21.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.i j ) H n +— z 1 . (21. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a.i) . n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i .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.n+2a. The second integral is an integral representation of the step function. and + 2 2 l n + 2 ( x n + 2 . n + 2 . (n+l)A n n j(a.3.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). The first.2y(xn+2 .
an expansion in rising powers of the coupling constant (contained in the potential). 21. The Feynman Path Integral Formalism Im t \t>0 Rex Fig.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. Replacing 9(t) by its integral representation (21. (21. This is the type of expansion most people describe as perturbation theory. Thus in first .28) this becomes K0(x.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. i.e.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.Q (21.2 Contours of integration. 21.514 CHAPTER 21. The first order correction contains one factor of the potential V.*' where 2mo — r. The generalization to three space dimensions (required below in Eq. ( 3 b) ° The expression E+ V — it 2m.51)) is fairly selfevident. the second order two factors of V and so on.
t/.5) a n d (21.tf. (21.. Eqs.^ J_00 V(xk. .t) JSo dtV(x.21.20)) and with elS = exp i S i o / hi dtV(x.tf. .Xi. (21.U). li f f 515 order (cf.t) the first order correction is (G\ ~ G — Go) Gi(xf. dxr.4 Including V to First Order: Perturbation Expansion h=l. dxn exp i ^ —.32) Since — we recall this for convenience G(xf. } e °° f l f •/ oo /"oo F ( a .. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.^.i V e / D { a .ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.Xi.(xfc .xi.i. \2meJ n+l J.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/._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 ... ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x. .t)\.xfc_ OO J— OO J—OO L J.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq. .33) xG0(xi. x (n+l)/2 \ lim .)} (21. (21.^){iV(a.t)] n+l .ti.i.
t') + • • • . t2.t)K0(x. in writing down the expansion exp i dtV(x.516 CHAPTER 21. (21.e.xi.ti) t2)K0(x2. . (21.ti). i.t2)V(x2.t. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). dx2Ko(xf.xi. = i dt2 / dx2Ki(xf. t) Ju Thus the term containing n times V contains a factor 1/n!.xi. OO /"OO / oo dt / J—oo dxKo(xf. Jti Consider the term of second order.e.ti).x2.t) =li dtV(x. i.x.tf.35) We also observe here that the problem of timeordering arises only in perturbation theory.ti)K0(x. This is cancelled by n! different time orderings.36) {if •j.tf. / 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'. Xi.i / dtV(x.tf.t2)V(x2. They must therefore be deleted.t)V(x.x2.t) (21.37) Jti The last two contributions do not describe the time evolution from ti to tf.t) ! Jti r*f dt'V(x. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.I dtV(x.t2.U) xV(xi.t2) xKo(x2. (21. The factors of "V" originate from the series expansion of the exponential exp ftf .
Xi. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj.xi. The diagrams represent mathematical quantities and hence are designed on the basis of rules. These may be formulated in configuration space as below. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. x r l i 'V(x. called Feynman rules.t)Ko(x.E). . x"^ X x"2.t)= f dpdEeipxiEtV(p.t2 K o (x 2'l2 .t)V(x. tf.t) is V(p.tf.ti) = K0(xf.x. The Fourier transform of V(x. 21.t) ) K K K0 'V X K o Fig.i ] P I dt J dxKn(xf.ti).Xi. (21.t. For various reasons — such as illustration.3 Feynman rules and representation of the Green's function.t.t)V(x. or in momentum space as mostly in field theory.Xi. E) which is given by (in onetime plus onespace dimensions) V(x.ti) i dt = K0(xf.39) dxK(xf.tf. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.39).38) .tf.ti) (21. of course. Here. U) (21. Such diagrams in the context of field theory are called Feynman diagrams. 21. (21.tf.40) The rules for the present case are given in Fig.ti).21.t)K0(x.tf.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf.3 along with the diagramatic representation of Eq.x. we are concerned with quantum mechanics.Xi.Xi.
T = duration of time for which potential is switched on.518 CHAPTER 21. ^(x. * k 1 • . i. There the wave functions ipi.41) We have to pass over to three space dimensions.t) = . The Feynman Path Integral Formalism 21.e. and dn = dnxdnydnz. The plane wave function normalized to one over a box of volume V = L3 is.tj)K{xf.7 = e i k ' x _ i : l i with f i/}*i/jdx. i. to one particle in all of space. (3) The transition probability per second is = \A\2 . for the allowed values of k: kx = 2imx/L. (2) The incident flux (number of particles passing through unit area in unit time) is equal to the incident velocity times the density of incident particles and this is = n—7 particles per cm second.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit threemomentum interval** = V/(2irh)3.ipf have to be properly normalized. We recall for convenience from **Note that from the wave function we obtain. 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.ti). .= l.ti)il)i(xi.k = 2wn/L.e.Xi. with p = Kk. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics. here with E — k 2 /2mo. (21. (21. mo V since ip is normalized to one particle in volume V. dn = (L/27r) 3 dk.tf. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf.
117) and Example 10.21. u = ^\H'kfp(k). The transition probability per unit time is .47) . Then H'ki can be taken out of the integrand in Eq.e. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.— . 47T 1 3 7 (21.t) = e~^lT\ r where r = Ixl. H = H0 + \H'. Eq. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. # = ^an(i)VnM) n = a^(t) + \a^(t) + .k..= 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.. P(*) = j r (2145) (cf. V(x.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.44) and f_oo can be replaced by fQ. Proceeding in this manner one obtains Fermi's "golden rule". (10. «= — . (21.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. in which T is a large but finite time. i.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.4). and with these the following perturbation ansatz: ih^* with an(t) = HV.44) = u(Ek. (21.
by threedimensional dxf and so on.44) by contracting the propagators to instantaneous point interactions. The threedimensional generalization is (taking tf » T and ti <C —T. (21.52) . i. ti) = co5(t — U)5(x — Xj) etc.tf. (21.x.. Thus with the ansatz Ko(x.tf)K0(xf. we obtain A1 = clS(ti tf) / dxf(i)V(xf.tf.520 CHAPTER 21.ti).ti)'ipi(xi.i x (i)V(x. (21. This now has the form of the amplitude (21. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.f.).49) i.tf)Ki(x.44) (here in the scattering problem with initial and final energies equal). The lowest order approximation of A. (21..39) OO /"00 / dt / OO J —OO dxK0(xf. (x*. (21.tf)ip*(xf.yLi.x.>/>.tf)ipi(xf. so that 0(t) can be dropped) K0(xf.t. x. dt dxip*f(xf. In our previous onedimensional considerations we had with Eq. (21.^ / .t)V(x. t.49) together with the following wave functions (of asymptotically free particles) 1>i(*i. Ai = dxf / dx.f. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas. Xj. T a large but finite time.30a) K0(xf.t)K0(x.* / . (21. we have to replace the onedimensional quantities dxf etc.ti).U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .tf.t) (21.51) Inserting this into Eq.tfx. t.48) where from Eq. We can relate this amplitude to the amplitude ark of Eq. £.tf). (21. also called Born approximation.xuti) = ^ J°° d p e i p x _ i ^ * .Xi.tf. t)K0(x.e.t) = ^ J*ie*<*'*><<*'*>.50) Here x — xj — Xi and t = tf — ti.
k.p ) .iaT2/8: L With 0 0 • .21. (21. (21.k. 2 „ x v~ m o (21.T) = x exp L z/ q/2 k2 521 f dx. which is evaluated in Example 21.t)exp A(p. k2 ipx + i—t + ikx — i 1 2mo 2mo (21.T) = ^JdXJdt%*t2/T2exp i ( k . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .56) (21. this becomes A(p.T) = JdtJdx(^\V(x.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. . £).58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.54) Now inserting the expression for the potential V(x.).f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i 2mo (t — U) + ik • x 2mo . P2 . Then integrating over q and q' we obtain . Then we require the following integral.i ( k 2 .k.57) we have A(p.r) = = y .p 2.~ .fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .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 .59) .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21. (21.53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.k.
62) We want to consider the limit T — oo.T2f 8 V/ 2 ." + M ] p) ' "•"• ^2 . otot contains the factor \A(p. <rtot. ( 21 .\ 2?7lo / 1 (21.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.t o t= y dp ( 2 v r ) 3 "V 3 I W: 2mo y m0V k 32^a2S(^) y ( k y 22 [[ ( k _ p ))2 + M 2 ] 2 ' kP 2 (21. we obtain . / / P 2 _ . dE = PjV = ^ E ± .63) we have: Probability per unit time = 77777.60) Now.a ( — ) ^ .64) Inserting the result now into our expression for the total cross section. In our case this has not been done. { I i _ F T l } .^ . T ) ~ . i. Using the result (21. write down the probability per unit time which is ( 8 \1'2 _a2 16^ . k . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .522 CHAPTER 21.T)2/T.51 — 2 /w2 y ~ —. k2 . . (21. e (21. e 8{k) = lim 7=*—^ (21. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21.e.59). This expression assumes that the integration over time t is normalized to 1. (21. Using Eq.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .3. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T .„.P ) 2 + M2]2 4 VvrT2. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).65) But E=*.„ 2.< ^ £ \M VTTT2.k. mo m (21 .63) We establish this result for T — 1/e in Example 21. 66) o . y 2 [ ( k .
/Yukawa = ) Solution: We choose the 2axis along the vector q.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 — . i. setting M2 = 0.68) Inserting (21.e .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.21. (21.p ) 2 = 2 k 2 ( l . Then (x = r) foo „2jj.70) .67) and (21.M £ M * = / m0kdn = m0^2m0E (k .. (21.e. Example 21.65) with the cutoff M > 0. p 2 = k 2 .cos 9) = 4k 2 sin 2  .68) into Eq.69) This is the wellknown Rutherford formula. r\ poo Yukawa = = ^ Jo 2TT / r e~Mr y_i dre~Mr d(cos 9)e^r sin qr = — G / cos e = 2TT / J0 —e~Mr iqr { eiqr .e. giving the potential a coupling constant. —Mr 4w = . esfn (21. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .* 9 ' d r e .2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .^ . (21.j . 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.^ .4.67) T h e delta function here implies energy conservation. This result may also be obtained from lyukawa of Eq. 2mo / dft.59) by replacing there q 2 by ( k — p ) 2 .j .= e " f c a ^ o 2v^fa /4a = lim £^o —. and inserting the expression thus obtained into the final result of Example 10. so t h a t (21.
.71) with respect to k yields the same as °° 2 . Direct integration /4a . '<*> = . a simpler method suffices. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. We consider again nonrelativistic quantum mechanics. however. Va we obtain g(k) = Je~k V a = 0. In the present case. In configuration space the state of the particle is written x).524 The function required is CHAPTER 21.71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.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. (21.* . We write the state vector of a particle of threemomentum p in the momentum space representation: p).73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /.q).oo J / )eikx. ' i a = lim e —0 pk 2 /e2 e^TT 21.71) 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. We have x> = I J ^ l p X p l x ) . 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).^0 a^C 2^/TTO. Since dxe" 1 1 1 = « / . (21. (yx) = 6(y . 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 ) .72) co From (21.70) and (21. Differentiation of (21. a.75) .x) (21.
Ryder [241].t). The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.t).We now introduce the Dirac bra and ket formalism.x. pp. (with ft = 1) d/dt\ip) = —iH\tp).e.82) (21.t\ = iH{x. See also L.t).e. (px) = e~ipx.t)} constitute a "moving reference frame" in the sense that ip{x.160. Similarly we have a position operator X with Xx) = xx). which describes the propagation of the particle from its initial position x. (21.e.i) = e iH4 x.i'x.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. Thus (qPp) = p(qp) = p(27r)3<5(q . at initial time ti to its final position xy at time tf.^ l x ) . ) H = (x</>o)s. t) = (x. x.77) (21. x' — x infinitesimal.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc .t) = +iH\x.p).£) = ( x V ^ ' . (21.t) = e* H t x).t'x. i.t). we must have d/dt{x.81) Comments on signs in Eq.t> =iH\x. t\ipo)s. M^ip(x. Then ( X  ^ .i) = K(x'. .. t' — t.e.t'. 4) is to satisfy the Schrodinger equation. (21.If '/'(x.76) These relations imply the completeness relations 1 = J dxx><x. The time development is given by the Hamilton operator H. i.e. d/dt\x. The equation 1 = J ^p)(p. as had to be shown.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. i. i.We define x.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). so that \if>t=o)s = H>)H.0) = e""x).80) (21. so thattt d — x.83) (21. 159 .82): The general Schrodinger equation has the state vector to the right. We consider first an infinitesimal transition.t) = —iHip(x. Hence for the infinitesimal transition (x'.. Then the states {x. (21.21. i. (21.t\. H.
x y(x)e i p .V 2 < 5 ( x . "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.88) Thus P has the operator representation P= /dxx)V(x. (21.85) The momentum operator P satisfies the relation P = /(2^3PP)(P' (2L86) so that it projects out the momentum q of a momentum state q).* .P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«..y<&(p'ix')(x'ivix)<Xp) = f dxe.526 CHAPTER 21.85) the amplitude or expectation value or position space representation (x'#x) = .x).90) (P'IVIP) = y<&.x = V(p . The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P '  e ""'"'' )  P ) < P  X > e.e. i.p').x') + V{x)5(x' . 2mo Since (21.91) . (2187) (21.89) Using this and previous relations we obtain for the Hamiltonian (21.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).i p '. (21. Pq) = J 7^3Pp)(pq) = J dppp>*(p .
93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21. We now consider the transition of the particle from Xj.t) (cf.82)). We have <x'. But e'ipx + 0[{t .6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .p') + V(p . the signs are reversed when these operators are applied to the ket vector \x.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.p'). t).i'x. (21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx .84).92) We now return to the expression (21.21.t') \^(27rf5(p 2mo +V(pp' where we used (21.tj+i\x. EQ — Hi .p) + i{t . i.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 '.j. ti to x j . Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.e. 527 (21.92).p') • • • j W < 5 ( p ' . (x'. (21. i. for the amplitude of an infinitesimal transition.95) when applied to the bra vector (x.fx. tf subdivided into infinitesimal transitions (xj+i.tj).
. i. d x „ _ i .tf\Xi.xk (21.e. .97) H hH(pn.2. t n _ i ) t„2><xi.ti.x i ) H h pn • ( x n . ..t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o .t 0 ) + H(p2.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.tixo. .tn respectively. then corresponds to the sum over all paths.+ V(x f c ).528 CHAPTER 21.98) tk — *fcl # ( p f c ..x 0 ) + p 2 • (x 2 . . x n which the particle passes at times tQ. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.xic) by fc — Xfej H(pk.e. t n  x n _ i .a — —e/2mo and divide b o t h > sides by 2TT. . . (21. . d x i . d x i ( x n .tj) = / ••• / d x „ _ i d x n _ 2 .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. T h e integration over the n—1 intermediate positions..94) and thus obtain a phase space expression. (Xf.. i.i/xj.tfei) fe=i ..tf\Xi. dx... x f c ) = ^ . We now insert for each of the infinitesimal intermediate transitions the amplitude (21.xn)(tn in_i)]. d x n .. x i ) ( t i . zmo T h e n t h e integration over all p ^ can be performed by using Eq.27) in which we make the replacements x —> eifc. x exp [ .99) . x i .dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .X n _i)] h) (21.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .ti) rx/=x(i/) d x i dx2 . = 1 X X<jPfc We now replace H(pfc.i { f l " ( p i .
] ex H ^I (.103) is another form of the pathintegral for which the integration with the help of Eq. Smolyanov.e. i. Truman [254]. (21. G.98).e.101) Then Eq. G. to / m0 \ 3 / 2 las. .tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. / {xf. A.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . > Thus the phase space representation of the pathintegral* (21. Tokarev and A.ti) = J D{x(t)}D{p(t)}expi [' dt\pi. (21..8) as N(e) so t h a t the integral does not vanish for e — 0. (21. i.H(x.tf\xi.p)]\. (21.100) 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.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.21.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.102) We see t h a t the factor (with h = 1) m0 2m(tk 3/2 3/2 [N(e)}< = tki) \2irie) is precisely the normalization factor inserted earlier for the onedimensional case in Eq.
xfc). Thus finally we have Feynman's principle: (Xf. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. (21. Garrod [106].104) where for the particularly simple Lagrangian L under consideration here m0±l . this agrees with the corresponding discrepancy in the onedimensional case of Eq. t to x. pk) = L(xk.x) (21. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. (21.x) . x. (21.iJ(x fc .ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x.t'}. operator (21.8).H(xfe.tf\xi.e. £ /  X J . pfc) = Pfc • xfc .ti) where nl D{x(£)}exp '. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. £ J ) with no operator in between. i. .105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. Consider the following expectation value with primed and doubledprimed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'.tf\xi.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). I hi dtL(x.V(xk) = m0±l . We have now obtained the path integral representation of the matrix element ( X / . nl (xf.8). p . t.530 CHAPTER 21.
* * .0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. ••• 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 (21.108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x". With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'.v ' x ' and thus {x".107). (21.t'}. x fdxG  ^ P <p"x3)<x3e^V) x (21.109) . ./ ^ I < « .t"\x5){x5\x'. .21. so that we obtain a cnumber position coordinate which we can shifted across other cnumbers. Thus <xV'xx'.* * f dp = j (2^ p) ^ d .77) and thus have 531 *!*') = *M = * 7 ^ I P > « . (21. (px).6 Path Integrals in Dirac's Notation For the verification we use Eq.76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). With the wave functions (21. .t"\x\x'.i p '• x .5 l^ t 'lp / > e .*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.
113) Inserting for each of the infinitesimal intermediate transitions the amplitude . (21. (21.t') dp . The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.112) where x" is a cnumber. ] .0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .0 dp x„ / (27T): 1 .£ n xx n _i. we obtain (here and in the following the limit n —> oo at the end being understood) (x/. £ n _ixx n _ 2 .tf — tn and x$ = XQ.tj) at equal time intervals e. We now proceed from there along parallel lines.94) for the case with no operator in between. we obtain with (21.t'){x"\H\x') = X + •••} (21.£/xxj.i t " ( x "  #  x 5 ) ( x V ) +. This is the expression corresponding to that of Eq. ( x " . x / / e .ti = to as before.x') .£'> = x"[«5(x" . tn2) • • • (xi.t n _i) x ( x n _ i ..t"\x\x'.85) (xVxx'. = x"6{x" .f ) i f e i p .x ' ) (2^ i^+vM + D «p(x"x') (21.iit!' 1') 2mo + vtf) + <5(x"x').532 CHAPTER 21.tf subdivided into n transitions (xj+i. tixx 0 .90) we can rewrite this expression as (x".74) of the delta function. . We now consider the transition of the expectation value of operator x from Xj.i(t" .tj+i\x\xj. Inserting here the integral representation (21. (21.x') . .111) 1 .• dx1 (x n .£j) = / ••• / dx n _idx n _2 .it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .i{t" .*/<x"x5)(x5tfx'> + . Thus with x^ = xn.110) Recalling Eq.i ( t " . t0).£j to Xf. (21.t') (xV'xx'.
x (21.ti) 533 rxf=x(tf) = / • • • / dXidx. tj) = xf (x/.7 Canonical Quantization from Path Integrals (21.114).iyxxj.{x".t') = (x" + 5x"..7 Canonical Quantization from P a t h Integrals We saw above that in the path integral method the evolution of a system is expressed by a basic functional integral like (21.x . (21.t"\x'.115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] . the string of factors xi X2 .112). .t') sider the variation S(x". we obtain* (xf. tj  xj1 x». We demonstrate first that the operator representation of P given by Eq. For simplicity we consider the case of one spatial dimension.x 1 )(t 1 Z & ) +H(p2.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq.21.t"\x'...10) or (21. n—>oo *•»• 21..t"\x'. Xn _i X„ xi dx.x 0 ) + p 2 • (x 2 .xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. .{^( P i. tj \ xt.t') (21.\Xi.t"\x'..114) x exp [i{pi • (xi . x n _ i x n reduces to the single factor x^. ..tf\x.89) is consistent with the conjugate expression dL (2L116 * .« • > We reconsider therefore the transition amplitude (x".2 .x n _ i ) ] . Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.t"\x'.2 .. (21. (21. dyi l xi x 2 .x2)(t2 ~ h) + • • • + H(p ••• + p „ .3 " n d x f c .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.. dxnn _i Xi X2 .y (27F (27F •" (2^(2^)3 exp[.and hence may be put in front of the integrals to yield (xf.( x „ . U).t') .t') = and con 5x"Vx„{x". x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .105).
Thus in this case of classical mechanics dL_ d fdV. .x). (21. dL r /•*" .(x". jt/ 37 ( £T. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x.l&cdi.t i_(dL_ x't' h\dx J t„ (21.534 CHAPTER 21.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').122a) = 17) 5x"Vx/. t". and we demand the validity of the equation of motion. (21. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t.118) JI(x)/h .t'). <9L' dt —dx + — 5 x ox ox f .119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.t"\x'. Hence w « = (t). (21.t"\x'. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21.10) and set D{x} = D°{x}/JV(i/ . dx dt \ dx _ (21. we separate the metric factors from T){x} in Eq.t') =  6— / 1 V°{x}t .tt). Then.t'): 1 fx" 6(x". The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.118) — = (21 5x"(x". x) Jt Jt dL.i/(x)/fi ) j ^ ) dtL(x. " (t). with N = N(t" . For the purpose of enabling some formulations below.
125c) Corresponding to our relation (21. .x).21. F.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. (21.125b) and the measure property (3) N~\A + A') = A^. S. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). (21.t').115) in three space dimensions. x' (21. 6 (21. Streater and A. / Jxi V°{x} / Jx' V°{x}.7 Canonical Quantization from Path Integrals or JL(x»y\x.Xf = x(tf).1 (A)iV.tf\xi.122b) By comparison with Eq.126) R.t')(x'. ^2\x'. (21. 535 (21.124) where with A = tf — ti. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21.1 (A / ).1>) = ^{x". we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. Wightman [264].t'\ = l. I(x) = / JA dtL(x. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.125a) (2) the completeness relation holds.t"\p(t")\x'.. § The evolution of the system is expressed by the Feynman functional integral (xf.
(21.127)) (x". Then the relation follows with the help of the properties (1).tf\xi. tn lie correspondingly in consecutive order between U and tf.ti) f ' V0{x}xlX2eu^lh.tf\xi\x'.131) = ih(x".t"\X'.t\x'.t"\x'.tf\x(ti)\xi. (21.t"\x'.t').536 CHAPTER 21.133) = ih{x".t'){x\t'\x2\xi.132) The generalization to higher spatial dimensions proceeds along parallel lines.t"\x'.128) where tf is later than ti and t\. We have (cf.ti) = Xi(xf.tf\T(xix2)\xi.ti) = —!— p V0{x}xx • • • xneu^h.127) We can now write down an expression for a timeordered product of operators.ti). we have the relation (xf.t>). Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.t') + x"^J^{x".t'). Eq.t"\x'.t"\x"\x'. (21.t"\x'. We can rewrite this expression as the commutator relation (ITU*" x"^](x". (21.ti) = x' 'Y^{xf. (21..130) ~i5x7jX"{x". But for t = ti we have (and equivalently for t = tf) as we can see from Eq.t') (21.t').t') \ i ox" i ox"J or in commutator form [P.t') = ih(x. .t\x'. Thus if T denotes such ordering.tf\T(Xl • • • xn)\xi.x](x..129) = ^ y With Eq.114) (xf. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.. We consider the case n = 2.. Thus (xf.122b) we can deduce the nontrivial canonical commutation relation arising here. (21. (2) and (3) above. (21.t')=x"(x". (21.
N. In physical applications such cases are rare. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. Expansions of this type arise. divergent. nonzero radius of convergence in the domain of the coupling parameter.g. W = e x p ( l / z ) . Such expansions are strictly speaking. z = a + i/3 with a — oo or f3 —> oo or both. i.g. p. integrated over) beyond its radius of convergence. Consider e.e. although the first few terms indicate a convergent behaviour. the series converges. for sufficiently large r. Watson [283]. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e.g. E. It was therefore natural to search for alternative methods of expansion. > 537 . In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless.e. If the expansion is mathematically well defined. see e. Whittaker and G. 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. 102.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. An essential singularity is really an ambiguity rather than an infinity. as we have seen in Chapter 8. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. in the sense of rigorous convergence tests.Chapter 22 Classical Field Configurations 22. such as the fine structure constant of quantum electrodynamics. i. i. The vast majority of perturbation expansions discussed e.e. in the context of quantum mechanics are not convergent. T. so that successive contributions to the first approximation do not invalidate this approximation. Besides. but asymptotic. Such an expansion parameter is usually known to be or assumed to be small in some sense. it has some finite.g.
such as soliton theory.e. In the following we consider various typical examples. These methods are often described as methods of collective coordinates.538 CHAPTER 22. one considered in particular an expansion which is such that the first approximation is purely classical in a certain sense and such that this ignores quantum effects. On the contrary.h2. i. The procedure requires a change of variables from the original ones to collective and fluctuation variables (in analogy to centre of mass and relative coordinates). There. cnumber. 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. i.g. A further challenging task was the development of methods of quantization of theories which incorporate classical. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0).. and has led to insights which previously seemed unimaginable. although this terminology is not precise. 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. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties..e. the latter's topological properties if any. and in general this change is accompanied by certain constraints..h3. This type of expansion is decsribed as "semiclassicar. nature of the dominant approximation does not imply that this describes classical motion. Many of these aspects are interesting by themselves . higher order corrections are of orders ±. The classical. One therefore faces the problem of quantizing a system which is subject to constraints. The fundamental extension of the method of canonical quantization (i. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. A method which achieves this is in particular Feynman's path integral procedure.h. In searching for other means of expansion. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative".) l/h2.e.h°. However. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. such > a series begins with a contribution proportional to (e. The study of expansions of this type has turned out to be extremely fruitful. and the corresponding expansion is called the "loop expansion".. 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. cnumber first approximations.
However.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. V{</>) = m§0V + A0(<^)2. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking".* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics.2 The Constant Classical Field 539 and in any case deserve detailed study. This means theories with field densities which are therefore infinitedimensional. a static (time independent but spacedependent) solution of the classical equations of motion. 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>.22. 22. 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). It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. We shall not be concerned so much with the case of a constant first approximation.V(4>). (22. Tong [272]. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. 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. modified by going to imaginary time). to keep this case in mind.1) ' T h e reader who wants a highly advanced overview of instantons. Thus references to field theory will be exceptional and hopefully do not irritate the reader. vortices and kinks. d^] = d^d^ . . monopoles. it is important for a better understanding of the considerations which follow.g. may like to consult the article of D. or even a solution to a modified field equation (e. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). 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.
cnumber) solutions <j)(x) = 4>c = const. Classical Field Configurations where d^ — (—d/d(ct).2) We ask: Are there classical (i. as illustrated in Fig. 433. B. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. Felsager [91].e. V ) .3) See e. 22.S£/<90(x. i.540 CHAPTER 22. p. § (22.1 Different potentials for different signs of m§. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0.1. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. m l + \04>*4> )$ = <).£). then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. An even more familiar example is the GinzburgLandau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape). and hence here [d^ + m20}(l) = Xo(f (22.e. £)) is bounded from below we must have Ao > 0. 22.g. V[<t>] m 2>o o Fig. of the equation of motion? For these we must have all derivatives of <j) zero. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. . The EulerLagrange equation is d„ dC d(d.
The Hamiltonian density H is defined by the Legendre transform H[<f>.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig.2 The spontaneously chosen phase. We can convince ourselves that <fic is the field associated with a state of minimized energy. H = —C = V.22. like the solution x(t) of the simple Newton equation mx(i) — —V'(x) violates the invariance of this equation under time translations t —> t + 5t (i. 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>.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. TT = —J. w A iP 3ml 5* "XT" (22. i. we have 541 0. But for Ao > 0. aq> For (b = const.5) . 22. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. the equation has this invariance.2 The Constant Classical Field For Ao > 0. Fig. TTIQ > 0. the only such solution is the case we wish to consider.e.e. not the solution x(t)). and so ft[<M=m6:(^* 1 + Ao(0*0)2. 22. with 0 = 0. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. (22.ir]=7r<fiC. mg < 0.</>2)plane.
we set cf>(x) = <f)c + ri(x).3) obtained above. and we wish to investigate the behaviour of the field cf> in its neighbourhood.3). Classical Field Configurations the density Ti is minimized (i.g. (22. We examine this in more detail by setting <f>(x) equal to the classical cnumber configuration plus a fluctuation field n(x) which is again complex like <f>(x).e.6) Inserting this into Eq.2. e. 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.3ml = 2ml = ml + TA 0 A 2 = 0.542 CHAPTER 22.Then m\ m\ = ml + A0A2 = m2. For every value of the phase (3 we have a different constant cnumber field configuration 4>c.e. along the trough of V (which we can call a longitudinal direction). the one for (3 = 0. i. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . V?) (22.7a) (22. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. i. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. Suppose we choose one such configuration.7a) vanish on account of Eq. and partly by climbing up the parabolic wall on either side.7b) In Eq.e.2) and separating real and imaginary parts. i. (22. (22. This is the condition (22. in a transverse direction. as indicated in Fig. &H/d(fi = 0) if [m20 + ^>]</> = 0. >° for m 0 < °> . (22.tp2.7b) the constant on the left and the coefficient of A on the right of Eq. . one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 . Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.e. 22.
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. which. The elimination of the . tangential to the classical path. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. Thus the Green's function is similar to that in electrodynamics. We can see the problem here by looking at Eq. whereas ipi(x) is y ' " " " " " ^ . the existence of individual perturbation contributions). then a massless boson exists.r')S(t . 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. (3=0 Fig. We have here an example of the Goldstone theorem which says: If the solutions of the equation of motion do not possess a continuous symmetry of the Lagrangian.3 The spontaneously chosen phase seen from above. t.22. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. 22. There the wave function with an associated vanishing eigenvalue is called a "zero mode". t') = d(r .e. the component which climbs up radially outward along the profile of the potential implying a tp2(x) term with the potential V(</>) and so in the Lagrangian density £. i.7b). Goldstone's theorem applies to fully relativistic field theories.e. (22. has to be removed in order to allow a well defined perturbation procedure (i. ^ = h eip / \ ^2 l . 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. of course. irrespective of the question whether the perturbation series as such converges or not. r'.t'). though not identical phenomenon. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory.
C. To insure that the energy (see below) is positive definite we require V[$] > 0. instead we shall consider static and timevarying solutions in important models of one spatial dimension.e. We see therefore that the classical cnumber field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties.d^} = ~d^d^V[^) = ^2^(^j V[$].3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. Later we wish to develop a perturbation series about a classical configuration of $.t In particular the sextic potential can be considered along parallel lines. W. Lohe [179]. Khare [17]. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. we shall need only at a later stage. spinless field $(x. 7711 = — 1 ( 1 6 Minkowski manifold). J. (22. F. see e. equates to zero the coefficient of this wouldbedivergence.* We consider the theory (later: quantum theory) of a real. t Potentials of this type have been discussed by M. Soliani [237]. Behera and A. Rebbi and G. MullerKirsten [291]. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). The Lagrangian density is taken to be C[^.g. Zimmerschied and H.* Potentials of higher polynomial order in $ than $ 4 can also be considered. . *S. however. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. Classical Field Configurations other components results from the vanishing mass and the constraint called the gaugefixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution.£) in one spatial dimension. 22. A. i. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc.544 CHAPTER 22. We make one more assumption concerning V which. and we choose the metric ?7oo = + 1 . N.
but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.e. From Eq.22.e. In a quantum theory the field $ is an operator defined on a space of states. 1].t) has an explicit timedependence it is a Heisenbergpicture operator.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$. The first such condition to be imposed is that the energy of the solution of interest must be finite. the reason for this condition being that we visualize the classical solution as representative of a lump of energy.13a) We write such fields <j>(x). — m$ 1 — cos Im 1 I ^ $ (22. i.«{x) = V'(<p).l}. at least in the present case. 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). 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. (22.11) (22. (22. such that J > — — — = 0. (22. (22. ^ ^ * + V'[*]=0. i. (22.l].§ Unstable classical configurations are also important and will be considered in later chapters.12) we see that they satisfy the Newtonianlike equation <j. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. .12) where the prime denotes differentiation with respect to $ . Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.9) with quartic potential is defined by ^V\g$\ = ^V\g*.13b) We are not interested in just any solution of this equation. The other condition which we impose (and study in detail) is that of stability.g] = \ v m For instance. Since $(x. the socalled ^theory = ^V[g$. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. The field $(x.
(22. (22.T = <I>.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22.7r) where IT is the momentum conjugate to $ defined by dC ? r = . (22. d. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.14) where the overdot implies differentiation with respect to time t. .e.e. dC • Too = r?oo£ + =r$ = C + vr$ = H.c = <£z + . i. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. n = & . 22. We observe here already that the integrand of the integral in Eq.17) Since we have to integrate over all space.T^ = 0. HI tanh m(x>k) 9 Fig.15) The zerozero component is equivalent to the Hamiltonian density H.546 CHAPTER 22.2\dx 2 ' 1 fd& + V[$].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>. the classical solution will therefore have to be such that this integral is finite. (22.4 The wellknown soliton of $4theory. <9$ i.
4m .11).20) (22. (22. i. 22. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .1 . The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq.777.18) is depicted in Fig. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n".x n ).5 The soliton of sineGordon theory.21) This sineGordon configuration is depicted in Fig. "The name sineGordon is derived from the analogous KleinGordon equation or theory.4. 22.13b) for the specific potentials (22. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. the $*theory.5. m ( x .22. it is not difficult to find classical cnumber solutions 4>c to Eq. — XQ). In fact. See J. in the case of the first. 22. O±m(xxo) (22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . Rubinstein [240].3 Soliton Theories in One Spatial Dimension 547 As we shall see below. .18) Here XQ is an integration constant. < = — tan [e >  ° ] 9 2 rrm/g — Fig.10) and (22. (22. > .e. (22. Again XQ is an integration constant.
± 1[tanh.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. or \(<P'? = V(4>)+ const. and hence 4/71 9<t>{x.20) are obtained from Eq.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . <^>(x) = ± dx x=g4>/m \ / 2 ( l — COS X I Am dx lx x=g4>/m A / 4 s i n 2 ±1 In tan . With we obtain g(/>(a. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ.1 "»JtKxo) 9 1 . (22.o) =m7r. . 4m In tan tan 5 l/e±m(*xo)tan^£^)U. Inserting the potential (22.13b) which we can write d \W? = !"(•».* ( m' m 2 r)] ^2" .^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). In the case of the sineGordon theory defined by the potential (22.548 CHAPTER 22. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .e. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .9 2 4>{x) = i.18) and (22. Classical Field Configurations The solutions (22.
22.21) for the energies can also be obtained from Eq. this does not require the action to be minimized. in the onedimensional Ising model. for instance. 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. Jackiw [141]. instead of doing this now. 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. (22.17) 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) . (22.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks".4 Stability of Classical Configurations The concept of stability can be complex. the discussion of Bogomol'nyi equations) in a simpler way.17). however. Then we have to demand that the functional derivative of the energy E(<p) be zero. we obtain these expressions later (cf.22) and that its second variational derivative be positive semidefinite. The expressions (22. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. i. SE((f>) 5cb(y) 0.22. (22. A state of stability is therefore associated with minimized action and/or energy. . (22.19). The relation (22.e. Suppose now that we demand stability in the sense of minimized energy. the others as antikinks. the curves rising monotonically from negative to positive values are known as kinks. However.23) See for instance R.
the function 4>c(x + a). Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. if 4>c{x) is a solution. Eq. and the EulerLagrange equation retains this property. The Newtonlike equation (22. Thus the vanishing of the first variational derivative yields again Eq.28) . it is also called a "translational mode".e. namely (p'c.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. i. (22. In fact.a / 0 is not the same.550 CHAPTER 22.13b) with classical solution 4>c. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. if we apply the generator of spatial translations d/dx to the classical equation.13b) retains the invariance although the solution (f>c does not. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. 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. of course. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). (22. Since it arises from the violation of translational invariance by 4>c. written down in Lorentzinvariant form. dx (22. (22. (22. dx2 + V"{4>) (22. in particular the invariance under spatial translations. the integrated contribution vanishes. its eigenvalues w\ > 0.A. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. a very general phenomenon which can be established as follows. The eigenfunction 4>'c{x) is for this reason called a "zero mode".25). i. = 0.e. with eigenvalue w2. (22. we see that the latter has one eigenfunction. This is. Classical Field Configurations Since y ^ ±00.26) Comparing this equation with Eq. i.13b).24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite.e. in fact. The original Lagrangian was. in fact ip0(x) const.
i. In the onedimensional case under discussion we can rewrite Eq.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. / oo Thus for E(4>c) to be finite. lim [V(<f>c) + const.e.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.22. this integral has to be finite.33) x—>±oo .]. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero. (22. (22. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. We can choose the potential V such that the constant is zero. We see therefore that the occurrence of these zero modes is a very general phenomenon.18) and (22. and that S[UC] is finite (or equivalently the energy). 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.e. i. (22. = 0. (22.17) we obtain oo dx[V(4>c) + const. we are interested in classical cnumber field configurations of finite energy. Thus TUC is a zero mode. We return to our considerations of stability. in fact.] = 0 or x—>±oo lim V{(/)c) = 0 for const. and we have seen that Eqs. (22.32) = V'(</>).20) are such configurations in two particular models. it is an expression like (22.31) Inserting this into Eq. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). As stated above. 0 = S'[UC] = S'[gUc}.27). Then the vanishing of the first functional variation implies that 5S = 0.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22.
10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. ± l . Thus the classical energy or vacuum configuration is not unique. 22. 22.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. (22. » In the <fr4theory defined by Eq.35) and it is seen that in this case q can be ± 1 or 0. n = 0 . the classical field (bc must approach one minimum for x — . Classical Field Configurations V Fig. the potentials have degenerate minima (i. ± 2 .6 The minima of V in $4theory at x — ±oo. In the case of the 3>4theory we can define a conserved current k^ by (22. . In the sineGordon theory > • defined by Eq. oo J—oo y The topological quantum number q is defined by 0.552 CHAPTER 22. .e. V{4>) = 0 at different values of </>). . This number is defined like a charge in field theory by the spatial integral of the time component of a current. (22.1).34) jfc" = e^dv(l since d^k^ = 0. (22. The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22.6. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory.c o and another for x — oo. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. in view of the antisymmetry of eM^. . For a configuration to have finite energy. .
7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo).A2)2 + 0 except for A2 = 1.11) are given by ( so that for x — ±oo: > — J. Again we see that E(\<f>c) is finite only for A = ± 1 . . x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition.cos(27rnA)]. First. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. continuous values. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. Tfl \ (22. . 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 .7T (22. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. . one may ask how topology comes into the picture. . and exp(ix(0))<pc reduces to 4>C when x(#) = 0. In the case of the sineGordon theory the minima of the potential (22. zero is not a meaningful topological number. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. in general.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . and we call this "topological stability. A property is.38) 4 V(\<t>c) >m ~2[l . termed topological.37) for the onedimensional case here. n = 0 . 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. We can see this as follows.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. ± l ± 2 .22.
We observe (a) the topological current k^ of Eq. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.40) (22.554 CHAPTER 22. Classical Field Configurations involves the difference An = rioo — n. (22.43) **See e. Thus with one space dimension higher there. W. (c) the time component k° is a spatial divergence. Eq.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. (22. (22. if oo / dxy/2V{<l>)<f>'{x). (For a broader view we consider briefly in Sec.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. 00 (22. 22. 425 .9 the foregoing arguments in a higher dimensional context. The inequality is saturated. the energy is minimized.428. 22.41) dxy/2V{$)4>'{x). where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. J.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22.31) to (22.^ = N.g.42) Thus the energy has a lower bound given by the right hand side of this inequality. i.e.17)) f 00 > 0. (d) the topological quantum number iV arises in a completely classical context. MiillerKirsten [215]. H. the factor A = exp[ix(#)] varies over a complete circle).** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). (22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). . 00 (22.13b) or ^ ' ( * ) ] 2 = V(</>). Eqs. its integral is nonzero only on account of nontrivial boundary conditions. pp.
44) This first order equation which saturates the inequality (22. (22.^ Clearly it solves the second order equation or (22.5 Bogomol'nyi Equations and Bounds We observe that (cf. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. 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.39).21). for instance. Bogomol'nyi [27]. 555 (22. (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.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi. n E . . We discuss this in the following Example.22.n > 1.40)) this occurs when 4>'(x) T V2V(4>) = 0. 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. Eq.19). B. 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. Here (V(c/)) > 0. In the sineGordon theory we can consider.
Classical Field Configurations Example 22. 22. Construct an expression for the energy of such a classical system and its Lagrangian.e.t) <j>(xn. Assuming that the pendulums can move (i.cos</>(x„. Solution: Consider a string along the xaxis as described.7.t) 4>{xn. } n= — oo roc —> / L J .1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.t) measured e.556 CHAPTER 22. rotate) only in the (y.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i.g.7 The pendulum analogy. Initially the pendulums are suspended freely in the gravitational field. Chapter XI. 22.g. 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.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. xn+\ — xn = a) 00 a —> dx. Fig. H. 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. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. z)plane and only such that their elastic strings remain taut. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) . ka —> K. Goldstein [114]. Assume that the pendulums can move (i. all being identical.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . We can now write down an expression for the energy of such a system of (say) n pendulums. At equidistant points xn = na along this string we attach strings with pendulum bobs.e. rotate) only in the (y.t)\2 +mgr{\ .46) **See e. t)).2 V dt + k[(f>(xn+i. so that oo > u. from the yaxis as shown in Fig. with mass m and connected with short strings to neighbouring bobs. . V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na.
e.13b) we have for small fluctuations rj(x. (22. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. ' Equating coefficients of exp(—iw k t). 22.25) obtained earlier.7 linking the pendulum bobs). 22.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. If some w\ were negative. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). and so would invalidate the procedure which assumes that rj(x. (22.47) where n(x. (22.50) This equation is identical with Eq.22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. of the continuous curve in Fig. This number. It is instructive to obtain this equation by yet another method.t) = <l>c(x) + r}(x.t) = ° (22'48) (22. (22.t). We can understand what stability means in the present case. Inserting (22. (22.49) exp(iwkt)ipk(x). t) is the fluctuation (field).49) would imply an exponential growth in the future t > 0 or in the past t < 0.t): {w With the ansatz r)(x. t) of Eq.13a) and (22. (22.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. The equation is called stability equation or equation of small fluctuations. the topological quantum number. Hence we set ${x. Eq.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.47) into (22.t) is a small . which represents an expansion in terms of normal modes. is therefore also called "winding numbed. wk would be imaginary and so the factor exp(—iwkt) in rj(x.
51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . V"((/)c(x)). and need 9 = tan" 1 j e±m^Xo) 1. Classical Field Configurations fluctuation. Inserting (j)c of (22. we shall see later that the fluctuations n(x. i. In fact.e. The equation of small fluctuations (22. V"(<b) = 2m2 + 6g2(f>2. we obtain V"(<t>c) . 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£^)*.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction. This means that the stability is not universal but only local in a certain sense. In fact. . the zero mode. (22.1. For x — ±oo the quan> tity acting as "potential".52) tan 6 := e±m^xo). we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. If that is necessary. i.A \ Inserting <f)c of Eq. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists.2 m 2 + 6m 2 tanh 2 m(x .e.20).558 CHAPTER 22. Thus for stability w\ must not be negative. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory.18). cos46» = 2cos2(2(9) .50) has the general form of a timeindependent onedimensional Schrodinger equation.
A  *(* + !) cosh 2 z tp(z) = 0. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. (22.e.53) = m(x — (22.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. We can therefore construct solutions of definite parity.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 . and as a consequence a limiting form of the small fluctuation equation of all basic potentials.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation. (22.ijj^ respectively. i.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.2 (22. which satisfy the boundary conditions V>(o) = o.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable. — XQ)] cosh[±m(x — XQ)]: 1 . We can investigate the spectrum as follows.55) The differential operator is even in z. ip+. . even and odd solutions. ^+(0) d 0. Those of the other case are complex and periodic at infinity.* For such a potential the spectrum can be b o t h discrete and continuous. Consider the equation fd2 dz2 .22.
Z)2.( 2 n . i. A * An = .e. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . when in the even case n ( n . i. 2n < Z. A ^ An = .Z ) + ^(A + Z2) = 0.60) terminate after a finite number of terms.e.57) Then the equation for x{z) 1S (22. (22.e.l2 = . .60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.4 n ( n .Z ) 2 .560 CHAPTER 22.58) = (1 + sinh2 z)~l/2x(z). (22. Classical Field Configurations We now set (with ip either ip+ or ip) tp(z) = (cosh z)~lx(z) and change to the variable £ = sinh2 z. + l2) = 0. and in the odd case n+\)U+\l)+\(\ i.59) Setting (22.e.( 2 n + l .Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. i.61) Discrete eigenvalues are obtained when the series (22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n.I) .
11).50) we obtain with z = m(x — xo) tp(z) = 0. i.65a) _dz2 cosh 2 z l=2.2. w2 = Am2.63) we see that the continuum starts at A= 2 2^ = 0. We can now derive the spectra of the stability equation for the potentials (22... d w = tanhz. N = 0.10) and (22. (a) ^theory: the equation Inserting (22.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0. the continuum is w > Am2.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz. for N = 0.54) by setting X = k2 Then 1 .. 2 . We can see the continuous spectrum of Eq.2Z = .is normalizable provided 2(n + .62) we see that the equation possesses discrete eigenvalues w —2 . (22.( 2 . (22.\=Kl2 From (22.N<1.51) into (22.1: w2 = 0.1. m i.u2)—^V 2 dio and ip(z) = elkzV{to). From (22. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 . (22. 3m 2 ..22. m N<2.Ny. (22. (22. °2 i.6 The Small Fluctuation Equation 561 and the function ip.e.62) where even (odd) iV are associated with even (odd) eigenfunctions.e.j < I .e.
m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. 22. the zero modes. also be deduced from d(j)c/dx. In the $ 4 theory we saw that (cf..2 The discrete eigenvalues are now given by w ^l m^ = (lN)2.562 CHAPTER 22.e.. Thus either spectrum contains the expected eigenvalue zero. 22.=1A=^_.66) We see that this is a nonvanishing function for any finite value of x. Eq. This property can. We know that 4>c is a monotonic function.8 The zero mode as typical ground state. Classical Field Configurations (b) SineGordon theory: Inserting (22. so that its derivative is nowhere zero (except at ±oo). .50) we obtain with z = m(x — XQ) the equation (22.53) into (22.27)) . %(x) Fig.65b) dz2 cosh^J. i. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. . Thus 4>'c(x) has no node.8 and we see it has the shape of a typical ground state wave function. It is now particularly interesting to look at their wave functions. (22. . and — as expected — this eigenfunction is even. m d . N<1. The zero mode is depicted in Fig. for N = 0: to2 = 0 and the continuum starts at w2 = m2. . of course. . d .
6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . — XQ) OC Hence the conclusion is similar to that in the previous case.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 . .68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. (22. without quantum corrections). d . i.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a.oc (22.22. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. Similarly in the sine—Gordon theory we write so t h a t 4 /oo * " / > & • ) W171 dx oo cosh m(x — XQ) TYl = 4 — [ t a n h m O r .e.= E(<f>c).70) . . Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.e.t a n h x .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).z 0 )]oo = 8 . . Finally we consider the normalization of the zero modes. 1 o rrr tanh x — . (22. (22. .
Derrick [68]. it is important to understand their physical implications.72) with T{$) = I' ddx]{V(j))2 > 0. . Kastrup [145].> 4>a{x) = 4>(ax).73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . J As a first example we consider the expression (22.e.564 CHAPTER 22.4) <j>(x) —• a<f>(ax). I.g. Goddard and D. ddx {V4>f + v{4>) (22. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. (22. Affleck [5] who uses in his Eq. P(</>) = f ddxV{(f)) > 0.2d. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . however. we consider E(<P) which we also write as (22.T(4>).g. H. They may help. *The considerations of this section extend in part beyond quantum mechanics and into field theory. A. for if such configurations exist in a realistic theory with three spatial dimensions. One can have different types of scale transformations — see e. P. *See e. i. H. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces.g.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. This question has important consequences. (2.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. to underline the fundamentally different nature of topological and nontopological finite energy configurations.41) for the energy of the classical configuration but now with respect to a ddimensional position space. Olive [113]. ^For basic aspects of scale transformations see e. f G . Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). Classical Field Configurations 22. I.
78) For the energy of a static configuration.e.. 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}. i.V(<P) (22.79) In order to emphasize that (f> — (p^ in any direction. we set 0 dE(<pa) da J a = i (22.75) Since both T and P are > 0 and d is integral.71). (22.d)T(<f>) = dP{4>). x = (x1. a=l (22. we write • = lim J r .22.e.80) . Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. the second derivative is positive) only for d = 1. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).74) From (22.. to be finite the field <P ~^ 4>oo £ Mo for x — oo. (22.. Consider first the Lagrangian density £[&M = \{d^)(d^) .e.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. Next we consider more complicated cases. d = 1 minimizes the energy.73) we see that this implies (2 . 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.76) we see that E(<f>) can be minimized (i.xd).x2. (22. (22. since d2E(<pa) da2 2(2d)T(<f>). Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension.\ eMo(22. In fact.. i.
38). —00. i. — 1). Thus in this case S^1 is a disconnected set consisting of two points. that we have a set of disconnected classical vacua (</>oo><^oo. Table 22. are the same. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. A correspondingly topologically nontrivial field theory would therefore have to possess a!7(l) symmetry as in the complex Higgs model (recall that in our considerations of $ 4 . is a connected region. If we go to two spatial dimensions (d = 2).81) and (since d = 1) S = ± 1 . x = 00.and sineGordon theories we assumed $ to be real. one set can be mapped into the other. These observations are summarized in Table 22. in fact. ( with — = lim <pc(x) 9J *>±oo TYl \ (22.1.e.e.566 CHAPTER 22. i. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. —00 or equivalently S = + 1 . and. i. as is the case for the constant solutions. of • course.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. (ptx = <f>oo. 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). the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. S1.e. We see.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . therefore. —1. . Then each point at infinity is characterized by a different value of the polar angle 9. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00.
map into.e.g. the field must map position space into field space at infinity. Thus. i. i. if cj> were real and position space twodimensional. the one point S = + 1 . given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$.e. Thus. assuming invariance of C under the transformations of the group SO(3) in internal space. On the other hand. —00 or S = 1 .2. we would have only two points at infinity.83) . This will be "nontrivial" in the above sense only if 0oo is independent of 9. for instance. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. 9) would depend only on r = x and hence the problem would be effectively onedimensional. $(r) : Sl .e.7 Existence of Finite Energy Solutions Suppose. * n reach any other <^(oo) by simple phase shifts.3 (^real) v{4>) = / (22.e. We see.• M0. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. If position space were onedimensional. i. by allowing <fi to have three components in some "internal" (isospin) space. phase transformations 6Q ^> 9Q + 59. that if the problem is to be topologically nontrivial. (22.82) Now Mo := \ 4>(oc) ^24>2 = a2\. VMx)) with (0(oo) = linvxx) (p(r. therefore. . Similarly.e. i. i.22. to have topologically distinct classical configurations. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. cj)(r. i. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. 00.e. we have w e ca {Mx)}> and 1 3 1. if <p is real.1 . This can be achieved e.
Summarizing we see that </>(x) is a mapping from S^T1 to <S^1.e.e. A) + T{A) + P{<j>). Georgi and S. the appropriate rotation. it is effectively a surface (and so a continuous set) S2.V{4>a]. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM.568 CHAPTER 22. 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. Olesen [220]. .e. a = 1. the subsequent discussion requires only the general form of the Lagrangians.a M J . + i ( D " f t W ) a . i.2.. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. Example 22. $ — $ + 27 > • T in sineGordon theory.3. $ — > > $ exp{i Xlo=i Pa. 11 H. If. . circle.„ where G r = 9 M £ .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) £(</>. so that details of these models are irrelevant. invariance under replacements $ — —$ in $ 4 theory. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. A) = T(4>. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. the winding number is n.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". Nielsen and P. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. we cover <S^_1 n times.e.</>] = ~G^Gai. % . i. The symmetry of the potential (and Lagrangian) — i.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. Glashow [108]. $ — $exp(ia) in a U(l) symmetric theory. through each point at infinity). B. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3).. L. going once around S%~1 (i. sphere.
d)T(4>.Ax) dX Since T((f>..d T ( ^ .A). >• A„ x ( x ) = AAM(Ax).22. A).84) and investigate the existence of finite energy classical field configurations. Thus we set dE(<fix.d T ( A ) + AdP(</>).dP(4>) = 0. ^ A ) = A 2 .87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. (i)^ 2d T Considering the contributions of (22.86) (22.i T(AX) = X4~dT(A). i.84) in this way one finds that the quantities involved transform T ( ^ . A) + (4 . it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . (2 . T(A) = f ddxV(4>) > 0. (22. (22. and so B ( ^ A . A. Mx') = E with a scale transformation x — x' = Ax. = 0.7 Existence of Finite Energy Solutions where T{<j>.84) is positive semidefinite and so must be finite by itself. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.e. We now consider scale transformations with A £ R.89) P{4>) are positive semidefinite and not all zero.e. i.T{A). > Since T>IJ. (22. (22. (22.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 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'>. we have {^/»)A^ (22. some contribution must be .87a) C?M„(x) — \2G^{\x).<^[x) transforms like a vector. Solution: Each contribution in (22.d)T(A) . J 4 ) + A 4 . negative. l j ddxGijGai] > 0. A A ) = A2dT(^.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). <0 >• <£A(X) = <KAx).
A) or P{4>) for d = 3.89). (22. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. Landau [111]. Thus such a contribution could counterbalance T{4>. (22.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. exemplified by soliton and sine—Gordon theories. 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. ft V . The GinzburgLandau theory assumes that the static energy density is then given by £W0 = ^ W ( x )  2 + 7 + \aMx)\2 + \m*)t (2291) The parameter (22.570 CHAPTER 22. From this development one knows that in the superconducting state of the metal the electrons combine to form pairs called "Cooper pairs'" which then have bosonic properties. (22. L. of course. Ginzburg and L. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity.89) the contribution (4 — d)T(4>).** e. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. One can thus define a scalar field or wave function ty(x. Skyrme [253].g. T^{<t>x) = \4dT(</>). (e) For d > 5: In this case it is not possible to stabilize Eq. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. H. J. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. W. D.90) and we would have to add to Eq. a > 0. 22. is temperature dependent and changes sign at the critical temperature Tc. . a = a(T) = a T —T c . We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory.92) The expression ^ ( x ) ! is known as the "order parameter. Tchrakian and H. (22. We then have a situation of the potential like that described in the context **T. In the following we consider a related theory. MiillerKirsten [268]. R. H. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example.t) such that  ^  2 describes the density of Cooper pairs.89).e. (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). This case is. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models.
94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ .(tp) = 0. (22. from x < 0) means — as is familiar from electrodynamics — replacing V by V .^ 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. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. i.94) to check the dimension) 1 = I—a. but for T < Tc it has a double well shape.93) J' (22. The parameter 7 serves to put the minima of the double well at V = 0.2): For T > Tc the overall potential has a single well (minimum). Clearly the length (see Eq.22.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. assuming there is no superconducting material in this domain). ^(00) = a (22. which we recognize as one half of the kink solution we obtained previously as the topological instanton solution of the Schrodinger equation with double well potential (in the present case ip(x) = 0 for x < 0.e. Then the static energy density becomes (cf.e. Example 22.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. 22.2) £(^. 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. Variation as before.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. setting S£. Sec. is a measure of the distance from the surface of the superconductor to where . i. (22.
h [^DV'V'DV].g. 2mi Setting < / > \P\e vp B = ViA. 450. 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. MiillerKirsten [215]. (22. exhibits a completely different behaviour — in fact. as indicated in Fig. the order parameter \ip\ attains (approximately) its asymptotic value. i. The magnetic field on the other hand. V x M = j s .96) ipH.98) where M is the magnetization resulting from atomic currents. (22.572 no superconductor CHAPTER 22. We can see this as follows. yJ—a/[3.9 Behaviour of ip and B in the superconductor. i.e. H.M. W. Classical Field Configurations superconductor (x>0) Fig.e. In the Maxwell equation See e.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 . i. p. The density of the Cooper or super current is given by the typical expression of a current.9. J.e. (22. . 22.
5 to 10~6 cm). (22. and V x (V x B) = / J 0 V x j s .100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. K i ™*. those of the socalled type I and those of type II. There are two types of superconductors. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 .L B = O.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. This is zero in the present case. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field. Using the relation "curl curl = grad div . (22. Also dD/dt = 0 in a static situation like the one here. In the onedimensional case we have B{x) = B(0)ex/XL. Hence 0 = V x H = V x ( — B .M i.e. V x B = [i0V x M = /lois.99) Inserting (22. this becomes A B = no V x j 5 . Approximating p by its equilibrium value. m xi A B . i..div grad" and V • B = 0.22. These vortices therefore carry magnetic flux which is necessarily quantized .e. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. 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.
i. 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). 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. *A. A. De Gennes [67]. (22. See also P. we must have that lim V ( H ) = 0. (22.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22.2 — 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. of two points on a line. in fact a periodic function of the polar angle with period 27r. Finiteness of E requires \<t>\ = \<fioo\ on this circle. (22.9) ^e^ '>\<poo\. i. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. .101) (p(r. The existence of these vortices has been confirmed experimentally. 'See the remark in H. We can think of (e. where exp(ix(0)) is an arbitrary phase factor.e. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. the space consists of two diametrically opposite points on the circle. G. in other words.36). In the NielsenOlesen model defined in Example 22. Classical Field Configurations (see below). 22. J. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).g) twodimensional space as being bounded by a circle at r = oo. A.37).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 two models of Example 22. Abrikosov [2].9).102) In the onedimensional soliton case of Eqs. Schaposnik [69] after their equation (3.574 CHAPTER 22..e. de Vega and F. Thus the asymptotic field </>(r.
The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. commutative). In mathematical language S1 is therefore a compact. Fig.e. . 2n] remains continuous if the function assumes the same value at the two endpoints. The 1sphere is the unit circle in the space of all complex numbers. 0 < 0 < 2TT.X ( 0 = O)]. Suppose <j>. (22. Finkelstein and C.22. topological group with the usual multiplication as group operation.e.1).10. X is called the domain of / and y its range.* 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. W. (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". The two theories we considered above in Example 22. Naturally an integer cannot change continuously. 22.1. i. defines the orientation of this line element.118)). 22.10 States of winding number 0. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. Thus smooth deformations *See D.102) is called the exponential map of S1 onto S1. We may define this as « = ^[X(0 = 2 T T ) .2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. The map A : S1 > S1 defined on S1 C R2 by (22. abelian (i.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. Sl := {zeC\\z\ = 1}. Misner [96]. Then a continuous function of (j> on [0.
A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite.. The gauge field required in the asymptotic limit (22. one hole. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example).e. 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. i. no hole. if it has one twist as above or a hole. elements in the same class have the same topological characteristic.) remain the same under homeomorphic transformations and are therefore called topological invariants. the winding number must be a constant of the motion. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n.104) so that d2x hJ Since A has direction e#. r^oo 1 dx(9) n r o0 (22.105) is described as "pure gauge" since it is determined entirely by the phase x(#). The different kinds of connectivity of these topological spaces (e.105) the corrections falling off faster than 1/r.. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below).g. The winding numbers n are such topological invariants. For that reason n is called a "topological invariant?.. finite. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism.i'ldx(0) 9A Thus for (22. i. we required that lim r—>oo w o.576 CHAPTER 22. Thus two homeomorphic spaces have equivalent topological structures. (22. In other .e.g.104) to be finite we must demand that g . The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. connectivity. we can write 1 d h J ee rde w At h r^oo . i.e.
105) insure that B ^ O . Ffiu. the vector potential A. we can consider transformations which describe continuous transformations of a configuration with fixed n. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).e.e. i. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = .105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22.110) . 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 higher order contributions in (22. into S i . cannot be deformed continuously into one with two loops (n = 2). (22. Of course.e.9 Introduction to Homotopy Classes words.106) J g j r d d g Thus the magnetic flux is quantized. if A = V x . Therefore. it is determined by a gradient (l/r)d/dd. cannot be pure gauge everywhere. (22. An example is the mapping H(t. It follows that the field Fij. 9){^ and hence the functions (22. i.108) map the spatial circle S^.107) The fields <f>(r. is also zero at r = oo. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.9):=t(f>£\9) + (lt)(f)W(9) with te[0. the number of flux quanta 2irh/g being the winding number n. the group space of (the phase transformations of) U(l) continuously. (22. we have B = V x A = curlgradx = 0. going once around S^. However. the gauge field.109) A W(0) = e^ W.22.<b rd9—4^ = n. with n = 1. However.l]. we can go n times around S^. for B / 0 . i. A field configuration with one loop (or twist).
3: Homotopy an equivalence relation Verify that the homotopy map H(t. Reuter [77]. Example 22.111) as a map from the space described by 9. these two maps (fin from S^.e. Classical Field Configurations which is continuous in the parameter t and is such that H(0.lQSl^S^ with H(0. The relation (22.g.e.e.9) is called a homotopy (emphasis on the second syllable). We can consider (22.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). f T h e standard texts on homotopy theory are N. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). like <pn (0). Hu [134]. The map H(t. A brief introduction can be found in the very readable book of W. (fin (9) in the above example.e.8) itself must be a configuration with definite winding number. i. the phase factors of (f)(9).9) = <fiW(9) and H(l. 9) is an equivalence relation. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. it is clear that this is also a homotopy. S%. — S\ > (22. i. e. H.1].578 CHAPTER 22. i.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity.T. we have to show that the map satisfies the three properties of (a) identity. Since the passage of time is continuous. (b) symmetry and (c) transitivity. Dittrich and M.113) .e.0) = $\6). H(1. A superposition of classical configurations with different winding numbers does not minimize the energy. i. (b) symmetry and (c) transitivity. the unit interval I = [0. i. time cannot make a field jump from one homotopy class (see below) to another. In fact. n' ^ n. But this type of combination is excluded sind H(t. (22. We verify these in the following example.110) could be written down for configurations with different winding numbers. Solution: As stated. 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. to the internal group space of £7(1). 8) = cfinl\9).111) In principle an expression like (22. (22. Two curves which can be related to each other in this way. Thomas [270]. Steenrod [261] and S. n. and the space of t.9) = ^°\0). Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). H.
so that H(t.t. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. below). i? is a homotopy connecting </>(°) with <^>(2'. 9) = H(l. is isomorphic to) the set of integers Z (positive. being the continuous map connecting </>(°) with <j>(1'>. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). ' .9) = <f>(0\6) and H(l.e. 0).0) = </>(0) is continuous there. 9) connects <pn (0) with itself continuously. these groups are then called homotopy groups (cf.e. i. 0) = <£{2).9) = . <f>W .* 22. negative or zero).ffi(O. U(l)). we see that of the set of these maps some are equivalent and others are not. i<t<l. Homotopy classes are sometimes also called Chern—Pontryagin classes. Returning to the maps \(x) : Si ^ S1^ (i. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. Moreover. 0) = ^ (0) and 6(1. 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. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. is such that H(0.9) H(0.0) for for 0<t<.22. . Then H(t. as they are called. which therefore demonstrates that the homotopy relationship is transitive. 0). Hence <f>(0^ ~ <p 2 ). 9) = H(0. On the other hand H(0. a class being a set of equivalent maps. (b) Next we check that if 0<°) ~ <p^. 9) := H(l .0). then </>W ~ <t>{0)• We define an H(t. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). . at t = 1/2 both i?i and H 2 give c/^1'. 0) = tf2(l. <^(2). \ M)\ H!(2t.e) tf2(2il. We now define the function H ( f 0 ff . 0) = 4>W (0). 9) = <f>W (0).e. since and tf (1.10 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. (c) We can demonstrate the property of transitivity as follows.
which is to be mapped into another circle Si indicated by dashed lines. to make the result plausible. S 2 the surface of a sphere in a threedimensional Euclidean space. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. but TTI[S2] = 0: S 1 is a circle. How can we see this? We first sketch the idea. p. 22. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9.e. n=0: t=1: t=1/2: i. Rajaraman [235]. T (22. 52.580 CHAPTER 22. Fig. and then introduce some of the related mathematics.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. .g. 22.11 the circle drawn with the continuous line is the circle S%.11 The trivial n = 0 map allowing shrinkage to a point. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. T < 9 < 2TT. Sketch of idea demonstrating that nilS1] = Z.115) Varying t € [0.1] continuously to 0 one regains xo(#) = 0(b) In the cases n = 1 and n = 2 we have Xi(0) = 9 and xi(0) = 2e for al1 e  Similar considerations are given in the wellknown book of R. In the diagrams^ of Fig. 22.) why 7i"i[S2] = 0. In order to really appreciate how topology comes into the picture here one should understand (e. Appropriate maps.
2 the maps are wound around the dashed circle. 22.e. 0 < X(P) < 2vr. . i. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. For winding number n we have the mapping (i.117). (22. 22. and hence the winding number.e. the transformations^ (the result may be looked at as 8n of expression (22.117) in ±. i. of (22.s.s.116) In fact from this we obtain (see below) 2  n and this means r.h. 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.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. to the case of winding number zero.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.h. One can prove that the expression (22.e. (22.117) 2^ d6X(x)\1(x). Coleman [55]. We make another observation at this point. is invariant under infinitesimal (continuous) deformations.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. The maps for these cases are illustrated in Fig.12. d (22.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).22. In the cases n = 1.
Sec. Thus there is only the trivial map into a point. only one homotopy class. Classical Field Configurations Under this transformation n changes by 5n Now. In fact.8. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point.582 CHAPTER 22. 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. i. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided.e.e. 15.e. there is a theorem which says thatH TTg(Sn) = 0 for q < n. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). whereas 7ri(50(3)) = Z2 (i. the cyclic group with two elements (since 50(3) is doubly connected). 1 1 See N. an integer modulo 2) i. . Steenrod [261]. the identity class (the group space S2 is simply connected.
and in particular the exponentially small level splittings of the double well potential. 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). different from literature. 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. 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. 23.e.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. 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. Since we have not yet introduced periodic (i. in such a way that the factors appearing in the result are exhibited in a transparent way. 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.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.Chapter 23 P a t h Integrals and Instantons 23. We consider such a theory now with action S((p) depending on the real scalar field 583 .
i. (23.2.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. Chapter 6) 1 2 .e. Gildener and A. in particular E. ! " ^ j = ^ 2 . Fig. Path Integrals and Instantons S(<f>) / dt \i>2 .1 The double well'potential. note that here the mass of the equivalent classical particle is taken to be mo = 1. and ^2(0) : m* . "For comparison with literature. Patrascioiu [110] and Chapter 18.2) Fig. 23. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.* /•tf CHAPTER 23.v{4>) <?VV m4 __ 1 . (23. n = 0.1. 23.584 <)>{t).1)) and hence E0 = mh/y/2. The Hamiltonian of the simple harmonic oscillator (cf.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . 2 2 4 ™ = ~.
2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. 23. This tunneling affects the eigenvalues.g.2 The wave functions of the lowest states. 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.x) = x2 V(x).e. The states of the system here must be even or odd under x — ±x. 23. and in the above 4> corresponds to x. e. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . as we did in Chapter 18 — but as we saw.23. W(0) = 0. this alone does not yield the level splitting. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. </._(0) = 0. must have definite parity as in a physical > situation. we have the Lagrangian L(x. We can develop a perturbation theory around either of the minima.In mechanics of a particle of mass 1 with ~ position coordinate x(t). i. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering).2 — and we can deduce from this the boundary conditions. but there is tunneling between the two minima if the central hump is not infinitely high.
£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).  . dr We observe that Eq.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.2 V'(<l>) = . (23.8) V(4>) + const.2. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.0o.3) for tf — oo.7) so that (23. (23.e. Euclidean time T = it. Equation (23. classically forbidden domains then become "classically" accessible. i.e. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.e.U) = J'D{<f>}exp[iS(<l>)/h] (23. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. that given by Eq. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.5) and iS = dLE d dr[d(d(f)/dT)} i.586 CHAPTER 23. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g. dLE = 0.6) d*l dr 2 . But quantum mechanically it can tunnel through the hump from one well to the other. is finite if we go to imaginary. (23. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. For small h we can attempt a stationary phase method. The transition amplitude for this.4) J <bi =—(An rTf (ITLE = I JT.e. dr m+™ (23.7. r . SE. i.3). i.2 m 2 ( l m z (23.<h.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo. In the semiclassical path integral procedure that we want to use here. (23.9) .tf\ .£'(0) = 0. Then (0o.
2 Instantons and AntiInstantons 587 In fact.e.e.e. is depicted in Fig. 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). from fa — —rn/g. we see that the equation is the same as the classical equation for a static soliton (i.18). i. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). Fig.3 The instanton path. . this means the total energy of the particle in Euclidean motion is zero. 23. — (23. 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. Its trajectory. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously.23.3. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. the classical path. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton". in the present case (T — To) = m tanh m(r — TO). 23. i.10) where TO is an integration constant. and hence kinetic energy gained is equal to potential energy lost.
10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .^ . Using Eq.7) admits another configuration which is the negative of (23. (23.' . 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. 23. This antiinstanton configuration rises monotonically from right to left as shown in Fig. (23.13). The configuration (23. thus indicating the nonperturbative nature of the expression.10) is given by (23. Inserting the result (23. (23.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .TJ T fro.4. the constant must be zero.13) into Eq.8) we obtain T f SE dr[2V((j)c) + const.7"/! ).588 CHAPTER 23.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.5).10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.].Tj) oc exp 4 m3 m^2 (23. but with the same Euclidean action (23. Tf = oo. The classical equation (23.11) For this quantity to be finite for TJ = —oo. we obtain lim f+ OO.4). Path Integrals and Instantons The Euclidean action of the instanton solution (23.13).r 0 ) ' (23. We also observe that the expression is singular for g2 —> 0.13) 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 22.14) But this is not yet enough.
so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.15b) since we require configurations from vacuum to vacuum. . 0c(r) = &(T).15b) a solution of the classical equation? Is it a topological configuration — if so.g. as we can see e.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . (23.TO far apart and T'Q 3> TQ.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r).15b) is not exactly a solution.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.15b) has the shape shown in Fig. b u t if TQ.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig.5.23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g. The sum (23.5. / > Looking at the instanton (23.15b) In the case of (23. 23. e. 23.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig. #(r) = </>C(T . 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. from the sketch or from the variation of SE. in the case of (23.g.4 The antiinstanton. (23.TO) + ^ c ( r .r0).15a) we have a simple superposition. 23.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. Clearly we are here interested in configurations of type (23. 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. and an antiinstanton localized at (say) TQ.
Thus it belongs to the socalled vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero. 5^ In order to answer the second question we observe that (23. g cosh m(r — TQ) (23. Ignoring exponentially small contributions the Euclidean action of the configuration (23. it also does not exactly stationarize the Euclidean action.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.15a) is a configuration which starts from 4> = —i^/g and ends there. we see that in the domain [—oo. vanishes as T\ —> ±oo (of course. at oo the variation is <<> = 0). T\] with LECIT = 0.15a) is not exactly a classical solution.5 Superposition of an instanton and a widely separated antiinstanton.15a) can be argued to be zero. of course. 23. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. since (23.2 m ( T o . In fact.T 0 ) } ] .16) .CHAPTER 23. Finally. #(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 . 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 .e x p { . Path Integrals and Instantons '.
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).
64) as above and consider the determinant of the expression. i..TI) K(T2.n) i=l N lim exp £*(T. T. we obtain exactly the nonvanishing terms on the right of Eq. r i ) + . (23. ^This is the method explained by B.64) Inserting the explicit expression for the kernel given in Eq.67) i=l v ) = lim exp V In ( 1 + ' L K(T.T2) I exp I K(n. H. 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.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i.64). . Felsager [91]. p. [44]. . 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.n) K(Tj. Thus D„ = l + 5 ^ K ( r i . 192. so that (with T *T) 1/2 •Dz N(T) (23.lnJV(T)}] N(T) N(T) (23. (23.63b).64) D (23. r ' ) is to be evaluated at r = T ' .7 Y. 2! .e. n )d? (23. X(TI. Martin [43]. We observe that if we discretize Eq...4 Field Fluctuations 607 This equation represents a set of n linear equations which has a unique solution provided the determinant Dn of the coefficient matrix on the right does not vanish. Cameron and W.68) D above.63a).62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .23.T')dT' (23. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. 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.n) o K(TJ. where a f 1 See R. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits. (23. are summed over all values from 1 to n.exp • /_:T JV(T')" dr' = e x p [ .j..N{T) Finally we add the following remarks concerning the discretization (23.T1)dT1 + J dn J dr2. Thus in our case one of the kernels is zero and we have to decide what we do when.a { l n i V ( r ) .n) z=l x (23. the value of the determinant is half (or arithmetic mean value) along the diagonal.Ti) i=l : exp \ £ K(T. Thus the factor a inserted above has to be taken as 1/2. (23. This case has been considered in the literature^ where it is shown that if the Volterra transformation is considered as a Fredholm transformation whose kernel vanishes on one side of the discontinuity or diagonal. whereas the determinant arises only once in the original determinant Dn.TI) 0 K(T2. = 1 52 K(n. the kernel K(r.66) •Dr) . In the limit S —* 0 .
Z i .71) Hence (with h = 1) l = j . In every integration / dzi the contribution €7.2o) 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 21.73) .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . Path Integrals and Instantons Example 23. Evaluating Ii. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). T + N(T) / T' = T JT dr' N(T') .59) the endpoint conditions »)(±T) = 0 and evaluate the integral.N(T) . In = dzi dz2 • • • dzn e x p .l + £7t) «i (23. Eq.0. (23. represents a constant translation. n J=. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo.— . (23.608 CHAPTER 23. f(T) = N{T) J^ dr'^^z(r') (23. Briefly. (23. rn+i = T.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. (23.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23.e—*0 In. Eq. I2 and using induction one obtains (cf.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). (23. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23.( . N(T') .70) and this now requires also integration over the endpoint coordinate z(T).(T) + /_ dr — . With partial integration (in the second step) we obtain (since (cf.59)) + f(T)] = — f dz(T) f°° dae^W+f™. and hence may be ignored. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . .72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.Z ( T ). (21.22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l .^ . Solution: From Eq.
N.59) and agrees with that in the literature. 11 L. (23.^ 23. £n = ~2 dTT](T)7]n(l (23. This result verifies Eq. (3. ^E.£Q) — aTo]dTod£o. Patrascioiu [110]. 1/2 T W 2 (r)J (23. J co § dr Recall that / f ^ doce^ = yfVfp. . / — / / We started off with V(T) = ^CnVniT). D.11).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.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.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. a = const. Eq." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function.e.4.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const.74) where we reinserted h.10) and (21.. Gildener and A.12). (21.23. Popov [89]. 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. i. (23. Faddeev and V. Ignoring the remaining metric factor with the reasoning explained > between Eqs. j / dTT]n(T)r]m(l r . / d£o <9r0 5[F(TQ.
We establish the function ^(To>£o) with the following reasoning.81) the rdependence is integrated out and hence the integral does not provide the function F(TO.£o)/a. demand validity of) the constraint £o = 0.^O) with dependence on ro which the replacement (23.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T).77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration.75) enforces the collective coordinate ro to be given by F(ro. (23. in the direction of the zero mode. where L = in + cn. (23. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). i /*oo (23. .^n{r).610 this means that CHAPTER 23.79) (23.81) 2 / dr»7o(r)0c(r) = . at r = ro.76) we set 4>c{r) = ^CnVnir).e. We now impose (i. i.75) requires. Thus. e.J drm(r)Mr + T0) (23. i. (23. (23. (23.e. Path Integrals and Instantons _^ ^ w . The delta function 8[F(TQ. / dTm{r)n{T). We consider the following integral / ( r 0 ) := . in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. The number £o is the component of the fluctuation rj(r).e.£Q) — aro] contained in Eq. In analogy to (23. (23. In the integral of Eq. Then cn = ~2 / dr(pc(T)rjn(T).83) **The collective coordinate in the present context is also called "instanton time". if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .78) <t>ij) = Y.g. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). this constraint implies that we go to the subspace which is orthogonal to the zero mode.80) i r or •i /*oo dr7?o(r)0(r) = 0.
./ drri0(T)<pc(T) + PJ (23..88) ro0 Then in Eq. We evaluate this quantity as follows in the leading approximation.91) " F o r more discussion of this point and related aspects see also B.^) .84) (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.86) = Jdrrjoir^Mr F(TO.J dTrj0(T)[(t>c(T + TO) + ??(T)] i r /• I a rfT77o(T)—0C(T + TO) + sr — / dTrj(T)r)o(i # C ( T ) . Felsager [91]. 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) .90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). (23.y/S^ro] (23. p. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.y/SET0 = 0.87) l/M) (23.85) PJ W i t h Eq. We have AFP = /F(T0.Zo)y/SETQ We now define A FP '• =0 • (23.<t>c{r)} . 158.75) 1 = AFP f dT06[{F{rQ. so t h a t with F(TQ. (23. (23.23.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.(23. .£O) OTQ = = £ " .
e. 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.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To).75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. the "static energy" invariant. if ^ + d£o = 0. =~ (23.612 CHAPTER 23. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). i.0AFP8 4o H P .94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. P the relative sign being of no significance. if the coefficient of 770(70) vanishes. 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. Inserting (23.91) into Eq.e. Consider. ~ There is a quick way to obtain the result (23. i.e. 770(7").93). (23. 7 Thus there is no fluctuation along the direction of the zero mode in the sum. 0 d r (23./ dT0AFP. i. tangential to (J)C(T)) leaves this and hence the square of this quantity. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i.90).e. in d(j)c(To) + dr/fa). .93) where AT = TT — 00. /(TO) ^/SE~T0} V~SET0 c d£. P J AT (23.92) We find therefore that the value of the FaddeevPopov determinant is given by the square root of twice the classical kinetic energy or simply of SE^ The relation (23. using the relation (23.
60a) in not involving the zero mode contributions.97) is a constant still to be evaluated.55b).52a) and replace in this J d£o by the expression (23. (23. Since w\ is for finite T given by WQ of Eq.4 Field 23. we can set I 2TT ^o = l'o\ wlp2'' which with Eq. 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.4. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.60a) 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. (23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. We determine I'Q by observing t h a t this differs from IQ of Eq. (23.93).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.96) lim .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 (23.23. (23. We observe t h a t the result (23.98) .55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.96) does not yet allow a comparison with (23.43)). (23.
~ rKo')e{Tx . Although we choose p2 = 1.100) in such a way that it exhibits the crucial factors of which it is composed.5 I n s t a n t o n .4>C exp h (23.i n s t a n t o n contributions As already mentioned. ^See E. (3.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. (23. the result becomes h lim 2 m r ^ ^ 1 e .102) *For comparison with the parameters of E. i.e. (23.a n t i . their Eq. In view of their normalization of the normal modes. they have to introduce the inverse of that factor on the far right of their expression (3.55b)) w0 = v0e VQ 23.TiMiy+'Mr «>. the approximation is then known as the "dilute gas approximation".4.2 m T e x p 2 irgh ' T»oo (23. Gildener and A. .20). *For the validity of the calculation the instantons or kinks have to be far apart.100) We observe that the dependence on the normalization constant p cancels out.101) 2T > oo (cf.614 CHAPTER 23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{. 0c M. we drag the factor p2 along for ease of comparison with literature. Patrascioiu [110]. Gildener and A.r). Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 . 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 .* Finally we rewrite the expression (23.32). T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. 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.
r 0 ( 2 ) = 0. the total Feynman amplitude is given by (^/.36) (/. The factor ( .(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. 1 0 6 a ) and SE{(J>I. (23. (n = 0 implies the single instanton solution). ( 2 3 .102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left). 2 . T2 = T/3.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.106b) SE{<t>l.<t>2)eSE{<t>2. Since these endpoint configurations. (3) (23.r 0 ( 1 ) = . here (f>2. this means we have to consider the amplitude (23.T2}((t)2. r0(3) = 2T/3.Tl}{4>l. .<j>i) .17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed. (23.<Pi)2 + • • • . 23.4>I) ld2SE SE(c/>i. .T2\(f)l. (23.4 Field Fluctuations 615 for n = 1.6.23.4>l)eSE{4>l.104) Considering the case n — 1 of a kink with one additional kinkantikink pair.<t>ff + h . Consider (with T 0 = T.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . are varied.l ) i + 1 in the sum of Eq.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .2 T / 3 .(l>i) (23.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.Tl\(j)i. Proceeding as before we now consider first in analogy with Eq.Tf\(p2.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig. Taking these configurations into account.H)2 + . 7 i = .Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.r / 3 .r/l&. . we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. T3 = T) d(f>l((pf.fo) = Id2 St SE(<Pf.4>i.
109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations. i. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. the dilute gas approximation. d2SE{4>) d(j)2 2m.(01 . i.107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.2J.4>i) / d(f>2 e x p {<h . the . (23.106a).e.2d2SE ~c 4>i d(f)i e x p SE^f. of course. dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .616 CHAPTER 23.12) we have JM~°°) i. each of the three intermediate kernels contained in Eq. that each applies only to a fraction of the overall Euclidean time interval of r = 2T.4>i) { 2 " 2d SE .3 9 L 3m 2 and 9SE((/)) ± . On the assumption that the kinks of the instantons are widely separated. (23. (23.e. With our remark after Eq.105) can in other respects be treated in analogy with the single instanton amplitude.108) where the last expression is obtained as follows. ±m/g (23. The allowed positions of the individual kinks have to be such that their order is maintained. rn ^2 J2H 9 1m' ±2gcj).. Path Integrals and Instantons where the first expression in Eq.e. i. in the case of l2n+i therefore ry2n.4>f) d4>\ 4>s72 (23. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.e. It has to be remembered. each of which implies a multiplicative factor 7.106b) is to be substituted into the second relation of (23.
23.105) t h a t each of the three amplitudes contributes a factor e x p [ . In the case of (<^f.93) has been applied.r/^.110) Finally we have to recall from Eq. a 2 0.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23.101): (^/. Gildener and A.2 T / 3 this product becomes that given by E.6 the collective (i) coordinate TQ has to be to the right of TQ . Patrascioiu JT/3 ° ir< >2T/3 1 dr.4 Field Fluctuations 617 replacement (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.r.Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23. i.T/0j.112) We now compare this result with expression (23. and in the case depicted in Fig. T ] with ] T 3 = 1 A » T = A T = 2T.)(3) 1 o A T 3 Ao lim .e.^ .V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.7 3 ^ — . (3) ° JrW: 2T/3 (2T)3 .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 . 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) .(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23. (23.(*) o ax = 2 T / 3 . we now have to integrate over three collective coordinates. Am3 2m J rW ( r ( 0 and this followed by setting aj. then sign reversals TQ i ) . 23.m A .a 3 [110].
W. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . (23. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state.e. has been obtained without explicit use of canonical commutation relations (i. 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. MiillerKirsten [163]. Finally we realize that the computation of (23. GP for those of E. J. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S. J. 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. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — . Achuthan. . we obtain AE = 2A* and agreement with the result (18. explicit quantization). ^See P. (23.113) is very complicated. MiillerKirsten [30]. Liang and H. H. K. by either replacing A T by 2AT or doubling A*. Wiedemann [4]. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton). l CHAPTER 23. Patrascioiu [110]. — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92.618 and hence AE A. To help the comparison we note that — with indices indicating the authors.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs.' 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. even without explicit specification of the prefactor. 23. Gildener and A.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. W.113) for qo = 1: A1E „ .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. . We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals. We observe one more aspect of our calculation.With this correspondence A\E of Eq. (18.182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2.^r . here as 1) is also taken into account.Q. W. J.113). The difference between levels.W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). AE. Bose and H. MiillerKirsten and A.
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.1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1. i. ™=£('^° < 2 «> Here the two symmetrical minima of the potential are called vacua (classical or perturbation theory vacua as they are sometimes called because ordinary RayleighSchrodinger perturbation theory would be performed around these points). 619 (24. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp. Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 .a 2 ) 2 + c^T2. and their interpretation and uses. We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues.e. in a somewhat different formulation. Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance.Chapter 24 P a t h Integrals and Bounces on a Line 24. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling).2) . however.
a spontaneous symmetry breaking leads to the double well (i.2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. its symmetry under . J 4. Path Integrals and Bounces on R1 Here dVT{4>) vanishes for . A2 _ a — \„2\ u ) ~r u/\y> 2 (24. We see that <fi = 0 minimizes VT(</>) for temperatures T > TC.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 . T = 2\(<f>2 . at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) . but implies instability for T < Tc.e. = 4(ACT2 cT 2 ). We can destroy the reflection symmetry of V((f)) (i. 24. Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool. > / V[<H 'true'vacuum 'false'vacuum Fig.a2)4> + 2cTz 2 0.1 The asymmetric double well.A a 2 ) .e.620 CHAPTER 24.
L. the vacuum being understood as the classical.g. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions. also in contexts of string theory. <44 2) ' This is now an asymmetric double well potential as in Fig. perturbationtheory vacuum). see e. T False vacua and bounces occur. Costa and J. which (b) are saddle point configurations of E (or SE).t In the following we consider first (after some recapitulations) a simple model of a false vacuum. A.1 with (24.7) *See e. Penedones [59] and references cited there. S. 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). tf(*) =^ 1 . (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE). 24. H.^J >o. Since the true vacuum minimizes the potential. On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory.e.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e.Y. Cornalba. (24.5) We see that one minimum of V now lies higher than the other.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . and (c) determine the lowering of the energy to the true quantum mechanical ground state (i. V(<j> = m/g) = 2e.g. (24. a linear term as in w^^'A1^'*0V(<l> = m/g) = 0. the quantum mechanical vacuum). The Lagrangian density of the theory is given by C = dp*d»*U(*). the higher minimum is called a "false vacuum". for instance.24. False vacua play an important role in models of the inflationary universe* and elsewhere. and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only. .) . Pi [126]. M. .g. Guth and S.
11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . i.14) .9) 9 for the doublewell potential given above. dx2 + u"(<j>) 5(x y).e.622 CHAPTER 24.10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24.12) Since <fic(x) solves Eq. t) = <f>c(x) + Y^ iPk{x)ekt (24.8).8) This equation possesses the kink solution <t>c{x) m tanhm(a. 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. h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x). we see that (24. (24. — XQ) (24. i. Path Integrals and Bounces on R1 Static fields given by d$(x.e.t) dt 0 are written (f>(x).7) yields the following equation called stability or small fluctuation equation. so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention). (24. (24. Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24. which is an important point as we shall see.13) For small w\ > 0. We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional. (24.
. eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue.oo \ x== + CX) Fig. thus the zero mode here is given by d .15) cosh mx From our study of the solutions of this equation in Sec. o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = .e. We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. kink. i. . the zero mode). i. 22.1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite. underneath the particle's velocity.17) . In the case of the double well potential (24. configuration (for our </>4theory).16) V[<H x= oo 0 d2 . 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24. i.e. 24.2 The inverted potential. d \m . 3m 2 and w\ > Am2.24.8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero.e. the other a positive quantity) and a continuum which are given respectively by w\ — 0. (24.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24.
Then the energy satisfies f°° . and — clearly — we can resort to a classical interpretation of this function.e.2. the associated zero mode. The classical configuration is then called a pseudoparticle configuration or instanton. with Euclidean time. F (24. quantum mechanics. where it is then again at rest. (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.624 CHAPTER 24. Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function.e. (24.19) E((f>c) > / J00 dxV2U(f)'c(x x0) = —x &9 1. i. i. The associated zero mode represents the velocity of the pseudoparticle as indicated in Fig. 4 TTT. We define the current k» := e^dvfc. i. Integrating the classical equation (24. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. 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.e.21) The charge is defined as the space integral of the time component of the current. The kink solution can now be looked at as the classical solution of a onedimensional field theory. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function. In this case the usual Newton equation reappears with an opposite sign of the potential. 24.e. Below we shall see that the configuration described as a bounce is given by an even function of x.20) The number 1 on the right is the socalled topological charge which is obtained as follows.18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U. so that its derivative representing the pseudoparticle's velocity is odd.8). by Q = \(°° dxf^eoid1^) = l[M°°) ~ M°°)] = 1(2422) . In the pseudoparticle consideration the coordinate x plays the role of time. which has the solution 4>c(x). 4>"{x) = U'(4>). Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton.3 (24. i.
M. Tchrakian [208].A W + £/(A0C (24.A2)2 ^ 0 except for A2 = 1. Bender and T. Liang and H. See also S.5 = 15. See in particular Sees.* In the following we discuss a simple model which allows an explicit discussion of bounces.1 to 6.4 and 6.• ±00 : U(X(pc) = ^ ( 1 . Jianzu Zhang and D.24. of course. possess other solutions which have Q — 0. T. Coleman [41]. + H. The requirement of finite E therefore does not permit \<j)c deformations.Q. S.24) Thus the energy is finite only for A = ± 1 . We consider the family of configurations {\<fic(x)} where A is some number. In the preceding chapter we performed this calculation with the help of the path integral method. Wu [18]. MullerKirsten [166]. J. We now turn to corresponding considerations for a potential which allows tunneling to infinity and hence the definition of a particle current. (24. Callan and S. the kink solution having a monotonic behaviour on the way from one endpoint to the other.2 The Bounce in a Simple Example 625 For reasons discussed earlier this charge is called the topological charge.37)) 4 x . 2. W. Coleman [56]. Then E[\4>c f dx . In Eq.5.11) of this reference the number 35 of 8/35 is misprinting of 3. (22. [19]. Our main objective here is in the first place the recalculation of the result of Bender and Wu* for the decay rate of a particle trapped in the ground state of an inverted double well potential by means of the path integral method. We can introduce the concept of topological stability as follows. . MullerKirsten.23) J —( In this expression we have the potential with the behaviour (see Eq. The theory does. ^ *C. [57] and C. We call this topological stability. H. We recall finally that the onedimensional theory with the double well potential allows the calculation of the difference between the ground state energy and that of the first excited state. These are the constant solutions < = ±m/g which are therefore > / called nontopological or are described as topologically trivial. f J. J. Coleman [54]. It is important here to realize that the nonzero value of Q results from the different boundary conditions of the classical configuration at +00 and —00. (4. We see already here that a configuration which has the same value at both endpoints has Q — 0 and so is nontopological (this will be seen to be the case for a bounce).^ 24. We are here interested in these nontrivial solutions with Q / 0. W.2 The Bounce in a Simple Example The concept of bounces was introduced by Coleman. G.
z . the formula there obtained (i. Ohmori [222]. .25) In field theory the </>3 potential does not make much sense since it is unbounded from below. 24.g.6). the result (24.* However. We do not carry through the entire calculation to the explicit expression for the decay rate. (24. (24.e. Fig.. We therefore proceed immediately to the Euclidean EulerLagrange equation for particle mass TUQ = 1 1 fdq 2\dx = V(q). which we do here for purposes of illustration. 24.3 The inverted cubic potential and velocity of the bounce. See e.5 (for a WKB treatment see Example 14. Path Integrals and Bounces on IR1 We take as the definition of the lifetime of a state the inverse of the decay width T defined by 3E r := .2 .86)) is then applied to the cubic potential in Sec. since this will be done in detail in the next section for the inverted double well potential.626 CHAPTER 24. K.26) *A cubic potential like the one we treat here found application in a string theory context. we can discuss it in the context of quantum mechanics.
24.e. cosh [ab(x . Thus this derivative. Since the zero mode is now the odd eigenfunction of a Schrodingerlike equation with eigenvalue zero. 24. we evaluate the Eucl