•
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
one problem being that there are few nontrivial models which permit a deeper insight into their connection. that is the Coulomb potential and the harmonic oscillator. with degenerate vacua) were first and more easily derived from the Schrodinger equation. but arise also in recently studied models of large spins. we recapitulate the origin of quantum mechanics. basic postulates and standard applications. its mathematical foundations. in Feynman diagrams in quantum electrodynamics). 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. Chapters 1 to 14. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. like periodic potentials. screened Coulomb potentials and maybe singular potentials. 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. be treated to a considerable degree of satisfaction perturbatively. even though the expansions are mostly asymptotic. With various techniques and deeper studies. has not always been understood well. whereas for the calculation of discrete eigenvalues the Schrodinger equation. the aforementioned exactly solvable cases.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. With the growing importance of models in statistical mechanics and in field theory. To what extent the two methods are actually equivalent.e. numerous problems could. However. Excluding spin. in fact.g. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. In the first part. These basic cases will be dealt with in detail by both methods in this text. again point the way: For scattering problems the path integral seems particularly convenient. 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. 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. but also about complex energies that he encounters in a parallel course on nuclear physics. Thus this first part .
In the second part. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. i. periodic instantons. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. The potentials with degenerate minima will be seen to reappear throughout the text. we deal mostly with applications depending on perturbation theory. 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). Chapters 15 to 20. and the elliptic or Lame potential — here introduced earlier as a generaliza . which is presumably the only such singular case permitting complete solution and was achieved only recently. The most prominent examples here are the double well potential and its inverted form. After a treatment of power potentials. and their eigenvalues.XVII deals mainly with standard quantum mechanics although we do not dwell here on a large number of other aspects which are treated in detail in the longestablished and wellknown textbooks. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. 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.e. The solution of this case — however in nonperiodic form — turns out to be a prerequisite for the complete solution of the Schrodinger equation for the singular potential 1/r 4 in Chapter 19. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. In the final chapter of this part we discuss the large order behaviour of the perturbation expansion with particular reference to the cosine and double well potentials. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. Using perturbation theory. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. Thereafter the concepts of instantons. the chapter thereafter deals with Yukawa potentials.
In fact.g. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. quartic. however. the Mathieu equation. 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. anharmonic oscillator potentials on a comparable level with.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. cubic). the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. 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. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. for instance. from our precise reference to unavoidable elliptic integrals taken from Tables). The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. except that these are no longer of hypergeometric type. With a fully systematic perturbation method and with applied boundary conditions. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. an approximation whose higher . such as limiting procedures. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. We then illustrate the relevance of this in the method of collective coordinates. 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. This puts Schrodinger equations with e. 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. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. 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. All results are compared with those obtained by perturbation theory. and whenever available also with the results of WKB calculations. it is. This method is therefore more complicated. for instance. we can make the following observations.
For ease of reading.* H. Andrews). Whittaker and Watson [283]. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. Dingle (then University of Western Australia. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. also for the verification of results.g. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. D. The line of thinking underlying this text grew out of the author's association with Professor R. in particular Professors J. K. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. As a rule. not the least for permitting a more detailed and hopefully comprehensible presentation here. are relegated to separate subsections which — lacking a better name — we refer to as Examples. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. the references referred to are never cited by mere numbers which have to be identified e. The author is deeply indebted to his onetime supervisor Professor R. T. W.Q. 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. Park (Masan). . like E. B. J. we also consider at various points of the text comparisons with WKB approximations. Their deep involvement in the attempt described here is evident from the cited bibliography. as well as applications and illustrations. B.g. with the source given in the bibliography at the end. e.g. 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. Whittaker and G. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. N. H. Thus when instantons became a familiar topic it was natural to venture into this with the intent to compare the results with those of perturbation theory. Instead a glance at a nearby footnote provides the reader immediately the names of authors.XIX order contributions are difficult to obtain. thereafter University of St. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). MiillerKirsten *In the running text references are cited like e. Liang (Taiyuan). D. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). shows him that each such formula here has been properly looked up). at the end of a chapter (after troublesome turning of pages). Throughout the text some calculations which require special attention. Watson [283]. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. Since this comparison was the guideline in writing the text. Nonetheless.
The best approximation to such a body is a cavity with a tiny opening (of solid angle d£l) and whose inside walls provide a diffuse distribution of the radiation entering through the hole with the intensity of the incoming ray decreasing rapidly after a few reflections from the walls.Chapter 1 Introduction 1. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it. the amount of radiation absorbed by a body is equal to the amount the body 1 . 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 .2 cm at moderate temperatures T) is the radiation emitted by a body (consisting of a large number of atoms) as a result of the temperature (as we know today as a result of transitions between a large number of very closely lying energy levels). which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy).1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. Kirchhoff's law in thermodynamics says that in the case of equilibrium.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. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. We do not enter here into detailed considerations of Planck. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Instead. 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.
(1. v + dv in equilibrium with the black body at temperature T. 1. T)du is the mean energy density (i.e.2 CHAPTER 1. In Eq.. (1.1) c is the velocity of light with c = u\.e.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.T) = 2*?£(?)kT. These expressions are: (1) Wien's law. the formula (to be explained) u(u. 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.3 4 J s.1 Absorption in a cavity. y J c 3 \ex . It was found that one expression agreed well with observations in the region of small A (or AT).1) ' y Here u(v. Fig.e. energy per unit volume) of the radiation (i.T) = dv3eC2U/T. The parameters k and h are the constants of Boltzmann and Planck: k = 1.e. Let us look at the final result of Planck. radiators. Black bodies as good absorbers are therefore also good emitters. u(u. i. h = 6. and then measuring the increase in temperature of the heat bath. A being the wavelength of the radiation. and the other in the region of large A (or AT). Introduction emits. T) were known and tested experimentally.38 x 1(T 23 J K'1. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. kT kXT (1. i.626 x 10 .2) and the . How did Planck arrive at the expression (1.l ) where x = ^ = ^ .
1. and he had succeeded in finding such an expression of the form u(v. (1. (1. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. Here W is a number which determines the distribution of the energy among a discrete number of objects.1. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". where the Fig. When Planck had found this expression. To this end he considered Boltzmann's formula S — klnW for the entropy S. we obtain: u(i/. We see. This is the point. 3 (1. that the formulas (1.1) as approximations. "small" (i.e. (x large). Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. and thus over a discrete number of admissible states.T) = av e6"/Ti' where a and b are constants. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously.47TZ/ 2 {x small). which are indistinguishable) among the N indistinguishable oscillators at .3) are contained in Eq.T) 2^^kT. e xhv. T) u(i/. Indeed.3) Ci. but — here the discreteness appears properly — only in elements (quanta) e.2) and (1. exp(x) ~ 1+x) and "large" (exp(—x) < 1). Considering Eq.T) = 2^C3T.2 Distributing quanta (dots) among oscillators (boxes). in the first place Planck had tried to find an expression linking both.1) in regions of a. discretization begins to enter. he searched for a derivation. C2. . C3 being constants.
940.e. as in most other Tables. I. W.2.343(2). e. Oberhettinger [181]. with riydu — 2 x —w—dv. one obtains (cf. (1. u(i>.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. v + dv.3. 1.e.4 CHAPTER 1.4) IniV! ~ JVlniViV + O(0). e = his. i.6) Fig. . Agreement with Eq. Magnus and F. and the second law of thermodynamics ((dS/dU)v Example 1.g. M.3 Comparing the polarization modes with those of a 2dimensional oscillator. (1.T)dv. Gradshteyn and I. p. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. p. formula 8. Then W is given by w = (N With the help of Stirling's formula* {N + piy.7) *See e. as for small values of e). . We now obtain the energy density of the radiation. there not called Stirling's formula. 1. We visualize the iV oscillators as boxes separated by N — 1 walls. S. (1.2) requires that e ex is. = 1/T). U{T) being the average energy emitted by one oscillator.g. Ryzhik [122]. Introduction temperature T. The Stirling formula or approximation will appear frequently in later chapters. i. N * oo.1)!P! (1. with the quanta represented schematically by dots as indicated in Fig. h = const.
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. i. We observe that u(v. We obtain therefore u^T) = Unv = 2^fJ^—i..3 (as for ideal conductors). The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^. is given by .3.2.. . L for i = 1.965 and xmin = 0. .i = 1. KT = I J .2.1)) The solutions of this equation are ^max = 4.1.7) for instance.e. dj\l dM .. so that (lvL\A 0 I I = rr.. (1.8) This is Planck's formula (1. 2 2 2 r2 L K — 7T n .V 2 ) E = 0.UJ = 2KV.. (1. v + dv. Then L^j = nrii. as in electrodynamics.T)dX = ^ehc/*kT_idX.1. nvdv per unit volume. so that the derivative of u implies (x as in Eq. In terms of A we have u(X.1).3. '''From the equation I \ JW . We obtain the expression (1. as indicated in Fig. where we have for the electric field E oc elwt \ J eK sin KI^I sin K2X2 sin K3X3 K with the boundary condition that at the walls E = 0 at Xi = 0. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.UJ2/C2 + K? = 0. where^ 2 [2KUY .3. (1.2. so that . rii = 1. The number with frequency v in the interval v.7).T) has a maximum which follows from du/dX = 0 (with c = vX)..
8) could be derived much more easily in the context of statistical mechanics. that it is easier to consider first the case of T = 0. of course. 2 . One might suppose now. such a procedure leads to contradictions. the behaviour of the system at zero absolute temperature.—v <L9) We observe that for T — 0 (i. Introduction he 4. . i. n = 0. x — oo) the mean energy vanishes (0 < U < * > oo). . 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 . If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. We are not dealing with the linear harmonic oscillator familiar from mechanics here. One expects. (1.e. but one can expect an analogy. .10) Thus here the socalled zero point energy appears.2 (1. and from which the constant h can be determined from the known value of k. Lorentz and Planck that Eq. This is Wien's displacement law. 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=—.10). Later it was realized by H. However.965K = Const.1. which can assume the discrete energies en — nhv. which had also been known before Planck's discovery.l.e. ra = 0. which can not be eliminated without a different approach. Thus we have a rather complicated system here. which did not arise in Planck's consideration of 1900.6 The first value yields AmaxT = CHAPTER 1. that of an oscillation system at absolute temperature T ^ 0. we then have at T = 0 independent oscillators. . . that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. A. Since temperature originates through contact with other oscillators. We therefore examine such contradictions next.
so that 1 f ds\ T — \dUjy .In . with Heisenberg's discovery of the uncertainty relation. around 1926.1: 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. . {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) .k dU 1+U\ f ln(l T + U\  U U k ( e .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. kT/2. we obtain T V S =fc[ln(TV+ P .e. We "This is what was effectively done before 1925 in Bohr's and Sommerfeld's atomic models and is today referred to as "old quantum theory".2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.i n .1)! . 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.1. In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente"). for 2 degrees of freedom.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. i.2 Contradicting Discretization: Uncertainties 7 Example 1. 1.1)! .1 ' u~ e .. d .= .kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .ln(TV .
8
CHAPTER 1. Introduction
shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = mx2 +  w V .
ZJ
(1.11)
Zi
The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A
A A
^ =\[Ef+l
(i i2)

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

(1.17a)
In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at
1.2 Contradicting Discretization:
Uncertainties
11
any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 Any'=
(1.17a) 2TT/L
since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)plane, we can write more precisely ApAq > 2irh. (1.18)
This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.
Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by
12 the limits 0 < q < q + Aq, 0<p<p
CHAPTER 1. Introduction
+ Ap,
what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is reinforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.
1.3 ParticleWave Dualism
13
the number of such photoelectrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particlelike, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few pointlike spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particlelike nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum
The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f
L. de Broglie [39].
14
CHAPTER
1.
Introduction
1.4
ParticleWave Dualism and Uncertainties
We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a doubleslit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.
^y
Ax
Fig. 1.7 Schematic arrangement of the singleslit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].
1.4 ParticleWave Dualism and Uncertainties
15
figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.
Si
t
Ax
\
»v
"^^J
B ^ ^
*
w
p3
Q
Fig. 1.8 The wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)
Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — rexp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T
5 = 2a =
lim NSN = — A a ; s i n 0 .
(1.21)
If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.
16
The intensity at the point Q is therefore
CHAPTER
1. Introduction
h = h
where from Eq. (1.21)
•K ,
.
a = flisinB = Aisint A 2
.
k
Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is
Ie=Io
sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2
(1.23)
The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 )  , i.e. for A x sin0 = (2n + 1 )  = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.
A
(1.24a)
(1.24b)
The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the singleslit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave Dualism and Uncertainties
17
Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or
1.4.1
Further thought experiments
Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light
Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it
18
CHAPTER 1. Introduction
is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~±. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have L = rp' = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.
Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.
1.5 The Complementarity
Primciple
19
quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by
AE « T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax
vG
.
It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of Ai£.
1.5
Bohr's Complementarity Principle
Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The socalled Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a doubleslit (this being the apparatus in this case) the photons or
20
CHAPTER 1. Introduction
electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wavelike or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*
1.6
Further Examples
Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with timedependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —yEo, w o (1.28)
where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui
u> = u>o a,t t = 0, w = ui at t = T,
.
_
i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w mx • 2\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt
dt The energy of the oscillator is l „ E — mx1 2
l 0 , z l 29 + mu} (t)x , y 2 '
so that
dE 1 — = mxz dt 2
9
dJ1 > 0, dt ~
(1.29) v '
where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain  [mx + mur (t)x\ = 0, i.e.
1 1 1 — —mx 2 H—mx 2 dt u22 2
1 mx 2
2d
— . 2 dt\u)
"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].
1.6 Farther
Examples
21
d ( E\ 1 , d / 1 \ — — ) = mx2 — ( — = dt\uJ2J 2 dt\u>2)
mx2 dw < 0, UJ3 dt ~
(1.30) v
where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, ie. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E
E dw2 < 0, UJ4 dt ~
i.e.
1 dE 1 dw2 < —2 . E dt ~ u) dt
(1.31)
JEo
or
E  Jui
u, '
> °
"o
E0 ~ UJ22
' ^ u'2 — < —
E'
<
^EQ.
Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with
e + wgeo, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2  =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < 5Bo.
E0<E'
Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and
22
CHAPTER
1.
Introduction
the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average
(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence
(L32)
E i.e. in
If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then lmw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (mx2\ =
If we now insert in the equation for dE/dt,
/I 1£ 2\ (  mx ) = , \ 2 / 2u2' we obtain dE_/l
~dt ~ \2mX
and hence
2\dw
2
_Edw2
dE _ 1 dw2 _ du
/ ~dT ~ 2w2~dT'
E — = const. w
°r
~E ~ ~iU> ~~ ~u7'
In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by
~~ p
2R\2nh)
~ AKR'
See e.g. L. Schiff [243], pp. 25  27.
Chapter 2
Hamiltonian Mechanics
2.1 Introductory Remarks
In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and reexpress Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the cnumber valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.
2.2
The Hamilton Formalism
In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
P.^ ^ j . The total time derivative of u can therefore be rewritten with the help of Eqs. M. 6t>0 5t Real quantities which are directly observable are called observables. Pi has to be considered as an independent quantity. the equation of motion of the observable u. (2. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq.2) as This equation is.3).) = £ [wm + WiK) = £ [WiWi . (2.4) as the generalization of Eqs. .1). S »(«. the standard reference for the application of Poisson brackets is the book of P. which can be observed directly at time t. du \ x^/dudH du dH\ . Hamiltonian Mechanics is wellknown.1). Goldstein [114]. (2. We can therefore consider Eq. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. one obtains the Hamilton equations* • OH .n n. this expression is simply a functional determinant. (2. All functions u(qi. A system consisting of several mass points is therefore described by a number of such variables.e.Pi) is contained implicitly in the canonical variables q^ and pi. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. i. ^As H. One can verify readily that Eq. chapter VII. (2.p% are therefore again observables. chapter VIII. in analogy with Eqs. Goldstein [114] remarks at the end of his chapter VIII. since .t). the entire timedependence of observables u(qi.1) as d . whereas the velocity requires observations of space coordinates at different times. (2. Compared with an arbitrary function f(qi. . It suggests itself therefore to consider more closely the properties of the symbols (2.qi(t) qi = hm f . Rather.4) contains as special cases the Hamilton Eqs. which all together describe the state of the system. x^fdu. A..24 CHAPTER 2.2) If we have only one degree of freedom (i = 1). qi(t + 6t) . (2.Pi.g. as mass times velocity. Dirac [75].1). (2.pi) of qi. The following properties can be verified: *See e. H.
B2}. {A.5d) is useful in calculations. Bx} + a2{A. Later we shall consider noncommuting quantities. {B. {C.6) give the values of these. As long as we are concerned with commuting quantities. As an example we consider a case we shall encounter again and again.e. (2.2. it is irrelevant whether we write {A. B}} = 0. the Poisson bracket {A. we wish to evaluate {A.5d) above. (2. A}} + {C.Qk} = 0. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. {pi.2 The Hamilton Formalism (1) Antisymmetry: {A. If.B}C or C{A. like here. (2. a i S i + a2B2} = ax{A.3) of the Poisson bracket. that of the linear harmonic oscillator.Pk} = 0.B}* (4) product formation: {A. Since Eqs. 25 (2.5c) The first three properties are readily seen to hold. in other words without any reference to the original definition (2. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. but could be multiplied by a complex number): {A. (2) linearity: {A.4).BC} (5) Jacobi identity: {A. which combines the Hamilton equations. where A and B are arbitrary observables.5b) (3) complex conjugation (note: observables are real.C}. (2. that the very general Eq.B}.B} = {B. we obtain the fundamental Poisson brackets. {qi.5d) = {A*. then the ordering is taken as in (2. These are {Qi. B} is completely evaluated.Pi. C}} + {B.B*}.6) We can now show. B}. Property (2.A}. The original definition of the Poisson bracket will not . If we evaluate the Poisson brackets for qi.B}C + B{A.5e) = {A.6). i. for example.5a) (2.Pk} = 5ik. (2. (2.
6). (2.11b) (2.t). (2.11a) 'q' = q. These are transformations qi—>Qi = Qi(q. or q(t) = qo+ Pot .26 CHAPTER 2.12) .H}. q = {q.10) In classical mechanics one studies also canonical transformations. "4' = q.p.8a) and use the properties of the Poisson bracket and Eqs.qot2 . (2.t).P2} + {q.9) Similarly we obtain from Eq.p.8b) p=q.8a) (2. (2. According to Eq. from which we infer that q(t) — qocost + posint.q2}) = = 2i{q.7) q = [q.yPot3 + •••• (2.H}.. and P={p. Then we obtain: (2.\(p2 + q2)} = l({q. Pi—> Pi = Pi(q..p) is an ordinary function is also irrelevant.p\) P. we consider as Hamiltonian the function H(q.p}p + p{q. Hamiltonian Mechanics be used at all. and so q = q...4) we have for u = q.9) and (2. (2.8b) We insert (2. (2.10) we deduce q = p = q.p) = ±(p2 + q2).7) into (2. In the evaluation one should also note that the fact that g.p. Since constants are also irrelevant in this context. (2. (2. and Pi are ordinary real number variables and that H(q. From Eqs. q + q = o.
i.* Hence in Example 2. Goldstein [114]. . i. those at a time t — 0. i.P is canonical in the sense defined above. Pi^Pi=Pi(Q.P provided the transformation q.1 we verify only Eq. i. . Eq.2 The Hamilton Formalism 27 for which the new coordinates are also canonical.pk} = 5ik.e. P) exists. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.e.P.e.Pi back to their constant initial values. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.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. (2. verify that the Poisson bracket of two observables A and B is invariant.13) We write the reversal of the transformation (2. chapter VIII. {PhPk} = 0.B}q. (2. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.3) we can now express the Poisson bracket {^4. {QhPk} = Sik. _ax p__dK_ (2. (2.B}q.Pk} = 0.e. and Example 2. as {A. Example 2.15a). i.t).P One can then show that {A.t).1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P). B} of two observables A and B in terms of either set of canonical variables. *H.qk} = 0. for which Hamilton's equations hold.p <> Q.14) With the help of the definition (2.2.e.B}Q>P.15a) °raS {AiB}Q. {qi. which transforms qi.P.p = {A. {qi. that (dropping the subscripts therefore) {Pi. which means that a Hamilton function K(Q. Example 2. (2.Qk} = 0. Of course. {Qi.2 below contains a further illustration of the use of Poisson brackets. (2.15a).3 deals with the relativistic extension.12) Qi—>qi = qi{Q.
1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. we obtain analogously dA dQk' Inserting both of these results into the first equation. p. Solution: Relativistically we have to treat space and time on an equal footing.+ dQ.s F 5 7 r ={A. OQk Pk}q.p(0) = poSolution: The solution can be looked up in the literature.15b) dA {Qk.28 CHAPTER 2.P^pr.p^v dP. 236. we obtain dA {Qk. § E x a m p l e 2.p = J2 d~Fk Replacing in the above A by P and B by A.( 9A = {Pk.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. Mittelstaedt [197].pc).pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. {A. Qk}q.p^— + {A.p = dB dA dB \ rA 1 {A'B^ E b r s r . Thus we extend q and p to spacetime vectors (t.B}QIP.A}q. (2.E(t). .Pj}q.A}q.q/c) and (E. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u. Hamiltonian Mechanics S o l u t i o n : U s i n g t h e definition of t h e P o i s s o n b r a c k e t a p p l i e d t o ^4 a n d B w e h a v e r „ „•. Thus whenever q and p are multiplied. (2. § See P. we have —Et. dA {Qk.15a) r„ ™ v . V V °Qk dPk dPk dQk J E x a m p l e 2. their product Et — qp being relativistically invariant.3: Relativistic Poisson brackets By extending qi.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).Qj}q. we obtain as claimed by Eq.p^ "Pk.p) .B}q.F}r ~di d~E ~ d~E~8t Consider F = H(q. Replacing here A by Q and B by A.
numerically zero.3 2.E(t)}r = Hence at Relativistically we really should have clu/dr. i. This probability is evidently proportional to AqAp. We distinguish in the following between two kinds of probabilities.. dudH s l u . We consider first the a priori weighting or a priori probability. (*. H(q. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . to .16) For example. so that Go(qo.3. we have A = (p apaq — J .3 Liouville Equation. it is the equations of motion which lead from Go(q. Since Hamilton's equations give a continuous map of one domain onto another.14) and area A of the phase space ellipse given by Eq.Po. one obtains Gi with q = q((lo.to..t0.p) to G\{q. .po is a point on Go. i.' ±=j1V c \dtj' u^. c dt 2 2 w^ du du dr .e. and the space spanned by the entire set of canonical coordinates is described as its phase space. (1.po.t). Instead we assume a case in which we know only that the system is located in some particular domain of phase space. du q\ P = —.e.p) . (2.)•'=(*)**£.p).p). Then .1 Liouville Equation. of course.2.y. But now we consider a system whose phase space coordinates are not known precisely. Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi. Let us assume that at some initial time to the system may be found in a domain Go(q. Of course.17a). 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.dt y/l  qZ/c2' 2.t). dE .. . but partial derivatives of F do not vanish.Pi are the coordinates of some mass point m. I . g oc AqAp.Po.po) = Gi(q. Probabilities 29 (This is. since H is expressed as a function of q and p.p) around some point qo. g. rrr ._.. in the case of the linear oscillator with energy E given by Eq. if qo. 2TTE UJ and hence g oc — . p= p(qo. also boundary points of one domain are mapped into boundary domains of the other. and E as a function of t). (1. du . and at time t > to in a domain G\(q.p)..
— dp A A — In(AqAp)s = — + — = Example 2.p^. Thus g must be independent of t.1): d . as is demonstrated by Liouville's theorem in Example 2. Example 2. Show that the total volume of phase space covered for constant E is 8n2IE. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. the (pg. (1. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. 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.4: Liouville's theorem Show that A q A p is independent of time i. dt dq and with with Hamilton's equations (2.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. = 2TrIEsm6.4. Pi 2 / \. which means. this has the same value at a time to.15) — an ellipse of area § dpgdp.30 CHAPTER 2..)curve for constant E and $ is — as may be seen by comparison with Eqs.eded<t> = 8TT2IE. Example 2. dt dq dq dp dp d2H dqdp d2H = 0./. dpdq and similarly dt = — Ap. Hence from the difference: — . 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.14) t o (1. Je=0 .— = — Aq. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . and hence g oc %ir2IdE.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . J0=0 Hence g oc 8n2IdE. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.
e.e. 2.p.2.p. whose positions in phase space are characterized by points. W=p(q.t).17) F ^GT^^ 0 Fig.3. i. leaving the basis of classical mechanics! Hence we set / «w)^ =.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.t)dqdp= 1. it is suggestive to introduce a factor 2KK with every pair dpdq without.Thus we assume a large number of identical sytems. by jr = P(9»P. Thus dn is that number of systems which at time t are contained in the domain q. (2. (2.Po. Since W has the dimension of a reciprocal action. (2.20) . i.p.p + dp. however.q + dq. d d 1 P = JJdqidpi.3 Liouville Equation. 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. Probabilities 2.*). (2. The total number of systems N is obtained by integrating over the whole of phase space.p.19) Thus W is the probability to find the system at time t at q.18) With a suitable normalization we can write this / W(q. dn — p(q.1 The system moving from domain Go to domain G\.t)dqdp = N.
t)dp( — I . we consider the domain G in Fig. t + dt)dqdp. i. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.t)dp{ jt Q. p + dp G P O q qndq Fig. 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.e. * q Fig.p.p / . The equation of motion for n or W is the socalled Liouville equation.a t dt + q+dq. i.p. how the system moves about in phase space. We are now interested in how n ov W changes in time.32 CHAPTER 2. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space.P dt p(q + dq. In doing this. In order to derive this equation.e. The number of points at time t + dt in domain G.p) denote the velocities in directions q and p — p{q.p. i.p) and vp(q. that in our consideration no additional points are created or destroyed. 2. — if vq(q.3 The region G. 2. 2.e.p. t)dqdp p(q. p(q.2 The ensemble in phase space. we take into account.p.
21) with Eq.p.t)vp(q.4) shows that p and u satisfy very similar equations. .19). so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . Comparison of Eq.20) and (2.p(q. . put differently.t)vp(q.t) dt p(q.p + dp.p.p. the probable motion of the system under consideration.P *' % = iH>P} ( 2 .t)} with JW(q. vp(q.p.t)vp(q.p) .p + dp)dtdq.p)=p = 3H —. p. Probabilities and thus p(q.p(q.p)dtdp +p(q.p.P)]K(Q. (2.t + dt) .p) .p. (2. dH Mq. p.p + dp) dp = or  However.t)vq{q.3 Liouville Equation.p(q + dq.p).t)vq(q + dq.p(q.p(q + dq.p + dp.t)vp(q. t)vq(q + dq. t)dqdp = p(q.p) = q = g^.p) dq p(q.p)dtdp . .P. (2. With Eqs. 33 Dividing both sides by dqdpdt this becomes p(q. (2.W(q.p)dtdq .p.2. .P)].t) dt {H(q.p.t)vq(q. = K(9.t + dt)dqdp — p(q.21) we can also write dW(q.p.21 ) This is the Liouville equation which describes the motion of the ensemble or.p.t)^ = 1.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . (2.
26) we described the time variation of the observable u(q.p.e. for i<«>!/«<*p>"w>(^)". Equation (2. We deduce from the Liouville equation the important consequence that ^ M = 0. Example 2.1 the area Go is equal to the area G\. since no systems are created or destroyed.34 CHAPTER 2.p)W(q.4 Expectation Values of Observables Let u = u(q. (2.p). We now inquire about the time variation of the expectation value (it) of u. (2.p) be an observable.t)(^J. p+dp.e. (2.p) the following expression: (u)=Ju(q. and hence that equal phase space volumes contain the same number of systems. 2. i. Thus in Fig. The first and most immediate possibility is — as indicated .p.24) since the total derivative is made up of precisely the partial derivatives contained in Eq. p. With Eq. q+dq.that the density or probability W(q. (2.4).24) implies that p is a constant in time. (. We define as expectation value of u(q.4). i.24). We shall see that we have two possibilities for this. that dtj\y v if qo. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.po are the initial values of q. t) depends explicitly on time t (if determined at a fixed .p (cf. 2.
Then Eq.p.t).t)}(^y. (2.t) = = W(g(q0. that (u) = (u)0. and hence u(q.p0.f(q0.4 Expectation Values. Reversing Eq.p) = uo{qo. of course.t).p.Po) at time t = 0.po.t).34) S e e also H. .22).t).2.p. Thus we can write.0) = W0(q0.0) =u(g(qo. since W oc p is constant in time according to Eq. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q. (2.t). (2. 8. Observables 35 point in phase space).t).f(qo. (2. 1 (2.t)W0(q0.Po. where we used Eq.32) In this expression the time t is contained explicitly in the observable u(q. Sec.po.p. we have Qo = g(q.0) = u0{qo.p.^ Solving the equations of motion for q. (2.24): W(q.p0)(^^J. However. at t = 0.33) po = f(q. (2.31) i. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. We expect.po.p) assumes certain values) that depends explicitly on time.e.p){H(q. we can express these in terms of their initial values qo.Po. we can also employ a more complicated consideration. so that Q = g(qo. ie.28) = Ju(q.t) W(q0.t).Po.p.29).po. (2. We verify this claim as follows.p).p.29) The distribution of the canonical variables is given by W(q.30) p = f(qo.t).t). (2.p) = u(q.Po. Goldstein [114].po. (2.8.Po.i).W(q.27) becomes !<»>  /«(*P>!"W>(^)" (2.t).
t).t).\„.p.35) Inserting these expressions into no we obtain uo(qo.PO.t) .t).p)})q=g(QO. (2. (2.29) q = g(g(q.u(q. we obtain («)o = / u{q.p0) = W(q.po.o. Eq.P0. (2.p.f(q. Hamiltonian Mechanics so that on the other hand with Eq. (240) .P0.po.Po)(^^)n.p.39) Substituting this into Eq. (2. (2.M 27Th i . Eq.f(q.30) u0(g(q. Inserting Eqs.P)}).t).t).t). (2. Taking now the total time derivative of (it)o. [ du0{q0.) f / u (q0.t).u{q.Po.t) = (2 30) (2.t) dB__ (du\ °P \°PS P=f(qo.38) we obtain £<«)„ = .35). (2.t)J(q. (2.P.p). / ^fdq0dpc .30) Moreover.t) 0H_ 0 q =  {{H(q.p.t)\—^\ = (u). (2.f(q.) Wo(qo.25) into Eq.31)) W0(q0.32).t) d t _ (du\ \ d<i dq / P=f(qo.t) M f 9u\ V °P / P=f(q0.t) (2.37) as had to be shown.28).0)=u(q.Po.p.t) dp M du\ ° q 'P=f(qo.p)W{q.t). we obtain an expression which is different from that in Eq.e.t) v{g(<j(q." V " " ™0' > • • » « » 0 .po)(~ .p.t).=9(.p).po. (2.36) = (2.t)W "(q0.P. (cf.p.32).p).p.p.(dq0dp0\n We deal with the partial derivative with the help of Eq. p = f(<j(q. d dt {U)0 = dt.f({H(q.36) and (2.t).30): duo(qQ.POit)i p=f(qo.t).t).0) u(q.t).f(q.f(g(q.Po.p.Po.p.(2. .t).37) and (2. (2. i.36 CHAPTER 2.
34) and use (2.W(q. x r „.42) the relation d .p). (2.u(q.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). In the first case. / d q d p \ n 2irh J (2. so that ~d~t <«>0 = J{H(q. (2.p){H(q.4 Expectation Values.p). from the properties of the Poisson bracket.p).45) in agreement with Eq. described as "Schrodinger picture".28).45) the Liouville equation. W}. Alternatively we could deduce from Eqs. However. uW} = {H. ..t) of u assuming certain values q and p.W(q.t) dqdp\n 2irhJ (2. The phasespace integral of a Poisson bracket like I (2. (2. u}W instead of u{H.41) This expression contains {H.2. u}W + u{H. ^ „.43) ——uW = 0.p.t)}. and the time variation of (u(q.W} in Eq. (2.p.p)) is attributed to the probability W(q.41) and (2.p). lim ——uW — 0.28) and (2.p.p.uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. we obtain zero after partial integration of I and hence from Eqs. Observables 37 Here we perform the transformation (2.p)}W(q. we obtain {H.31). dt <«)o v f u(q.25) and (2.28).t) dt {H(q. Consider {H. .(2.t)} . This timedependence is described by the Liouville equation dW(q. The considerations we just performed demonstrate that we have two ways of treating the timedependence: The explicit timedependence can either be contained in the probability W or in the (transformed) observables..p. the observable u is treated as a function u(q..
po. (c) A multiplication of the quantities among themselves is defined. which. Eq.p)." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. The equation of motion is then that of an observable. but does satisfy the usual associative and distributive laws.rt.e. the probability of the initial values Wo(qo. 2. does not have to be commutative.F(«. i. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. for which the axioms of a commutative group apply.p)}.t). the reason being that — since qo. Hamiltonian Mechanics In the other case. called u Heisenberg picture". " T h e author learned this approach from lectures of H. .** 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) dt du(q. The time dependence of the expectation values can be dealt with in two different ways.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. (b) As usual a muliplication by a complex number is defined. (cf.4)) ^ = {«(. (2.po) is assumed.t) dt __ ~ (2J30) ( 4) du0(qo.46) We thus also recognize the connection between the Liouville equation.p0.po.H(q. as the equation of motion of an ensemble or of a probability distribution on the one hand.. and probabilities W. " This means: Only in the Heisenberg picture dW/dt = 0. which describe the state of a system.38 CHAPTER 2. (2.p) It = {u(q. as we observed.p)}. This formulation deals with observables u representing physical quantities. and the equation of motion (2.4) of an observable on the other.po are constant initial values — we have du0(q0. 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. Koppe [152] at the university of Munich around 1964.
(2.26)) (u) = Tr(wW). Eq.48): (a) TV u* = (Tr «)* = TV^.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. by Traceu:=Ju(q. ~[q.p} = l {q. B] := AB — BA in the following way: 'i {q. it is helpful to introduce already at this stage some additional terminology. (c) TV (u*u) > 0. In Chapter 9 we attempt a corresponding approach for classical systems with .2.q}=0.q} = 0 {p. also written "TV". that its use here implies the essential properties it has in matrix theory. so that we consider this next. Introducing it here assumes. Moreover. Thus we can write the expectation value of an observable u (cf. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.48) In matrix theory the symbol "trace" has a welldefined meaning. That this is the case. can be deduced from the following characteristic properties of a trace. which all apply to (2. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2.p)(^y. therefore. and to be able to correlate these with the above classical considerations. also called "density matrix".6) a socalled "commutator''' [A.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. ox In view of our later considerations. (2.) = TV (vu). for a phasespace integral we define the word or symbol "trace". px > in—.p] = l. (d) Tr(ut. (b) TV(cm + j3u) = oTV u + /3Tr v. y:\p. (2.p} = 0 • — ^ ~[q. if u ^ 0.
Hamiltonian Mechanics a wavelike nature. electrodynamics. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. and.40 CHAPTER 2. These aspects will be considered in Chapter 27. it will be shown that in electrodynamics.e. i. excepting the Poisson brackets. . the Poisson brackets require modification to Dirac brackets. a constraint) has to be taken into account. However. since gauge fixing (i.e. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation.
= {ipi} is called a linear vector space on the set of numbers I 6 {C}. A set M.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. and selfadjoint operators in this space as representatives of observables. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. 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. we can define the Hilbert space as the space of states of a physical system. if the elements ipi of M satisfy the usual axioms of addition and K 41 . In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. We found that the Poisson algebra permits extensions to noncnumber formulations. which turn out to be those of the theory today known as quantum mechanics. 3.e. of measurable quantities.e. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. i. Building upon this. i.
1) aipi + aipj. f > 0 if V T ^ O . . (V>. n.^i) .02)* (hermiticity). (ipi.ip2).ip2.i = 1.ipn are said to be linearly dependent.ipn are said to be linearly independent. fa. . and n is the smallest such number. E (3.^) called inner product. n exist (not all zero). if every vector ip E M can be associated with numbers Q . 2 .aiV>i + a 2 ^ 2 ) = ai(ip. ^ i = 0 if ^ = 0. if numbers eti. A + tpj = tpj + i'i. (aptyi. . i. (3. ( n (3. i=l (3. —»• IK. (ipi + ipj) + ijjk = ipi + (ipj + ipk). . . (3 5c) + a2{ip.^1) = (01.. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M).3) If all on = 0. i = 1. ip2 of this space can be associated with a complex number (V'l. with the properties (a* G IK): (^2. such that n ^ = YJCi^i. . are linearly dependent. the vectors il)i. (3.42 CHAPTER 3. . so that n 5 > ^ = 0.4) The vector space . 2 . if any two elements ip\.5b) .2) Vectors ip\. x M. n is called the dimension of .e. In each case n linearly independent vectors are said to form a basis. • • • .tp2.M./?eK). If n + 1 elements ^ € M.  . . (a. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers.M is said to be a metric vector space or a preHilbert space. .~ip2) '• M.5a) (3. .
The first two properties imply ( a i ^ i + a2ip2.3. (3. for ip\. (310) (^2.7) In addition the following relations hold in a metric space M.9b) = 0 (Pythagoras theorem). tp2 £ M: 1(^1.V') +"2*(V.^) = ai*(V'i.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation). »2 2 .tp2 £ M.^ 2 ) =   ^   2 + 2 ^ i . (3.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.. In order to verify Eq.9c) HV'i + tp2\\2 + \\1p1 . which we can write 0 < ( ^ i .V>i + A</>2) > 0 . (3.e.)> (36) i.is defined by d(ipi. antilinearity in the first component (also described as sesquilinearity.^1) 2 h 11 0 <%«e + M .9a) (3. ^ i ) + A ( ^ i . ^ i ) + A * ( ^ .2.9a) we start from if) = ipi + \if>2 6 M. ^ 2 )  . A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. WA+Ml2 = ll^il 2 + ll^2 2 . The distance between two vectors if>i.9d)   V i   = sup  ( ^ i .rh) (3. meaning oneandahalffold linearity). (3.V. linearity in the second component. if Wi. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). for arbitrary A and ip2 ^ 0: (V>i + AV>2. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.2 Hilbert Spaces 43 where the asterix * means complex conjugation..*h)=\\AH\(38) (3.^2)1 < H^ill • 11^211 (Schwarz inequality). ^ ) + A 2 (V2.
a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0.* Two vectors i f i .9e). .11).9a). In verifying the triangle inequality we use this result (3. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X .\\ih\\. (3. o ^ V i = o.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. (3.  ^ l l = (HlMI +   « 2 .^)1 <IIV>il.13) the vector is said to be normalized.ii) thus verifying Eq. ^ e M are said to be orthogonal if (</>!.44 or CHAPTER 3. (3.V2) IIV'l2 + V'22 + 21(^2. Mathematical Foundations of Quantum Mechanics 1(^2. V 2 ) = 0 .^1)1 Vi 2 + ^2 2 + 2   ^   . (3. ll^ill + llVdl. (3. HIHI (3.9d). for which CO 2 := V^l^l 2 < 00. Examples of metric vector spaces: (1) Let M. (3. so that   ^ l + ^ 2   <   ^ l   + ^2. hence verification in each case is necessary. beginning with A = 1 in the second line of Eq.12) We omit here the verification of the remaining relations (3.9c). be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1.
The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. Then L2 is the space of all these equivalence classes. (3. i.5c).w) :=J2v*Wi. In order to avoid this difficulty. = C? be the set of all complexvalued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 /(x) 2 d 3 x < oo.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). (3. Js the space C? is not yet a metric vector space although for (cf. •'^ \ 0 otherwise. But this applies also in the case of any function which is nonzero only on a set of measure zero. / ) = 0. with inner product oo 45 (v. (3. for which the scalar product. 2 and so on. . all squareintegrable functions / which are "almost everywhere equal".5b).5a).e.5c)) / ( x ) = 0 = > ( / . */ x } = / /o for x = 0. which differ solely on a set of measure zero.3. and one defines addition and multiplication by complex numbers with respect to these classes. (2) Let M. i.e. Eq.g)= [ /(x)^(x)d3x. which satisfies relations (3.[$]):= / JM /(x)* 5 (x)d 3 *. is defined by ([/]. etc. are combined to an equivalence class [/] (with space L2). Elements of the classes are then called representatives of these classes. With the scalar product (f.
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. \v) E V are said to be orthogonal. if \u) = 0 (see (3.5c)). Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. if (u\v) — 0. (4. that the vector \u) E V is associated with a vector in V. Two ketvectors \u). Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV .9a): \{u\v)\2 < (u\u)(v\v).5) 4. The dimension of V is the same as the dimension of V.e. the norm squared. or (4. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. which we write (u\. This expression is linear with respect to \u) and antilinear with respect to \v).62 CHAPTER 4. (3) (u\u) = 0.(t.4b) (4.4c) (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). 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. It follows that we can now write the Schwarzian inequality (3. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*.3 Linear Operators. Dirac's Ket. V \u) (in the case of £) with infinite norm!).\u) = (u\0*.4a) \v) = J\(m<% and (4. if x\u) = 0 V kets \u) E V.and BraFormalism Furthermore we have: ( x  = 0. i.
i. Two operators A and B are said to be inverse to each other. defines the unit or identity operator.e. Furthermore. The product of such vectors is written: \u^u^) = \u)^\u)^. the space Vi <S> V2. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). 1 and \u) = B\v).. also acts on the dual space V. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. It follows that (r.4. the space Vi <S> V2 has the dimension .or Kronecker. Multiplication by unity. BA = 1. This does not always exist. This product is commutative. The product vectors l?^ 1 )^ 2 )) span a new vector space. as we can see as follows. denned as a linear operator on V. Then one says: (771 results from the action of A on (x. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) .3 Linear Operators..\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). ltt) = \u). Let the following bravector be given: (x\ £ V. if AB = 1. let (771 6 V be the bravector defined by this functional. Then the following rules apply: A + B = B + A. on bravectors are similar to those on ketvectors.e. if \v) — A\u) i. If Vi has the dimension A/i. Furthermore linearity applies. AB^BA. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. B. Finally we define the tensor. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). In this way the operations of operators A. i.or direct product of vector spaces. The inverse of a product is (AB)'1 =B1A~1.. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on.e. and one writes: (V\ = (X\A. The operator A. The inverse of A is written A .
i. Let A^> be an operator in the space Vi and A^ space V2. AW\uW)\uW) = \vWUW). (v\t) = (u\A\t). i.and BraFormalism an operator in the Mi A/2.e. Since we demanded the scalar products to have the property (t\v) = (v\ty. If (v\ = (u\A. here designated with the same symbol. and (v\ = (u\A.e. i. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4.g.e. also Sec.e. Every operator A^> commutes with every operator A(2\ i. 3. then \v) is a linear function of \u). \t) G V. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). As a consequence we arrive at the following properties (e. Dirac's Ket. [A^. Next we construct the following scalar products: (t\v) = (t\Ai\u).3). Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. we obtain the conjugate relation <tAt«> = (u\A\t)* (4.7b) (4.7C) («)(u)t = \v){u\.6) for all \u).7a) (4. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. we write \v) = A*\u).A^] = 0. Every such operator is then also associated with an operator.7d) . in the product space.64 CHAPTER 4. AW\U)W = \V)W.
as is also (u\u). Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . but only if these commutators commute. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH.K]^ = = [HKKH]* KHHK = = K^H^ [H. [H. The operator H is said to be positive definite.3 Linear Operators. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A].K}. Analogous considerations apply in the case of bravectors (u. or (u\A\u)* — {u\A^\u) = (u\A\u). It follows that a is real. An important theorem is the following. so that (a)(a)t = a)(a.K\. The commutator of two hermitian operators is antihermitian: [H. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H.4.e. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. Obviously the quantity \a)(a\ is an hermitian operator. 65 The linear operator H is called hermitian. i. i. hermitian The product of two hermitian operators is not necessarily hermitian. An hermitian operator A has the properties (a) that all its eigenvalues are real. if H — H' and antihermitian. . so that (n^4n) is real. if H is hermitian and (u\H\u) > 0 for all ketvectors \u).e. Verification: Since A — A^ and A\u) = a\u). it follows that (u\A\u) = a(u\u). for we had (\u){v\)i = \v){u\. K] = 0.
) = 0. then the hermitian operator defined on this space is an observable. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. Ps is idempotent.66 CHAPTER 4. i.9a) . Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. (4. i. Let S be a subspace of the Hilbert space !K.e. A continuous spectrum can immediately be included by passing to the extended Hilbert space. The above theorem remains unchanged. If these form a complete system. The entire set of ketvectors spans the extended Hilbert space.e. i. so that (u\Ps = (us\. Ps\ua. i. (u\Ps = {us\. but the continuum vectors are normalized to a delta function. (n\n')=5nnl.9b) (4. \u8*) E S*. a ^ b. Assuming A\u)=a\u).e. Next we introduce the very useful concept of projection operators.Au) = a(v\u) — b(v\u) — (a — b)(v\u). so that (v\u) = 0. (u\u') = 5{v .u'). as we can see as follows. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). Dirac's Ket. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). {us\us*) = 0. Ps2 = Ps(4. and let S* be its complement. we have 0 = (uArt) — (u.e. ( n  z / ) = 0 . since a ^ b. (v\A = b(v\. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.8) The projection operator has the following important properties: Ps is hermitian. since a is real. which includes vectors of infinite norm.and BraFormalism Moreover.
67 (4. so that \ua) = (a\u)\a) = \a)(a\u). as can be seen as follows.4. i. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). (1 — P)\u) are orthogonal. (1 — Ps) is projector onto S*. . 2).e.e. (4. P\p) =p\p). Ps is projector onto S..9c) This property can be seen as follows.p = 0. where £ is a continuous index. Then 0 = (P2P)\p) and hence p = 0. i. \ua) = c\a). Very similarly we can construct operators which project onto the subspace spanned by the continuum states. if the set of vectors 1). = (p2p)\p). so that (Ps2 ~ Ps)\u) = 0.. Let p be an eigenvalue of the projection operator P. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). \u) arbitrary. i.e. c = (a\u). i. Set \u) = \ua) + \ua*) with (a\ua*) = 0.. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. The quantity \a)(a\ is called elementary projector. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c.3 Linear Operators. iV) is a set of orthonormalized states.e.9d) Thus the vectors P\u).1. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. p 2 . . Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. Let a continuum state be written £).
Let \u\).114b). Dime's Ket. and Eq.1 and 11. 4. In general it is possible.68 CHAPTER 4.6.g. that one and the same eigenvalue Aj is associated with several eigenfunctions. Examples 8. Sec. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. which we introduced earlier.r)(ui. are linearly independent). .6.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.. Coulomb potential. Degeneracy will be discussed at various points in this text. 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. be a system of basis vectors in this space. Observables are representatives of measurable quantities.114c). 8. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. one says the degree of degeneracy is r . However.4 Observables Operators which play a particular role in quantum mechanics are those called observables.e.. If there are r such vectors.. the operator P is correspondingly P2\u)= f2\0m\u). i.e. (11. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). see Eq. (11. u2). P = £n)(n. which are orthogonal to each other (i. See e. in this case the spectrum also has a continuous part.r\ "An example is provided by the case of the hydrogen atom. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.
r)(ui.r\ = l. or also as subdivision of unity or of the unit operator. and APi = XiPi. The operators Pi are linearly independent. then PiPi' = 0. the relation (4. (4.10) follows from the fact that i Applying the operator A to the projector (4. ^ A i P i = A = ^Aiui.r) (the convergence. Together with the orthogonality condition.4 Observables The dimension of this subspace is that of the degree of degeneracy.r>(ui. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r \ui.r\u).r\u) i.r.r \(ui.r'} = Sii>6rr>. the operator A is completely determined by specification of the eigenvalues A.\i)Pi = 0.r (4. the projection of this operator is the entire space. in the case of degeneracy with (ui. i. which we do not enter into here. 69 i If A is an observable and if the spectrum is purely discrete. (4.e. i.e. and the eigenvectors \ui. Let us set {A .4.r\u)\2. is always implied).r =^ i.r)(ui.10) expresses the completeness of the orthonormal system.e.10) gives the linear combination it) = ^ i. Applied to an arbitrary vector \u) G "K Eq. i i (4.r\ui/.e.10) This expression is known as completeness relation or closure relation. PA = y£pi i = Yt\ui.10).11) i.r)(ui. we obtain i i i i i. If Aj ^ Aj'. The uniqueness of the expression (4. i.12) .
14) . (4. the eigenvector \ui): f(A)\ui) = f{\i)\ui). Dime's Ket. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. for instance. we have the corresponding generalizations. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1.13) 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).e. Then the operator PA which expresses the completeness of the entire system of eigenvectors is (all continuum states assumed to be in the interval (yi.70 CHAPTER 4.v').e. written in Dirac's notation. If the spectrum of an observable A consists of discrete as well as continuous parts. (3.and BraFormalism This is the expression called Parseval equation which we encountered with Eq. for instance exponential functions. now however. One defines as action of f(A) on. as explained earlier. i.17) previously. i. (4. Let v be the continuous parameter which characterizes the continuum. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). The ketvectors \uv) are orthonormalized to a delta function. In a similar way we can handle functions f(A) of an observable A. (u„zv) = 8{v . > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv.
and bravectors. x) € Fx.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. if (a) they all commute pairwise. for which the completeness relation is f dk\k)(k\=l. which all commute with one another. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.16) represent subdi .. with energy E. (4.58) in terms of ket. form a complete set of commuting observables. their commutator vanishes. Therefore we want to reexpress the Fourier transform (3. (3.15) and (4. For many practical purposes the use of the Fourier transform is unavoidable. here presented without proof.e.17) The Fourier transform provides the transition from one representation and basis to the other. if and only if they possess at least one common system of basis vectors.4.e. i. 4. \k)€Fk. and (b) if they possess a uniquely determined system of basis vectors.59). We shall return to this theorem.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. Since both expressions (4. again later. [A.C. Extending this we can say: A sequence of observables A. Eq. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. B].B.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.. In the onedimensional case we write {x\x') = S(xx'). i. First of all we can rewrite the integral representation of the delta function. (4. In the following we consider more generally ketvectors \ip) as representatives of the physical states. / dx\x){x\ = 1. as a formal orthonormality condition.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies).. where the symbol ip is already indicative of the Schrodinger wave function ij. (4..
18) According to Eq.and BraFormalism visions of the unit operator.72 CHAPTER 4.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). Rather \x) and \k) serve as basis vectors in the representation spaces Fx. we can rewrite Eq.e. ${k) := (k\if>).Fk.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.59) or Eq. (x\k) = ^=eikx.space representation. called wave function. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. il>{x) := (x\i/}) : "K * C. which are representatives of the states of our physical system. Dime's Ket. i. (4. (4.e.e.1a).22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).21) provides the ketvector \ij)) in the A. (4.L e " * * ' = {x'\k)\ V27T (4. this expression has to be identified with — / dkeikxeikx\ i. 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). <a#) = [ dk{x\k)(k\ip).19) Comparison with the orthonormalized system of trigonometric functions (4.2a) etc. i. .15): 5(x . V 27T (k\x') = . (4. Obviously we obtain this by inserting a complete system of basis vectors of Fk. (4. (3.
e.t). with the complementary system or *E. i. which is the quantum mechanical analogue of the classical probability density p(q.p. We mentioned earlier (in Sec.5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. 1. 73 .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. 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. if we do not take into account the complementary part of the universe. 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). the calculation of which for specific cases is the subject of Chapter 7. 5.2 The Density Matrix We shall establish a matrix p. Schrodinger [244].Chapter 5 Schrodinger Equation and Liouville Equation 5.
^ Since p is hermitian. Schrodinger Equation and Liouville Equation the measuring instruments...i. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. T . \a).. but also of the states of the complementary set of the universe. i. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. (a\i)=0 = {j\b). 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). which in Dirac's notation we can write as = J]C a »i) a.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. are called "pure states". we have to deal with a socalled "mixed state". where T means transpose: p = CTC*. \a).b. These are the quantum mechanical states which are also referred to as microstates. we have (a\b) = 5ab. a (5. The actual and real state of a system is then a superposition of two types of states.e. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij.. where i represents collectively all quantum numbers of the state of the system under consideration. (52) (5..74 CHAPTER 5. of an operator that acts only on the states of the system we are interested in. i.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix.3a) and N J  a ) ( o  = l in the complementary subspace.. p" ( C C * ) t = CTC* = p. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe.i. \i). is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a.b. The states \i) £ "K. called the density matrix.e. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete.3b) The quantum mechanical expectation value of an observable A.. that is.
a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.j(x'i)(ia. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).7) p = ^2\ii)(i. a projection operator. This transition to a new set of basis vectors in which the matrix becomes diagonal.x) := (x'\p\x) = yju. in a specific representation. as we discussed previously.3a).e. which today is a fundamental term in statistical mechanics. P ^ ) = i)(#) = (#)i) (5. but occupy in general different microscopic. or (5. e. (see below) eM.10) and is therefore. i i (511) In Eq. (5. If the ensemble corresponding to the system under consideration consists of Z elements (i..5.e. In this sense we have p(x'.. which are all subjected to the same macroscopic boundary and subsidiary conditions. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. in the position state representation or xrepresentation.9) is diagonal.3 with Pi^YtCaiCaj*. The ensemble can be represented as a set of points in phase space. quantum states.2 The Density Matrix 75 \i') 6 "K.e. (5.e.6) (i'\p\i") = Wi'Si'vi = real.2) and (5.g. . Gibbs introduced the concept of ensemble.ji(x / )i*(x).^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. i. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). In the following we write simply \i) instead of \i')..* We can specify the state of our system. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. Thus the real system is in the pure state \i) with probability uii/ J2i Ui. (5. where x represents collectively the entire set of position coordinates of the (particle) system.) = YJc t . i. This ensemble is interpreted as a large set of identical systems. as we saw earlier. seen macroscopically of Z copies of the system).\p\i")(i"\ = Y.4) we had the expression (A) = Y/Pij(j\A\i) i. a complete orthonormal system with properties (5.
Eq. We can also show that Ui > 0.14) which we can write (A) = Tr(pA). Thus Tr(j>ii)<i)=l. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). On the other hand.76 CHAPTER 5. (5.18) Substituting this into Eq. as claimed by Eq. (5.18).16) (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. according to Eq.) := (mm = (#') W> = l(#')2 > 0. we have Tr(p) = l. (5. (A. (5.5) above) (i\p\j)=Pij.17) Hence the sum of the classical probabilities is 1.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). 3 (5.3 = pvv = uv.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. Then the expectation value of the observable A is (cf. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. (5. (5. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) . (5. In particular for A = 1. Eq.4). Hence Wj/ > 0. i t (5.4)) (5.
We now consider the latter expression. In the following we are mostly concerned with considerations of systems in pure states.3 The Probability Density 77 We see therefore that p defines the mixed states.24) s (x\x') = 5{x . represents a quantum mechanical probability. We observed above.t) = \^{x. the only remaining one is 1. (5. For the case n we have TJ=J (520) (5. i. t) p= i The expression (5. i.8). (5.t)\2. If all ui but one are nought. in all other cases the state of the system is a mixed state. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. the system is in a pure state. since they do not take into account the (interaction with the) rest of the universe. so that p{x.t) = {x\i). not in a mixture of states. Equation (5. which is contained in the coefficients Cai. (5.3 The Probability Density p(x.23) Instead of i(x.22) In the following (x\i) is always to be understood as (x\i(t)}.t). Thus we have a system in a pure state.x').t) and instead of (x\Pi\x) we now write p(x. We see this more clearly by going to the position space representation. that these are actually not sufficiently general. by representing the vector \i) by the wave function i(x.e.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l.20) expresses that the system can be found with probability 1 in the state \i). but we see here. defines the statistical operator. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. ^2ui\i){i\. thus enters these states. i. The effect of the interaction of the system under consideration with the surroundings. (Question: Is the state of the universe a pure state?) 5. that — different from u>i — P.e.5. . one says. and in the other states with probability zero.t) we now write tp(x.e.21) p=\i){i\=Pi.
8).t) = \rp(x. (5.27) where H is the Hamilton operator of the system. To this end we have to differentiate the density matrix p. Schrodinger Equation and Liouville Equation The relation (5. We shall convince ourselves later that the postulated equation is sensible (in fact.t)\2.t) = / dx{i\x)(x\i) = (i\i) = 1. (5.27) is known as the Schrodinger equation.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. The generalization to a threedimensional position space is selfevident: JK 1 / p(x. This differentiation requires the time derivative of a state vector \i) E "K.8) and multiplying by ih. (5. . We now want to obtain the quantum mechanical analogue. (5. Equation (5.28) Differentiating Eq. t). t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).3.25) J Thus the particle whose state is described by \i). or the wave function tp(x. 2. 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. but the linear operator states are moving — even if the system is observably stationary). can be found with probability 1 in the space R1.22) can also be written (xi) 2 = (i\x)(x\i).e. We therefore postulate first an equation for the time dependence of a state vector. with respect to time.e. (5. (5.27) is iHJ\ jt= m . the Schrodinger equation is always postulated in some way.78 CHAPTER 5. and hence the integral over all space dxp(x. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. the expression (5.^ The equation which is the conjugate of Eq. it is not "derived"). which is chosen in such a way that the desired analogy is obtained. i. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).26) 5.
The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.28) the states are considered as timedependent. p also contains "quantum mechanical probabilities" (i. Eq. Eq. i. dp/dt ^ 0.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.e.e. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. i. The reason for this is that in addition to the "classical probabilities".5.e.31) The eigenvalues u>i of p determine the fraction of the total number of systems of the ensemble describing the actual system which occupy the state \i). We thus obtain the operator form of the Liouville equation. We can now consider the Schrodinger equation (5. not the partial derivative. we observe that here on the left hand side we have the total time derivative.28) we obtain (see also Eq. (5.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. (2. (3. both uji and i)(i). (5.p} + ihJ2^\J)(J\.32) with its classical counterpart (2. (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi. (5.32) Comparing Eq.27) and (5. (5.27).21). M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.34) .p}. (cf.Xi)\j).24)).37)) (5. such a formulation is described as Schrodinger picture. (5. the relation ih^ = [H. With Eqs.
the state \i) is still timedependent. P. in p = ^2i^i\i}(i\. we write V>(x.and temperaturedependent Green's functions. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1.38) (x\j)t=o = Yt(x\E^(En\J)t=o. but with respect to the parameter /3 = 1/kT. with respect to time t. n (537) This equation is described as the timeindependent Schrodinger equation. in Sec.t) = exp ip(x. (5. Feynman [94].35) can be written'! ip(x. 5. (5. 0) or \il))t = \il>)t=o exp (5. since the solution of Eq.e. i. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. t ) .. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.t) = Hld .36) in in The initial wave function or timeindependent wave function ^(x. appearing in the Boltzmann distribution.Xi){x\En) = En(x\En). which we shall need later.36) however. to) = exp[—iH(t — to)/H\.** In view of the close connection between time. .0): where H(ih—.x 1 ^ ( x . (5. we can replace these states by corresponding timeindependent states. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. a (5. „ / ith—. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x.8) for p. T meaning temperature..e. it is plausible to refer to this equation at this stage. however.4.80 CHAPTER 5. or H\En) = En\En). 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. With the help of Eq..35) Here we can look at the Hamilton operator as a "time generator3''. In expression (5. we shall describe as "time development ator given by U(t.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.t): > ih—</>(x. operator" the exponentiated oper . i. 10.4.
or ^ = ^ e 0Ei _m.1).45) since on the one hand •PEi i i i and on the other hand . (543) Since H(f>i(x) = Ei<f>i(x). i (5. (5.5. (5. (5.x'(3) = Si e ~^W*V).40) 1.39) therefore (x\p\x') = J2"i(x\^)(Ei\x').4 Schrodinger Equation and Liouville Equation 81 Example 5. we obtain p(x.e.42) Inserting here Eq. Eq. i. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. x') = ^ i LUiMxWix').41 ) or with 4>i(x) := (x\Ei) this is p(x.^ n tne position space representation Eq. (1.x')^p(x.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. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.44) r (TrePH)' (5. (5.40). we can rewrite p. i ( 5 .
x.x'. Schrodinger Equation and Liouville Equation so that p is given by Eq. This solution is the Green's function or kernel K(x.45) and (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.(3) := (x\pN(P)\x') = (x\e~0H\x'). (3).45). we obtain 9pN gpP) = SijEie^ = EiPNij(J3).47) with respect to (3.51) is seen to be very similar to the Schrodinger equation (5. (5. Tr(pN) = / dxpN(x.46) we can write the expectation value of an observable <^Tr<„. (5.x'.51) where the subscript x indicates that H acts on x of PN(X. (5. Example 5. [B.49) In the position or configuration space representation Eq. (5. Equation (5. (5.X'.L [B. (7.48) = 1. (5. (5.45) without normalization as PN(J3) := e~pH. (5.50) Differentiating this equation with respect to j3. whose calculation for specific cases is a topic of Chapter 7 (see Eq. [A.4> = ^ g ^ .x'.tt tt For a more extensive discussion see R. B] + L [A. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5./3).82 CHAPTER 5. Differentiating Eq.51). . A}} + • Solution: The relation can be verified by expansion of the left hand side. Wilcox [284].47) (5. We now rewrite the factor exp(—(3H) in Eq./?). (5. B}} .(3). we obtain = HxPN{x. (5.52) Hence this expectation value can now be evaluated with a knowledge of pN. With Eqs.46) is pN(x. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.35).55)). M.
its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. that the oscillator. * Comment of Y.. can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory.51) to (8. S.53)). 83 . Since the spectrum of the harmonic oscillator is entirely discrete. An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs. Kim and M. E. However. can — in fact — be quantized in quite a few different ways. This is a fundamental topic since the harmonic oscillator. like the Coulomb potential." Quite apart from this basic significance of the harmonic oscillator. its quantization here in terms of quasiparticle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator.the present form of quantum mechanics is largely a physics of harmonic oscillators. 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). The importance of the harmonic oscillator can also be seen from comments in the literature such as:* "..Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. Noz [149]. (8. we shall see on the way.
2) which for mo = 1/2. of which the only nontrivial one here is [q. which is called Weyl ordering. This can be done in a number of ways. 2mo dx \ 2 2 (6. which is — in the first place — representationindependent. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x).' ^mgLOh by setting and *~. Quantum Mechanics of the Harmonic Oscillator 6. . To this end we introduce first of all the dimensionless quantities rriQUi fl q. p^ih—. However. h = l.^ The first problem is to determine the eigenvalues and eigenfunctions of H.84 CHAPTER 6. we can also proceed in a different way.2 The OneDimensional Linear Oscillator In the case of the onedimensional. When such a term arises one has to resort to some definition.to = 1 reduces to the parabolic cylinder equation.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers.p]=ih (6. hence we do not use a distinguishing notation.1) for the hermitian operators q. This equation would then have to be solved as a second order differential equation. Also observe that H does not contain terms like pq or qp. so that there is no ambiguity due to commutation. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. in the present case + [EmooJ2x2)ip = 0. like taking half and half.p. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . For instance we can go to the special configuration or position space representation given by q^x. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions.
34d): [A.14a) we deduce for an eigenvector \a) of A^A.e.e.B}C + B[A.1) we obtain immediately the commutation relation [A. i.7) (6. A\ we obtain h A + A^ rriQLU y^ (6. + 1). i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .14b) [A* A. With the help of Eq. Prom these relations we obtain ( ^ = ^ .8) The eigenstates of H are therefore essentially those of N := ^ f A (6. (6. if A*A\a) = a\a). (6.13) ( ^ A ) i t = A\A^A From Eq. (6. C] + [A. and [AB. (6. for ^ A  a ) = a\a). (6.A^] = 1. We now use the relation (3.12) in order to obtain the following expressions: [tfA.6. A^] = A\ (6. q are hermitian). (6. (6. (6. C}B. C] = A[B.6) and p = y/motoh AAi =^. A] = [A\ A]A = A. i.BC] = [A.5) Reexpressing q and p in terms of A.14a) (6.10) then a = (a\A^A\a) = \\A\a)\\2>0.1 ) . A*} = A^A.C]. .9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a.11) Thus the eigenvalues are real and nonnegative.2 The OneDimensional Linear Oscillator 85 (p.
2)A a). Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. or \\A\a)\\ = Va.l)\a) = (a .20) .2A)\a) = A(Aa .15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). in view of Eq.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a).e. unless A 2 a) = 0. In this case we have (A*A)A2\a) (6 (6. A 2 a) is eigenvector of A^A with eigenvalue (a — 2).16) This means.l)A\a).86 the relation CHAPTER 6.17) (6.2A)\a) 2 (a .19) i. (6. However. (6. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).11) this is not possible. or Ata) = v ^ T T . but (b) An\n>\ llA An+1a)=0.l)A\a) = A(A^AA .  a) is eigenvector of A^A with eigenvalue (a + 1). (6.A)\a) A{AAU . This would mean that for sufficiently large values of n the eigenvalue would be negative. since this equation implies that the eigenvalues cannot be negative. (6.18) = '= a ) A(A^A . If we continue like this and consider the vectors An\a) ^ 0 for all n. (6. Thus for a certain value of n > 0. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a.l)\a) = A(a . Similarly we obtain from Eq. (6. unless A\a) = 0. Next we consider the vector A 2 a).. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .21) . unless A^\a) = 0. we must have (a) Let \an):= Ana)^0. (6.
But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0.e. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.18) we obtain for a = n: p t  n )   2 = n + l.' " /. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.20) we obtain from this that the right hand side vanishes. (6. (6.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).24) This is a very important relation which we can use as definition of the state vector 0)..20) we have for a = n = 1: A2\1) = 0.20) that A0) = 0. (6. Moreover. or a = n > 0.17) a by (a — n).n ) = " A»a>=» ' = A«a)F = 1 . \ I 2 2 x 87 Replacing in Eq.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. For a — n = 0 we deduce from (b) of (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . that an = 0. (6.. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. i. (6.25) (6. v.6. I „ .22) With relation (b) of Eq (6.n  a .27) According to (b) of (6. In the following manipulations we always use the commutator (6.5) to shift operators A to the right which then acting on 0) give zero. From relation (6.I n U . The state 0) is called ground state or vacuum state. \ 9 \ I / II 4 T ) . (6.28) .
as we can see as follows.(6.(A*)n] = n(Ai)n\ (6.88 and CHAPTER 6.26) the equality l) = A t 0). (A^)3] = [A.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).11) we have [A. (6. The states \n) thus defined are orthonormal. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0. (6. we have 2> = ^=dAi\0). and in general [A.11). (0^ 2 (At) 2 0) = 2. But using Eqs.29). Similarly we find 2)oc^Ut0). i.29) Hence we obtain and in view of Eq. A^A* + A*[A. In general we have \n) = ^(A^n\0) (6.e. (6.5) and again (6.32) (6. (At)2]A+ + (A*)2[A.33) .32) we have (nm) = .34) (6. and in view of Eq. According to Eq. (A*)2} = [A.31) (6.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. (6. A^} = 2A\ and [A. l>ocAt0). (6.^ ( O l A ^ n o ) .
in quantum mechanics. .34) and (6. (6.36) + 1) (6.'n\ and = v n ![ ( A t ) M i= + (6.1) = n\n). .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 .8) we obtain therefore H\n) = hu (A^A + ].41) The contribution hu/2 is called zero point energy. . we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). (6.1 0). 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 .39) A^A\n) = y/nA^\n . . whose number is given by the integer eigenvalue of the number operator N = A^A.40) (6. .J \n) = hw (n + i J \n).35) the operators A^ and A are called respectively raising and lowering operators or also shift operators.e.35) Inserting this result into Eq. With Eq. . n = 0 . Here.38) (6.6.1 (A + ) m .32): rf\n) = . (6. (6.1) (6. (6.1 ) . •(» (6.33).L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.34) A\n) = ±=A{Al)n\0) . ldnm = n\5nm. chosen in close analogy to the postulated "second quantization relations" of field theory. 2 . i. . l . The terminology is. however.37) n^)711}^) = v n  n . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv. we obtain (n\m) = 5nm We also deduce from Eq. In view of the properties (6.
The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. other elements zero. (6.6) and (6. other elements zero.. i..44) yfn. Vi 0 0 0 2mQUJ 0 V2 0 y/3 0 0 0 V3 0 y/l 0 \/2 0 V3 0 \ V • / P = mohu) VT 0 0 0 ° VT 0 V2 0 0 / 0 0 V3 0 (6. is defined by projection of operators on eigenfunctions of energy. (6. Eq.46) v^ . A^ from the relations (6. (5. En = hu(n + ).45) Correspondingly we obtain with Eqs.3 The Energy Representation of the Oscillator The energy representation. (6.1.g. ( 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..90 CHAPTER 6.43) = = y/n(n'\n — 1) = y/n5n^ni. and — in fact — we have used this already previously (see e.39)).. Vn + 1.. Quantum Mechanics of the Harmonic Oscillator 6. n = 0.2.42) We can deduce the energy representation of the operators A.39): {n\A.7) the energy representation of the operators q and p.'\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 . also called Heisenberg representation.e. (6.38) and (6.
6. (6. The ground state wave function (/>o(x) is defined by Eq. (6. (6.24).6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w . i. A\Q) = 0.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). by (cf. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h. This is a simple differential equation of the first order with solution (x0) = The normalization constant C is determined by the condition* OO /C 1 = (00) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / OO J —< emou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero. 2 .47) LP 0) = 0 .3)) ^ (q 2n \ + (6.1) and (6. mow / (6.e.5) are also satisfied as matrix equations. (6. Eq.4 The Configuration Space Representation We saw that the eigenstates are given by Eq.32).48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}.
e. i.2 n n! The functions Hn(£) are obviously polynomials of the nth degree.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 .56) < ^ = We have as an operator identity the relation . it suffices according to Eq. c/>n(x) := (x\n) = .l ) > n ( x ) .92 CHAPTER 6.32) to apply the appropriate number of creation operators A^ to the vacuum state 0). (6. Vn! Now {XlA (6. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. ) e"^2.51) \l 2h {X q \ m0uP)\l 2h \X m0u. they are called Hermite polynomials.l ) n f f „ ( 0 . These polynomials are real for real arguments and have the symmetry property # n ( .54) (6 55) i/4 rzri7r 1 / 4 vo. so that <t>n(x) = ( .0 = ( . (6. (6.L ( x  ( A t ) »  0 ) . In order to obtain the wave functions of higher states.
57b) • Generalizing this result and inserting it into Eq.58a) ' T h i s definition — apart from the usual one — is also cited e. (6. (202"2.12(20 .4 The Position Space Representation i.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f.58b) by W.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. (2036(20.g. .e.en_ dt. p. d£" (6. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.2)m = (±t)m for some function VKO. Magnus and F. e.12. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.6. ( 2 0 4 . 81. Oberhettinger [181].54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . 93 . 2£. (6.
60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.94 CHAPTER 6. £33£.He'n{£). Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. £. (6. £2l.6 £ 2 + 3. Magnus and F.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. 81.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 .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.^ + n) He n(0 = 0 (6. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. (6.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . (6. £ 4 . (6. = ffen+itO+ntfeniCO. 80. pp.59b) = (l)^2/2^. . (6.60b) J—< "See for instance W. Oberhettinger [181]. = nHe^iO.(e^/2) or Hn(0 = 2n/2Hen(V2ti).
p. Their normalization is therefore given by OO /"OO / OO J —OO z (6.4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( . (6. and so on.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S. W.64) n\ See e. F(S.l ) ^ e^HeniO 2 / 4  . .6..0) we obtain + s fdFs 1! V ds s=0 + OF ds 2 dF ds2 = s=0 2(f d S )^«)a s=0 = 2£ = ffi(0.^ = e^/42~n/2Hn(aV2). 93.63) The meaning is that if we expand F(£.61) e?l*2nl2Hn{Z). Magnus and F..g. Thus the generating function can be written F(Z. Oberhettinger [181].s) = eM?(Zs)2}(6. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.s) = J2 n=0 (6.s)=F(£. s) as a Taylor series.
But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. We also see from the eigenvalue t h a t the energy of the lowest (i. 6. i. In classical mechanics the energy of the oscillator can be zero. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig.e. ground) state is fko/2. i.1. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). Quantum Mechanics of the Harmonic Oscillator Fig. in order to estimate the lowest energy of the harmonic oscillator. has parity + 1 .96 CHAPTER 6. for x(t) = 0 and p(t) = mox(t) — 0. mow' Apx = ^m0Luh/2. We see t h a t the lowest state is symmetric. In fact we can use the uncertainty relation AxApx ~ H/2.e. 2 .1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator.e. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. and we see t h a t even and odd states alternate. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . 6.
Thus. we obtain the integral . and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q).6. This observability can be taken as direct proof of the uncertainty relation. The analogous Example 6. and is directly observable. which is the complete orbit. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. It is clear that the uncertainties are not due to deficiencies of measuring instruments (we are concerned here with a system in the pure state 0)). classical orbit i.. to —a. mass mo.plane pole at z=0 ^ . Example 6. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. The kinetic energy is always positive or zero. like twoatomic molecules. n*h. Sec.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. 27rmoi/ <f> V'a 2 — q2dq = n*h. pdq.plane Imz z . But it has a pole at infinity.: 27 T mov2a2. back to 0. original path q . to 0.1/a O l/a Rez all at infinity Fig.1. In the plane of complex q the integral has no poles for finite values of q. According to "old quantum theory" with integer n. In Example 6. E .4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. /classical orbit Here q goes from 0 to a.4) to obtain by contour integration the eigenvalues of the harmonic oscillator.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". 14. in corrected version instead with n* = n + 1/2. 6. setting q = 1/z.2 The original path and the transformed path at infinity.e.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf.
) hv. hence we consider the following differential equation in which n = 0 . We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . (m .a) m = 2 .e. 6.V62 a2).98 CHAPTER 6. « qdq + I2 = f dq . harmonic oscillator and Coulomb interaction. in q at infinity). i.( 6 .e. z = 0 Vi 2Kmov(Ka2) = n*h = E/v. Example 6. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points).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.1)! (6.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. are the most basic and exactly solvable cases in quantum mechanics. E = Z2e4/4(l + n+ l ) 2 (cf. (11. we start with a Laplace transform ansatz for the remaining part of the solution. .5 . . In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.1.c2 = (I + 1/2) 2 (natural units.2b = Ze2. . for instance — without deriving all properties of the solutions.. but here we leave this .5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. After removal of a Gaussian or W K B exponential. i.3.65) f ^Va2z2 . a J \d Solution: The solution proceeds as in Example 6.1. and obtain with this all properties of the solution. since these two cases. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o . Eq.5.2TTI a ^ = .2: Contour integration along a classical orbit Setting q = 1/z. / " ^ . In the following we investigate in more detail the important differential equation of the harmonic oscillator. 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). 6. With a2 = E. 2 .e.114a)). £ = n*/w = ( n + . 1 .
66) First we have to remove the £2term. Oberhettinger [181]. (6. (this determines the limits). W.J ±(ps(p))e*dp.e. . i.g. Substituting these expressions into Eq.t*ij& + ny{t) = O.68) § = J p2s(p)e*dp. so that the integrand of the remaining integral vanishes.g.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . Eq.69) **See e. (6. (11. This has to be true for all values of t.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) . . 80. We demand that C(p) := — \ps(p)e~pt] = 0. Then (with partial integration) *dt = ~*/Ps(p)e. (6. fp{ps{p)) = {v2 + n)s{p). Magnus and F.111)) the Mellin transform of e~p is n!.6. E. and / dpf(p)pn the Mellin transform of / ( p ) .67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6.i ^  (ps(p)) P n P Integrating this expression.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. p. (6. (cf. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). / dpf(p)e~px the Laplace transform of f(p).p t dp = \ps(p)ePt} .
Therefore we take ptp2/2 y(t) .. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. we can choose a path from zero to infinity (the condition will vanish at p = 0).3. = / • Im p Rep n not an integer Fig. The value of the exponential at this point is ~ exp(£ 2 /2).70) = 0 P. n must then be an integer. so that p cannot go to infinity.100 and hence CHAPTER 6. then for large t we have y(t) ~ exp(i 2 /2).67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. Hence a possible path of integration > is from — oo to oo. consider the exponential in the integrand. For singlevaluedness. 6. The quantum mechanical wave function is then (cf. of course).n+l between the two limits of integration. so that y(t) ~ exp(i 2 /2). when p = —t (which is possible. We now investigate possibilities to satisfy the boundary condition C(p) = 0. i. We can show that if \p\ is allowed to be large.3 The path when n is not an integer. (1) This will be satisfied if \p\ — ±oo. 6. (3) We can choose a contour once around p = 0 in the complex pplane. as indicated in Fig. This exponential assumes a maximum value when pt \—p2 = minimal. (2) If n < 0. (6. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity.e. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. To see this.71) . This can be attained if p is allowed to be large (t can be of either sign). Eq.
. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . . this becomes jfj. . ..if n is odd.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.5. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).. The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. We can also obtain the derivative form of this Hermite polynomial. We have 27T! Setting q — p + t. In y(t) above. if n is even and v = 1. 4 .„y2 v l { ' ^ r (^) \ 2 ' where v = 0.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( .. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . (676) .P 2 / 2 00 C—tY ( .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"i2/2.3. pt p2/2 (6. 2 .6.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.
n— 1) = n.g.102 CHAPTER 6.77) = t. 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). and elsewhere. Starting from Heo(t) = 1.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. one can construct the tower of polynomials Example 6. Reinterpreting the remaining integrals with Eq. (6.1 '2dt using Hen(t) = (l)n27 r . (n. Hen(t). (6. n — l)Heni(t).70). Solution: We have to evaluate °° 2 .n + l ) f f e n ^ i ( £ ) + (n.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. ( n . Pptp 2 / 00 /2 dp. we obtain tHe„{t) (l)"n! ~ i.59a). We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.e. for the evaluation of expectation values. 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+<.4. in radiation problems.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . (6. Hei(t) Henj. Example 6. 8.75) as Hermite polynomials. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6. n + l) = l.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. Example 6.3 and 6. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). as in the normalization of asymptotic expansions of Mathieu functions).
(6.q) = = e . so that /2 Q2(n+rrf in / ( p . g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. g) is n! Next.^ + 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 .82) ^ .66) is \Zn\V2w Example 6.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).78) \/27rn!.80) f(p. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(. 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.n)\ i s an+2rij. and 0 otherwise \/2n(—l)n+mn! if n = TO.6. zero otherwise (6.jQ) n n!I ^—' ^Q fi\(n ./(n __ 2r — /i)!.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.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.5 nm . Hence the normalized wave function of Eq.q). (6.
The result then follows with half the normalization and orthogonality integrals (6.r .v)\ ^ Multiplying both sides by the value of the normalization integral. n + 2)(n + 2.l . .77). 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) + .A i ) ! (6. n + l)(n + 1.3 we know that upgoing coefficients are 1 and downgoing coefficients (n. As an example we obtain for the integral from 0 to oo (also reexpressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e ." ) a2^) r coefficient of (2s)! in f^ (s . Prom Example 6. /i = (n — v) _ / r o o e .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . n + 2) + ( n .v)\(s .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . n ) ( n .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. n — 1) are given by the first index. we obtain the result (6. n .r .Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ .^ / 2 * .l ) ( n .2s _oo ( a 2 / 2 ) i . i.(t)e«2/2dt Now.( n + Jr).— — ^ . n + l ) ( n + 1. .e '' ^ o^M ! ( n .e. the relation (n.n+ l)(n + 1.ra + 3)(n + 3. n + 2)(n + 2. n + 3)(n + 3.78). n + l)(n + l . Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). n)(n. 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.83) = (n + 2r)!e a / 2 V „ „. i / ! ( n .' ..v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .n + 2) + (n.e. n.i / ) ! ( 2 r + i/)! > . i. K is the coefficient of a / ( 2 s ) ! in the expression J . n + l ) ( n + 1. the coefficient of T 2s »{sr u)\ As we observed.e. n + l ) ( n + 1.u)\ 2s<—" v\(n .104 CHAPTER 6. n + 2){n + 2. n + 2) + (n.79). 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.A o?(. i. by n!(7r/2) 1 / 2 . n + 2)]ife„ + 2 (t) + ••• .^ ) ! ( n + 2 r .° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . n + l ) ( n + 1. n + 4)_ffe n +4(t) + [(ra.
which is a point of unstable equilibrium for a classical particle. These are the Green's functions of the timeindependent Schrodinger equation or. we consider the sojourn time of a quantum mechanical particle at a point.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent.x').x'. 7.t). which is normally not discussed. With reference to this case. The first of these cases will reappear later. In particular we derive the timedependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. in Chapter 21 (with a different calculation). which we write G{x. and those which are timeindependent.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. and is thus an important point to be noted here. which we write here K(x. We consider first timeindependent Green's functions. This example also offers a convenient context to introduce the inverted oscillator potential. We 105 .1) be the timeindependent Schrodinger equation which has to be solved. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. more generally. in Feynman's path integral as the kernel or free particle propagator.Chapter 7 Green's Functions 7. Let H^ = E^> (7. of a differential equation of second order.
2) where Ho consists.EW)xG{x.g. since the generalization to higher dimensions is selfevident.x') is defined by the equation* (Ho .4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).13).e.106 assume a Hamiltonian of the form CHAPTER 7.x') or ^2^(x)^*(x') i = S(x .g.x') by recalling the completeness relation.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . cf.x'). Eq. The boundary conditions which G(x. e.e. (7. a perturbation part like (3q4. (7. Green's Functions H = H0 + Hi.x'). We can obtain an expression for G(x. aq2.g. for instance.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. 'The case of £(°) equal to some Ei is considered later. The timeindependent Green's function G(x. the relation X') = 5{x . (7. and Hi is some other contribution. . where £ ( 0 ) = lim E. e. square integrability. x') has to satisfy correspond to those of * (to insure e. (7. although the explicit calculations can be much more involved. i.x') can be written t G (7. i. of the kinetic part p2/2mo and some part of the potential. appropriately decreasing behaviour at infinity). We can readily see that G(x.
We can see the significance of the timeindependent Green's function for the complete problem as follows.) = f dx'G(x. (JL x'\ t) = H0K(x.2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei.EM)X*(X) ( .x'.x'.Hi{x'))^(x').7) with the initial condition K(x.7.E<®)xG(x. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .10) The contribution on the right hand side is the socalled perturbation contribution. We see that the initial condition (7.t) since ih^K(x.t). x'){E . i (7. .x'){E .t) (K for "kernel") is defined as solution of the equation ih—K(x. x'\ t) (7.0) = S(xx').E^ We can check this by considering (H0 .9) = i H0YieE^i%(x)^(x') = H0K(x.x'.x. (7.E^ of the eigenvalue. (7. (7.(*'))*fr') Hi{x'))^(x') f dx'5{x .t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').x'.. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).x')(E .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.£(0) .#. .E® (EE{0) Hi{x))m{x).11) = = 3) = fdx'(H0 . The timedependent Green's function K(x.
see B. Booss and D. But then the Green's function (7. Wiedemann [183]. MiillerKirsten and A. H. All such methods of solution are based on the analogy of Eq. i. however. (7. The theorem is also known as Fredholm alternative. (7.11) with a system of linear equations of the form Vi = Mijyj.12) Equations of the form of Eq. 17. Merzbacher [194]. that (E . then at best such a contribution would be something like cipi(x).x'. Bleecker [33].11) are called (homogeneous or inhomogeneous) Fredholm integral equations. W. Sec. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. see E.6) is not defined. In general.2 (not contained in the first edition!). integral equations are more difficult to solve than the corresponding differential equations. (7. In cases where an inhomogeneous contribution is given. if we assume that E^ = Ej. But if we assume that E^ ^ Ei for all i.. This will be different. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution.15) ''This problem and its circumvention can be formulated as a theorem.6) we would have (7. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x').108 CHAPTER 7.14) We can now see the significance of the timedependent Green's function K(x.11) an inhomogeneous term which is a solution of the nonperturbed part of the Schrodinger equation? If we keep in mind the anharmonic oscillator with square integrable wave functions ^(x). Instead of Eq.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. (7. (7.e. Examples in various contexts have been investigated by L. 2nd ed. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). Green's Functions Can we add on the right hand side of Eq.t) for the complete problem as follows. J. Maharana. . In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. D.
t > 0. We use this in Sec. 1 See e. p. We can write this expression also as§ ip(x.t) = J2 fdx/^n(x)^*n(x')^(x'. H\En) = En\En).0). .Q) = 6(xx').18) or. (7.t) = ^ e £ " f / « V n W < ( ^ ) .g.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. this relation provides the probability density \ip(x. (7. x'. t)\2.' which obviously satisfies the initial condition K(x. 2nd ed. 7.21) dx'K(x..2 for the computation of the sojourn time.158. x'. (7.15) which contains (7. for the oscillator potential.20) is the timedependent Green's function. e. i. (7.17) as initial condition is obviously W)t = Y.42)) Wi=0 = E l ^ ) ^ l ^ = o n (717) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o. (7. 109 (7.The solution of Eq. in different formulations.x'.20) the Green's function K(x. *>0. Merzbacher [194].t) is known. (7.17) is the initial condition of \4>)t. we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.g.x'.x'.0)eEnt/in. (6. t) with the Note that when K(x.19) i/>(x. Eq.7) with Eq. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. (7. E.22) According to Eq.t) = where if(x^'. \En)(En\ip)t=oeE^\ n (7.16) As usual.e.5.t) describes the evolution of the wave function from its initial value ^(a^O). Comparison of Eq.t)il>(x'.7. (5.
(7. 7.ie (7 .t) K(x. Inserting for GE+ie the expression above. Green's Functions difference that (5 = 1/kT plays the role of it. we obtain then also the corresponding density matrix.x'. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. x')6(t) (7.110 CHAPTER 7.1.x') and K(x.E . x'.«^«2 W .i /f e «/.e. Jc 27T „ hn . t) = J2 e £ " f M i ( x ) < ( x ' ) . In the following we shall derive the Green's function for the case of the harmonic oscillator. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. we obtain / W : = .x') = GE(x.23) along the contour C in the plane of complex E as shown in Fig. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.1 The contour of integration. t > 0. 24 )  ReE Fig.x'. between (with E^0' = E) G(x.t).25) in agreement with the timedependent Green's function . i. 7.
It is clear that it is nontrivial to solve an equation like Eq.x'.7). B m0 n _ 2ih 2iK (7. which is moving in one space dimension.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7. (7. (7.27). (7. 2mQK h h2 ih(AB) = —{4B2A). In this case the Green's function is the solution of the equation d hK(x. i. (7.3 The Green's Function of a Free Particle The timedependent Green's function of a free particle which we now derive is an important quantity and will reappear later in Feynman's path integral method. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.x'.0) = 1.31) The constant A has to be chosen such that K(x.28) B(xx')2 t2 DB(xx') 2 /t (7.7. we obtain *I£I = h2 —(2AB).26) This is the case of a free particle with mass mo. Thus we try the ansatz.X>. K(x.t).27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz. A and B being constants.e. / dxK(x. and identifying coefficients of the same powers of t on both sides.29) and (7.— d2 —2K(x.3 The Green's Function of a Free Particle 111 7.30) Inserting Eqs.e.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.29) tJJ 2 2B(x .0) = 5(xx').t) t h2 = . v ' 2m 0 v h i.x'. .x>.30) into Eq.
0)oce"Qx2+ifcox.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. t) = /jJo_ e mo(««') 2 /««. Green's Functions For parameter values such that the following integral exists.H We can insert the expression for K(x. (7.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. Merzbacher [194].(7. (7.112 CHAPTER 7..t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. (7. *>0. we > have to make the replacements V ^ fdk. 1st ed.e.38) . (7. Mx)^M^) „ n so that J =^ . p. 32) It follows that x'.x'. 158.zEnt. v27r En^^."737) \27rii/i in agreement with Eq. n For a free particle moving in the onedimensional domain \x\ < L — oo.20) and can then obtain ip(x.t) into Eq.t) = — / dkeak2+Pk 2vr J.21). from K{x. i.34) K(x. we have This is 1 provided A=JB= ^ (7 .35a) K(xy.ihMxWn{x'). "See the excercise in E.x't) = Y.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS .oo 27T (7. (7. (7. 2m0 (7.x'.t) — for instance for a wave packet given at time t = 0 of the form ^(x. j3 = i(xx').33).
i. . mQLu .7) is the solution of d ih—K(x. We rewrite this initial condition in terms of £ and use for a = const. I (7.X .40) Then Eq.33).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.40) becomes 2 . the relation* 6(x) = o<y(os). —ih—K(x.t) + ~x K(x. i.t).t).——K(x.. (7.7. 5{ax) = ± J eikaxdk = ArS(x). (7.e. (7. —KK(X. §jK(x.d r.25) — in the context of Feynman's path integral method.42) .37) will later be obtained by a different method — see Eq. z . .t) 1 + m0u2x2K(x. 7.e. . .. in this domain *Recall that S(x) = ± f eikxdk. x'.43) so that In the domain of small values of x (near the minimum of the potential) the Hamiltonian HQ is dominated by the kinetic energy.t) We now set h2 d2 = . .x'.x'.e.39) In this case the timedependent Green's function K of Eq. x'.x'.x'. x'./ ) + fK(x. K(x.3 The Green's Function of the Harmonic Oscillator 113 The result (7.8).t) = h2 2 d2 TW . . 1 2mo u>n ox n with the initial condition (7. (7.x usn ot i.. (7.x'. HQ = ^—p2 ZrriQ + \m0uj2q2. i. / ) = ~^K(x. / ) (7. (21.0) = 6(xx') at / = 0. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.e.
p. 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. that is. and we expect K(x.b2. (7.47) We insert this ansatz for K into Eq.e. 50.47) becomes (7.4a&£ + 2a . P.b2. a = .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7. i.50a) we obtain (7. (7. it • is suggestive to attempt for K the following ansatz^ m. we must have /o = 0. . so that for / — 0: > a^—. we obttain a' = l .45) in this sense.33). as the limiting case / — 0. (/ = 2a. Green's Functions to the particle behaves almost like a free particle.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.50c) (7. Identifying coefficients on both sides.) a'£2 + b'i + c' = (1 . (7.51) See R.4a 2 )£ 2 .42) and obtain the equation (with a1 = da/df etc.4 a 2 . c(0)^.114 CHAPTER 7.f) with a(0)»y.^ .Z'.48).50a) (7. (7.46) If we interpret Eq.45) in accordance with (7. W ) . Integrating Eq. Feynman [94]. f i. (7.c o t h 2 / .50b) (7..x'.e. (7. b' = Aab.t) become similar to expression (7.
with A. (7.50b) Kf) = —r^77) smh 2 / A independent of / .2. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. 7.* With this result we have another important quantity at our disposal. (7.45). if we return to x. Finally Eq.55) 'For an alternative derivation and further discussion see also B.i) (7.5.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. (7. we must have A = ('. K(x. as we shall see in the following. (7.50c) yields for c(/). (7./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. c(0) = £' / 4 / . we must have (besides A = —£') = /m 0 u. B independent of / .48) 6(0) = — £'/2f.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . p.x'.b(f). t. x'. Felsager [91].33) for the Green's > function of a free particle. and to ensure that we obtain the prefactor of Eq.e. x .27).c(f) K = 5 =exp Vsinh 2 / into Eq.t) = .48).7.40) of the timedependent Green's function with Eq. 174. we can use K(x.In B. (7. (7. In order to satisfy Eq.51) for the density matrix PN(X.X'.e. i. . (5.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7.t) to obtain this element (x. i. / —————rexp y 27rmsin(u. as one can verify. 115 To ensure that in accordance with Eq.54) For t —• 0 this expression goes over into the expression (7.53) or.x'. in particular for the derivation of the sojourn time in Sec. 2TT^ Inserting a(f). (7. Comparing the Eqs.0).
52): Tr (PNq2) Jx2pN(x.x. 11 For T ~+ 0: u>0 .56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.56a) the expression (7.52)) the expectation value of an observable A in the canonical distribution (i. we skip the algebra here..56a) {q2) = = Y.» l. 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"}. P. (7.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.e. Feynman [94]. [ [ 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. = jdxp(x. )x Inserting into Eq. .uij>0 >• 0. For instance we have with Eq. (5.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7.p)dx W ) = —^r = —p . p. (5. <2 (7. Eq.e. (5. Eq. ^—• Irpiv J pN(x.40)) 1 epEi 1 Ui = y.116 CHAPTER 7.x. we obtain § < Z> = ^ c o t h ^ ^ ° — .P) 2.. 52. Thus the system is in a mixed state and the expectation value "Cf. eP E i = i + y. Green's Functions With this expression we can evaluate (cf. at temperature T): {A) = Tl{pA) = ^ > .55).x. R.
x'.54) is K(x. We assume t > 0 and t —> oo. 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 (7.50) we obtain. of K(x.59b) n+ l \ ftwt here 2hf exp _ i ( n + .21).t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . (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.54). 1 mo^ exp lmow / 2 2 ft X (7.7. We return to the Green's function (7.QUJ (7. t) of Eq. If we consider the system in the pure state \i).3 The Green's Function of the Harmonic Oscillator 117 (7. x'.x'.56b) is that with respect to this mixed state (whose cause is the finite temperature T).J (w — ie)< e (n+l/2)Et e i(n+l/2)u. which means in the oscillator state \i) with eigenenergy fku(i + 1/2). (6. e > 0.59a) E 0 = ^fiw. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .e. This is the first and hence dominant term of the expression (7. 2 + x'2)le^2xx' to—>w—it t—>co 2 . /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. In this case the Green's function K(x. i.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.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. and we replace to by LO — ie.
(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7.5. However quantum mechanically in view of the uncertainties in position and momentum. lie on surfaces.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. In a very analogous manner we obtain the solution of the equation for the density matrix. which are planes.. In Example 7.59a). the particle will stay there only for a finite length of time T. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. 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. (7. is defined as vv = .1 we estimate T semiclassically. (7. the velocity of planes of equal phase.x'.118 CHAPTER 7.e.e. Green's Functions For t large (i. 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. to infinity) the contribution with n — 0 dominates.62) . as in Eq.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)]. Considered classically. i. The wavelength A is given by The phase velocity v<p. k = k. and with this we estimate the sojourn time T. of dpN as pN(x. (7. 7.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. i.e.
i ) = /*/(k')e i(k '" r '' . oo (7.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf.64) where / ( k ' ) differs substantially from zero only near k' = k. to the function ip given by oo / fik'y^'x^dk'. xt— + — = 0.<p(k"))}dk'dk".7. essentially again a Gauss curve. If we assume for / ( k ' ) a Gauss distribution. i. a > 0 .63) The relation u = w(k) is known as dispersion or dispersion law. e. u = w(k). (7.)¥'(*:")) +ism(<p(k') .)ll/(fc")l[cos(¥'(fe. (7.k ') 2 . t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.e. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. the fundamental postulate on matter waves in Chapter 2): E = hu. t) is the spatial Fourier transform of this Gauss distribution — as we know. e .e.65) Let f(k') = \f(k')\eia and <p := k'x . We now ask: How and under what conditions does the time variation of the function ^(r. The wave packet describes a wave of limited extent.e. ' t) dk'.Q ( k . / oo J—oo . ^ ( r . duo' da . One defines as centre of mass of the wave packet that value of x for which d(p — = 0. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. i. then ^(r. for which \ip\ assumes its largest value: OO . i.Jt + a. or dco' da The centre of mass determines the particular phase.
i. (7. p = hk the group velocity is equal to the particle velocity v.^ g=r a rad. we can use a simpler method. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. we obtain the deBroglie relation p — hk.fc w> vv = ~a^ = S r a d P E «i7 Sd u .120 CHAPTER 7. . For E = hio. a > 0.. = — (768) It is instructive to consider at this point the following examples. dE v 9 = . Example 7.— g(k). as claimed for =0.e. however. In the present case. (7. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. The threedimensional generalization is evidently du . . 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 = .67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)].69) In general one uses the theory of functions for the evaluation of this integral. Solution: The function to be calculated is the integral dxeax OO ezkx.
Example 7. (7. g . where c = 9(0) is a constant..2: Representations of the delta distribution Use Eqs. or see H. with partial fraction decomposition. (7. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. (7.75) ' E . MiillerKirsten [215].70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^.1 verify the uncertainty relation Aa.&)] for a + b..71) (7.74) Example 7..43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example. XZ (e > 0). (7.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a . W.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .69) and (7. (7. (7. J.72) Solution: From Eqs.70) we obtain 6(h) = lim — f°° dxeax2eikx = .Afc = 81n2. Appendix A..a) + 6{x .. \a — b\ (7.r „• With the help of Eq. .69) and (7.0 0 7T 7 0 7T e From Eq. 2 f + . —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration.7. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . (7.73) 27T 7 .
i. (7. Thus a slim maximum of the curve of /(a. According to Eq. 7. and hence implies large values of Ak. Thus a sharp maximum of the function f{x).2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. Green's Functions 1/(27i)1/2a — 1/2(2.77a) g(k) = e . where g(k) = max<?(fc)/2. Solution: In Fig. 7.) leads to a very broad maximum of the Fourier transform \g(k)\.t) is a measure of the probability to find the particle at time t at the position x. at time t = 0). requires according to Eq.e.r)1/2a Re f(x) Ax Fig. The breadth A/c of the curve g(k) around k = kg.77a) small values of a. i. the width Ax is a measure of the uncertainty of the probability.70) the Fourier transform of f(x) is (7.* o ) 2 / 2 .77b) (7.« 2 ( * . In quantum mechanics the square of the modulus of the wave function ip(x. when Ax is very small.78) It follows that the product of the uncertainties is AxAk = 8 In 2.g. The wave function corresponds to the function f(x) in the above considerations (e.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. is Afe = 2 V 2 1 n 2  (7. Physically .2 The Gaussian curve.CHAPTER 7. (7.e. where  / ( x )  = m a x  / ( x )  / 2 . A simple calculation yields A i = 2v/2~In2a.
<7*» We assume that initially. The function g(k) is therefore described as momentum space representation of the wave function f(x). (7.«> = / ^ C ' W . In the same limit the momentum uncertainty Afc grows beyond all bounds. For reasons > of simplicity in the following we set in addition t mo = l. so that Ki0(x.71) > and (7.79) According to Eq. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.7. The result (7.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. Thus G.81) *G. at time t = 0.75) implies therefore: The more precise the coordinate of a microscopic particle is determined. (7. (7.5. Barton recalls that W. O ) .. (7.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x.x'. This means the particle is localized at x = 0. ^In Eq.e. Lamb at Oxford set the following problem in an examination in 1957: "A pencil is to be balanced so as to stand upright on its point on a horizontal surface.2 ) and U)t. nr = S(x). 7. As G. 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..h = 1. Barton mentions at the beginning of his paper.".76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «.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.54) with the substitution ui — iu. Barton [15].u> = l. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. E. (7. In the limit a — 0 we obtain from Eqs. the less precise is the determination of its associated momentum. . i.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . which means all values of the momentum are equally probable.
Green's Functions Inserting (7.cosh t sinh t' Wisll2b ^ (7.2 (7.79) into (7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.124 CHAPTER 7.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.84) we obtain* ip(x.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.t) = J oo P .86) Evaluating the Gaussian integral of Eq.81) and (7. (7...(Ax' + Bxf \/27risinh + (B2 + ^cothi)a. = . . so t h a t (this defining C(t) and the imaginary part ip on the right) {B2 + icotht) sinh 2 1 + (i/26 2 ) sinh 2t ( l / 6 ) s i n h 1 . A . 9 (7. + cosh 2 1.82) i/f(x.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh . (7.85) = —r= o2 7 2 sinh 2i.83) ip(x.88) .t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7. ' 2\Aswht % cosh t (7.
89) = exp[x2/C2(t)} 7rV2C(t) (7.91) C(t)' (7. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.t)\2 Ji == 71 / V " Jo dO ? 6' (7.4 The Inverted Harmonic Oscillator The real part of Eq. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.t)\' 2 125 (7.7. (7.92) The sojourn time T is now defined as the mean time T given by T:= r*° ioo 2 (M)s r*(f4 d^e.. •&(t)=//c(t) = / or.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 . _2 1 exp C 2 (i) = l + 2sinh 2 i.90) with the expected limiting behaviour \im \ilj(x.t)= where Co(t) / dx\^(x. C2(t) (7. t) the phase ip drops out and one has (x.£ 2 (7.94) / dxf(x)=g'(y)f(g(y)).95) .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). We have therefore . In taking the modulus of ip(x.
+ 0(e'2) (7.From this we obtain C(t) C'(t) Z2 ^(r?2_.99) . B. 136.96) °° l Jr.(4 2 .126 We set I C(t)' so that dT ~d~l Now.2)(p_7?2)(7. we obtain dT _ 1 ~dJ~l 'See for instance H. Dwight [81].97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.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 /2^7.98) We insert this into Eq. p.=0 V^ C'(t) I Jt=0 V^ (7.= \ / l + 2sinh 2 t. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.. C(t) = ..97) sinh 2 1 = . (7.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.1 ) 2 \n J ^"•""ff^^V^K?. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.
It follows t h a t T~lnZ. 238. also motivated by this paper. U) (7. 7. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above.4. Fig. The equation there describes the classical "rolling down" of this tachyon. T h e following Example 7. this equation is the equation of motion of this scalar field with negative masssquared.7. Zwiebach [294]. this is T ~ ]n(ly/mou>/h). Explain the parameters entering the calculation.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7.g. 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. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. p. (7.102) 'As a matter of interest we add that in the theory of a free scalar field x. See e. B. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon".
/ 7 ^ 2 =2J(x }. CHAPTER 7. so that e. (7. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*).5)) Ax = <J(x2){x)2. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p).g. i. ^(x2)(p2). but instead are subject to an uncertainty relation of the form AxAp > h. For h —> 0 the time T —• oo in agreement with our classical expectation. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. Eq. Thus quantum mechanically x(0). and for simplicity we take (cf. Quantum mechanically x(0).e. the smallest macroscopic length. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. A p are defined by the mean square deviations given by (see Eq. x(0) = A. Green's Functions x = Aco sinh cot + Bco cosh cot. (x 2 ) h 2mou 2mou> and so l^J^ULl). and solution x = A cosh Lot + B sinh cot. v A x A p = \ (x2)(p2) v = m0u(x2). 2 x(0) ^ where A i . 2 10 x(0) = B u . so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). i. Ap= v /(p 2 )<p)2. (7. . pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.103) to VV h Here I is a largely arbitrary but macroscopic length like the length of the power of resolution of a microscope — so to speak. 2V ' Let A and Bio be the values of x and x at time t = 0.e.p(0) cannot be determined with arbitrary precision. (p2):=mW(x2). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I.128 with x2 = LO2X2 + const. (10. 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 .102)) p(0) = ^/(p 2 >.
in closed form) only for very few potentials.E^ is already a bigg problem. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. In general perturbation series do not converge. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. Frequently even the calculation of the next to leading contributions ip^. 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.Chapter 8 TimeIndependent Perturbation Theory 8. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. i. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.e. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). For mathematical purists the question of convergence of the series is an even bigger challenge.(i)+/^(2) + ".2h2 cos 2 a # = 0. one would set tt E = = ^(°)+/ty.1 Introductory Remarks The Schrodinger equation can be solved exactly (i.e. 129 .
concerning also the uniform convergence of the exponential series. see e. .• + (ir~ X X* 1' 2' r?l Xn + Rn(x). 0. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified.= 0 < 1. W. E = E^ + h2E^+hAE^ 71 72 7 + . ^For further details. TimeIndependent Perturbation Theory In the case of this equation. In the following we explain the difference between convergent and asymptotic series.. p. Watson [283]. J.2) n—»oo n It is interesting to compare the behaviour of the series (8. e. h hl are in the strict mathematical sense divergent. It will be seen later (e. E. Meixner and F . explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. . Whittaker and G.2 However.g.g.. (8.g. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . T.. N.g.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 .e. (8. Schafke [193]..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. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. the expansions in descending powers of h2.130 CHAPTER 8. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. 8. which is known as the Mathieu equation.* i.3) *See e. at the end of Chapter 26) — and by comparing the results of WKB. 581. This convergence can be shown with D'Alembert's ratio test since''' lim . 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. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •.+ .
. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . / ( I ) = 1 — 1! + 2! — 3! H • However.0..0. E. the better the approximation obtained. . and we see that the moduli of successive terms first decrease and then..000 002 00 . We can see this as follows.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. T. 235.. .3) diverges for every value of x. We observe in the first place that since lim n—>oo co (8. + •••.e. for x = 3. the series (8..2.0. .5) Cf. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. Comparing the asymptotic series (8.22222 .4) for every value of x. . Thus e.g..33333 .000 000 06 + • • • .9876.376 .3) with the convergent series (8. /(1000) = 1 .g..22222 . However.4938. +0. e. 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\. N.l n ( l + ^ ) n=l L ^ ' (8.0. + 0.222. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. after reaching 6/27 = 0.. p. for larger values of x... i. and in fact increase indefinitely. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ . Watson [283].. + 0. begin to increase again.0.29629.8. Whittaker and G.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. .001 + 0.. the partial sum of terms up to and including the least term yields a reasonably precise value of the function at that point with an error of the order of the first term of the remainder.. in spite of this the series (8. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20.1). 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.
Schiff [243].41) and (16. TimeIndependent Perturbation Theory where 7 = 0. ." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. in fact. (16. Merzbacher [194]. In many respects Messiah aims at more rigour in his arguments. ip) can be represented by rapidly converging power series in e. T. it is sensible to make the assumption. 152.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. E. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. 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.1 between Eqs. This is not appreciated by mathematical purists. p. In fact. at a later point in his treatise Messiah admits that the expansions are mostly § Cf.1.5) are practically unknown.132 CHAPTER 8. I. p. It is interesting to observe the comments of various authors on this point. meaning convergent. It is inherent in the nature of an approximation in a physical problem that in deciding between dominant or primary effects and those of secondary importance. although this has not been investigated except for a few simple problem^. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. the leading term of an asymptotic expansion. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. 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.e. 11 L. ^E.5). that E and \ijj) (i. Messiah [195].J In2. § However." These "rapidly converging power series" in physical contexts are most likely very rare cases. N. Sec. 236. This latter observation hints already at the importance of asymptotic series in applications. 16. Watson [283]. whereas applications of the convergent series (8. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. p. Whittaker and G. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). **A. 371. the dominant approximation is. Vol.
Vol. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.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.. solutions of second order differential equations. (8. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion .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. In fact.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.7) as / e'* dte* dt = 1 . *R. (8.* and we follow some of t h e considerations given there. Merzbacher.10) Vn Changing t h e variable of integration to u = t2x2." Of these three authors t h e first. 198 (German edition). seems to be closest t o t h e t r u t h a n d does not attempt t o give t h e impression t h a t t h e series has t o converge.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.8. 2 (8.= / e~t2dt. 8.7) This function has been considered in detail in t h e book of Dingle. Dingle [70].2.. (8. (8. p. we can rewrite Eq.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . B.. It is therefore suggestive to write iy poo <t>(x) = 1 .11) A . which have been studied in great detail.^=e~x2 / e^^dt. II. there is no need for such bias. ft Jx du = 2tdt. He says. We know from books on Special Functions t h a t all of these functions. (8. . Messiah [195]. Replacing t h e exponential in Eq.
(8.13) presupposes t h a t < 1. 7T (ni)!sin{7r(n+^)} ( „ .16) . Then e (n —\)\( n\ 2du u \n Xy/n o du VK n = 0 e 2 + XypK ~"(1+ ^ °° /•oo —u (ni)!/ n! (nn\ 2)! u du+ Jx : e u du 1 ^ ( 1 + n=0 oo /_u_ V x2 ^ ) K ^ n=0 (8. since (—4)! = y/n.12).12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.e.14) sin7T2.134 we obtain CHAPTER 8. TimeIndependent Perturbation Theory <P(x) = 1 . we obtain 1+ i.15) < 1 we can insert this expansion into Eq. « \~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. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .' we have 2 nH 7T 7T (8. Using the reflection formula (*!)!(*)! = for z = n + i .^e~x \Ar i.e.i)[(_!)«' Then. (8. du.^ X e U ~ ( 1 + * \ 2 .
In the foregoing discussion of the error function we assumed that x was real and positive.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq.15) into (8. argx = 0.8.e. (8. As in the case of the gamma function. Whichever way we look at the result. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned.12) ignoring the fact that the latter is valid (i. (8.2 Comparison of Asymptotic Series with Convergent Series 135 We can evaluate the first integral with the help of the integral representation of the gamma function. . and in fact so slowly.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. i. It is fairly clear that the above considerations remain unaffected for argx < 7T. Frequently "=" is written instead of "~". In an asymptotic expansion the first few terms successively decrease in magnitude.l)! = / Jo Then 2 cHnxdt. 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.e. much more useful than the convergent expansion.18) or that we insert the binomial expansion (8.—yS!L^LLZ e x OO / _ 1\ X2 /CO dueu L \ / V n = Q 1 and we write 0(X)~1 V 7 n=0 v 2^. (8. the asymptotic expansion of the error function is. We now see how the asymptotic expansion (8. in general. We can actually understand the deeper reason for this in practice.19) approximates <f){x) the better the larger x is.19) originates. The expansion (8. i.r<7r.17) 0(X) = i. (8. convergent) only in the restricted domain u < x2. We now want to relax this condition and allow for phases argx ^ 0. /•oo T(n) = (n .e. ' arg. whereas in a convergent expansion the terms first increase in magnitude.
20) where we set t = is.^ f V e ^ . thus with \v/y2\ < 1. We therefore consider this case in detail.22) we obtain 2 4>{iy) = . =dv. (8.2 y V . The situation becomes critical.23) Throughout the range of integration 0 < v < y2 the integrand is real. when argx = 7r/2.26) or else return . TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.21) = y2s2.^ = . however. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds. dv = 2sds = . i.vds. we can try to proceed with Eq. (8. (8.e. V7T Jo Changing the variable of integration to v (8.15). (8. VK Jo V^ Jo (8. (8. we set x = iy in Eq. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . (8.136 CHAPTER 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).24) We can read off the binomial expansion of the factor in the integrand from Eq.
30) yvn f Jo V y J Cf. which will be studied in more detail below.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! .^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv.29) where 5t means "real parf. (8. If we proceed with Eq.17): J£° e~vvndv = n\. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. (8. and the larger y.24). We know that for v < y2 the integrand is real. (8.^ ^ p . However. y y 2 ( n _i)! y arg(* = iy) = i * . i. e ^ ( i y ) ^ ^ . we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . 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. the better the approximation expressed by Eq. We have therefore <j>(iy) = ^e^TL fV e~v ( l .27) Then .28) we observe that the expansion (8.e. — (8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . Suppose now we consider Eq.28) n=0 We observe that this expansion differs significantly from (8..e. (8. (8.27). i.27) and (8.24). (8.28) ignores the integral contribution of Eq.^] * dv. i (8. (8.28) because then the correction term oc f°£ becomes smaller and smaller.19). Returning to Eqs. for large values of y. and so the integral is real. ey is much larger than something of order 1/y. .2 Comparison of Asymptotic Series with Convergent Series to Eq. (8.2 yir (8. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.8. Eq.
The corresponding phase lines are called Stokes rays.28). Since (cf.33)) would have • a maximally increasing exponential e+x .31) and (8. Looking at Eqs. i. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.31). y. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2.3!) which is (8.2 OO rz=0 V . This result is identical with that obtained previously.138 CHAPTER 8. (8.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. 7r = ." — E ^ ^ forarg.26). and • higher order terms of the associated series all have the same sign..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 observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. Eq. R. . (8. (8. (8. *Cf.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. is obtained for <f>(x) with arg x = —ir/2. since (8. We have therefore found that XTT n=0 ^ ' and . With a similar type of reasoning one can show that the same expansion. B. (8. . e xZ ^ (n — £)! „ ^ ) . Dingle [70].e.e.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i.25) and integrate from 0 to oo. (8. Chapter 1. TimeIndependent Perturbation Theory If we insert here the expansion (8.
8. —1 cancel out.34) the contributions 1. Eq. Oberhettinger [181]. apply predominantly to asymptotic series of the solutions of second order differential equations.37) Cf. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). 9 . Equation (8.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. and so may not be universally applicable. 5. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o .2). . '"Cf. in the sum of the right hand sides of Eqs.35) 8. Magnus and F. for arg x = 7r/2.36) 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. p. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a.2. 1/1 ipv{x) v1 vl (±ix). W. (8. (6.g. though.13.32) and (8. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. These rules. Dingle [70].e. i. 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 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. R. e. 91. See also Tables of Special Functions. i. B. Thus if we set x = XR + ixj.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.e.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. Chapter 1. pp.
B. i.38) are of the type (2i)! i\ i\(2x Y 2 (2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series.e.7. Chapter 17. 8. v — — f — 1.40) § Cf. (14). (8. The continuation of (8. TimeIndependent Perturbation Theory i.37).36) is invariant under these replacements. x~l/~1ipu(x). Eq. We also observe that the late (large i) terms of the series in (8. Chapter 1.ei<v+VxvHv{x\ argx = £ . with a real proportionality factor. (8. R. Thus D{J\x) (i.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. 0 < argz <  (8. 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. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. B. Dingle [70]. Chapters I (in particular p. > > The equation (8. Hence Dil){x) = xv(j>v(x). ti asymptotic series (8. ' S u c h symmetries are extensively exploited in the perturbation method of Sec.e.e. = 7r/2. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. R. We consider Dv (x). .x 2 oo «*)  — E^fi=0 v . 9) and XXI. Dingle [70].140 where^ CHAPTER 8. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. as can most easily be seen in the case of the Mathieu potential. 1 1 Cf.39) (as in the example of the error function). i.
i. the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first).41) the Stokes multiplier a is. (8. (8.42) T ^(eiwu i. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. On reaching ir the part containing ^ ( x ) . Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2.e.41) xuct)v{x) + a{x)~u~lipu{x). with \x\ = (—x) for arg a. i. sm(iri/) = —a/3/2 or a/? = 2sm(irv). is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. as yet an unknown real constant.44) + ^iap) = 0 .42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).e. (8. < 7T. Since Dv (x) is real when x is real. In (8. another half of the discontinuity appears on leaving the Stokes ray on the other side.41) to arg a. < 37r/2.8. beyond arg x — 7r/2. we must have at arg x = 7 in the dominant factor on the right of Eq. It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real. = ir. where in the first line on the right hand side the second term contains the real proportionality constant a/2.e. for xj = 0) and therefore possesses a Stokes discontinuity there.40). The value of Du (x) on the other side of the Stokes ray. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. (8. We therefore proceed to continue (8.e. = ir.
(z)\(ziy. (3{v .47) Equation (8.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. .e.42) that a multiplies ipv(x).1) = 2sin7ri/.e. TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. i.14). a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .v . (8.e. Dl \X) is (cf.45) is therefore given by a{v)a(v . argx = vr.142 CHAPTER 8.42) we insert from Eq. i. (8. i. (8.1) (8. (8.—^ ON • — ^ > + ^ ^ L .38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i . V i=0 \U0 v 2w .46) For a given function a(y) this equation determines /3(u).45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.51) .l)0(v .1)! TT.42)) for argx = ir (in the second line of Eq. i. 8.1) (8. a{y)P{y) = a(u .45) that af3 = 2sin(7r^)) D^\x) = (8.1). We can see that the equation is satisfied by (3{u) = ±a(u .48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .e. (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq.48) We compare this with the reflection formula (8.1) = ±a(y). (8. Eq. (8. Hence the left hand side must have the same property.
1 ) ! 4~L i\{2x2Y i=0 V2^(x).1. ' ^ (i/l)!z! ^ ..7). —f. i.7r)..e. Chapter I. solutions of Eq.e. the second contribution vanishes and we obtain £>«(*) . (2x2)1 = COSTTU(X) v ..8.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. For details we refer again to the book of Dingle. i.„ _ix2^(2ivl)l e 2 > —.e. It should be noted that in Eq. i.52) (argx = 7r). (8. (8. ' (X) (cf.I / .2.)(. (8. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator..35). the error .*"ei°> f ( ^ _  _ L _ = (!)««(.50a) because this enables us to define normalizable parabolic cylinder functions.1 ) ( . (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di. u 1 y> {2i + v)\ 1 n 0.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. which vanish at x = ±oo. We observe that for v — n = 0.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator.) (8. (8. (8.A / ± sin^Cx)"M* 2 V ^ .2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .36). Eq. 8. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.36)) in the domain 0 < arg x < IT.. However.e^ ..
60) into (8. Instead of dealing with Eq (8. S2. and demonstrate. i.e..38) can be obtained. ^ + 2 .56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8.^ + 1 ) .59) and equating to zero the coefficients of powers of x2. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. x.e.54) we consider again the important equation of the harmonic oscillator. how series of the type (8. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 .36)) 0 where + x2Q(x)y = 0. (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.36).54) (8. we obtain equations from which the coefficients So. Eq. (8. % = {. \ + U (8. 5„=4 Sl=o. Solving these equations we obtain = = 0.36) in the form (coefficient of — x2 here chosen to be ~. Eq. i. previously we had ^.37). x—>±oo (8..60) 2 x x l Inserting (8. (8. (8. We write Eq. + — + ^ + • • • . 0.55) The large x asymptotic expansion can also be obtained from the differential equation. • • • can be determined.. cf. x°.144 CHAPTER 8. S\. (8.
i. (8.38).91.z) is a confluent hypergeometric T. 2 X 2Wi.35) and (8. The derivation of the asymptotic series from integral representations (cf. etc.37) with (8. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). n 22e"T JziL (8. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ . \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  . function.b. 4~l~2 ' i i _i. » 1. Nonetheless we shall need asymptotic expansions like (8. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. Oberhettinger [181]. p.+ X M W = Q. T h e expansions (8.e.62) 4 \ 2 XiFi 2'2 .63) **W. and it is clear t h a t higher order terms follow accordingly. Magnus and F. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a.61) (the dominant terms of (8.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. are asymptotic series in a variable x. We. a z a a ( + 1) z2 which has the unit circle as its circle of convergence.37).8. therefore. In the following we are mostly concerned with asymptotic series in some parameter. / i N .19). do not enter here into their derivation and simply quote the result: W i t h \x\ . (8.36) have correspondingly e±x / 2 ) .38) in a variable for the solutions of approximated differential equations. .
Whittaker and G. Watson [283]. top of p.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially. 8.146 CHAPTER 8.1 a2 an a0 + — + ^ + . Thus there is no contradiction^ between (a) and (b). . 4 (8.65) We observe t h a t the series of (a) and validity 1 1 7r < a r g x < 4 In this domain 'Qx > 9ffcr and so e (b) have the common domain of 3 3 7r. then the series is said to be an asymptotic expansion of f(x) in t h a t range if for a fixed value of n lim \x\—>oo ft xn\f(x)Sn{x)\=0 3. N.0. TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2.+ — + ••• of a function f(x) for a given range of a r g x . T.67) Cf.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 . 349.64) (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 " ' (8.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. E. whereas e~x '4 increases exponentially.4x2 (8. 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.
(8. possesses a whole chapter on asymptotic expansions.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. first published in 1902. lim xneW x—>oo = 0 (8. . in particular with regard to possible exponentially small contributions to the expansion for which e.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (871) / dx%l)*m(x)lpn(x). (8.68) is (effectively) simply a statement of the observations made at the beginning.73) Vn + q/#> + e2T/i2) + • • • . which is assumed to be small.8. 8. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.e. In fact.g.Sn{x))  = oo n—>oo (8. A critical discussion of the definition can be found in the book by Dingle [70]. (8. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". i.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e.5 RayleighSchrodinger Perturbation Theory RayleighSchrodinger perturbation theory is the usual perturbation theory which one can describe as textbook perturbation theory as distinguished from other less common methods.68) The definition(8. The definition is due to Poincare (1866). we set E^En * > ^ n = En°1+eEnV = + e2EnV + . Chapter I. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.
(8. (879) We multiply this equation from the left by ip^ and integrate over x. The second equation can be rewritten as {Ho .V)i.n. i.148 CHAPTER 8.n = 40)V>n. TimeIndependent Perturbation Theory We insert these expansions into Eq. We obtain then the following set of equations: Hrj. °° i J or 0 = EWJdx{rnVil>n).76) The first equation is that of the unperturbed problem. .74) (8. ^ ) = 0. We insert the ansatz (8.e. Ho^+V^n = EW^+E^n. i.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.e.70) and equate on both sides the coefficients of the same powers of e.78) into Eq.i4 0) )V4 1} = {EM . (8. J f°° dxrn E(^ 0) .w » . (8. (8.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.e.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . 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. ( ^ . so that we choose / • dxrpM^O.75) HoW + vW = E^+E^+E^. 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.As a consequence the state vl/) is a vector in this space.4 0 ) M = / ^C(4 1} . With the help of the orthonormality condition (8. (8.
(8.e. i.e.83) makes sense only if Ef] ^ EnV for all i + n. we return to Eq. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . with En .i ^ 0.Viprt e (8.^(EW ~ V ) ^ + E& / J—oo dx^miln . the coefficients a\ .84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. (883) Thus the procedure forbids the equality of any E\ '. (8.76) which we rewrite as (H0 . (8.82) *> = ^ + E .Vil>n) (880) In order to obtain ipn .V)^ Setting + 42tyn.4 0 ) M 2 ) = ( 4 1 } . (8. multplying Eq. (il)i. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. We therefore consider Eq. (8. We multiply the equation by ipm.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.m ^ n and integrate: oo /•oo / oo dxrmY.79). It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.8. i. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. ^ n ) + 0(e 2 ). S ' ti^+Q^)• We observe that Eq.
V^n) .e.88) (Q) Inserting Eq. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.79) •/—oo oo / Using (8.^j)(V .150 CHAPTER 8.V'>pTl (E^Eny V'fc (8.^w .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.Vlpn)E.B i 0 ) ) ( £ f . Inserting the expression (8.87) (i>k. (8. ^ n ) j&iift + ^n .85) together with (8.£i°») 2 Hence +^£o (£«? .89) • • • where we guessed the term of 0(e 3 ).VV .4 0) ^+ e 2 E (lpk.Vlpn) E^ . V ^ i ) X (Q) (Vi.V4>n){lpn. ~Ej E E.^(8.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .2 (V> m . En0) (8.Vlpn] M _ P (0K 2 ^r . (8.(1) (i&0) . (8.^ ) + $& w £ (^fc.V^j)(^j. FVn) + e 2 ^ ( V > n .Vlpn)(lpn. j. n) £f) (8. ipn ) = 0 i.81) we obtain (2) _ _ {lt>m.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .
+ l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. i.g.n(x) = J2crtn+ii (8. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. The functions ipn(x) are eigenfunctions of some secondorder differential equation. (8.. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e..e.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.Vlpn) = y^Ci(lPk. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. Merzbacher[194]). 8.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.lpn+i) = ^ Cj5k. We then obtain (lpk. One would prefer to lump it together with the other contribution. An ugly feature of the expansion (8.8.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).89).90) is the term with only one summation J2 m the contribution of 0(e 2 ).7 is based on this procedure. .91) The coefficients Cj follow from the recurrence relations.90) can be found in just about any book on quantum mechanics. We can therefore rewrite tpn ': The perturbation theory described in Sec. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.
(8. (8.V)(c2l^n + ci 2 ^ TO ).97a) and (8. (8. . Equations (8. —I.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. so that En = 4 ° ) + eE™ + e2EP + ••• . TimeIndependent Perturbation Theory 8. Em = 4 ° ) +eE$ + e2E$ + . (8. i. .96) are effectively the known equations of ipnitpm and so yield no new information.ipm) = 0 = (^m.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . degeneracy. Em can now be linear combinations of ipni'ipm.e. + c22^m).70). (8.97b) with ipn y£ ipm. .97a) (8..93a).e. .95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8.e.4 0 ) ) ^ } = (Em] . In order to deal with the problem of degeneracy we consider the simplest case.93b) The lowest order eigenfunctions belonging to En.93b). Equality of eigenvalues. in a central force problem all have the same energy unless the system is placed in a magnetic field..97b) are .96) (Ho .4 0) )</4 1} = (En1] ~ V)(cu^n (Ho .94a). For example the magnetic states enumerated by the magnetic quantum number m. (8. is not uncommon.94a) into Eq. i.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. (8. m = I.ipn).152 CHAPTER 8. (8.We now insert Eqs. (H0 + eV)^n and (HQ + eV)Vm = Emym.94b) with (ipn. I — 1 . i. however. are allowed to be equal since otherwise the expansion becomes undefined. Equations (8. all n. that of double degeneracy En = Em .95a) (8. .93a) (8. (8.
y>m.H0^) .C 21 (^ m . cn(ipm.ipm ) the equations (iPn. V^n)\ . we obtain with (ipn.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .H0xl>$) .(lPn.V^n). ^ ) = ( V . and ( ^ m . Vlf>m)] .98b) + Culpm) + cuV^m)./>£>) = C2l[E£> . t f ) = a k 1 } 4 0 ) " Eg>aU = 0. Vl/.99) and ( ^ ^ O ^ ) " ^ ^ . H0lpV) .J#»a&> = 0j since ii4i = En (degeneracy). (8.m). V^n). c2iV^n + c22V7pm). Wm. We now multiply these equations from the left by ip* and integrate. (^n.H0il>W) = cn[EW .8.6 Degenerate Perturbation Theory 153 inhomogeneous equations. But (8.£ < « = ag)^) . Similarly Wn.cnVipn and tyi.(^04°))41)) = EM^CuTpn (ipl.EJ®(lPn.Vil>m)] V^m). (8. (^m.m. ^ o ^ ) ) = 0. .C22(Vn.^ ( ^ . £ aW^°Vi) .ci 2 (V„. ^ ) ) = CU[EW .ipn n) = 0 = (ipm.98a) Setting I = n. Then (^. H0lP$) = C22[E$ ~ (Vm. Vi/>n)] .^ ) ( ^ m . Equations (8.(Vn.
3). i> 4> Hence there is no degeneracy. It is then possible that a special perturbation. if)' (/>' — = —. Then / > i>" 2— = V ip Hence 0 = V ' V .This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate.ci2.Vrl>n) . i. (V>V</>V>) = const. 9} ip. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. like that providing the Stark effect (proportional to rcosO). A onedimensional system. Example 11.E™ (Vn.C22.Vlpn) CHAPTER 8. (8. dx i. If one wants to perform a perturbation calculation in such a case. d>" E=^. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). However. .154 and V (^m.101) This secular equation determines the first order correction En • The equations (8. In this particular case therefore nondegenerate perturbation theory can be used (cf..^Vm) 0. It follows that ip and <j> are linearly dependent.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. .E)tp = 0. At x = ±00 the constant is zero. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . 11. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r.100a) determine the coefficients Cn. i.e. the secular determinant.6. \nip = ln(j> + c o n s t .1. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.^ ' V = — (•>!>'<t><t>'i>).100b) determine C2i. as we saw. splits the problem into two independent onedimensional systems.Similarly Eqs. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. in particular in parabolic coordinates (see Sec. (8.e.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. Let ip and < be two eigenfunctions belonging to the same eigenvalue E.e. TimeIndependent Perturbation Theory (l/>m. Example 8.5). the degenerate perturbation theory must be used. 4. ip = c<p.
and such that one can even obtain a recurrence relation for the coefficients of the expansion. Applications of the method are. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. W.106) (8.t Since in general a solution is valid only in a limited region of the variable.7 DingleMuller Perturbation Method 155 8. . V = V0 + v.104) are chosen such that the equation Dcf) + (E0 . in particular in applications to periodic potentials. (8. (8. anharmonic oscillator potentials and screened Coulomb potentials. J. (8. T For instance spheroidal functions can be studied with this method. (8. a potential. Eq. Thus the method is a systematic method of matched asymptotic expansions.104) "This method was developed in R. We consider an equation of the form Dcj) + (E .* In various versions this method is used throughout this book.8. also possible in areas not of immediate relevance here. W. Muller [73].103).102) can be written D<f> + (EQ .V)(p = 0. B. cf. W. with some restrictions on the function V. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term.e. i. the considerations given below would be valid even if D contained only the derivative of the first order. Dingle and H.105) (8. Muller [216]. We assume that V can be expressed as the sum of two terms.e)<t>.V0)<t> = (v. See also R. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. H. the method has to be applied in each domain separately.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.103) such that if E is written correspondingly E = E0 + e. B. Dingle [72] and H. of course. It is immaterial here whether D does or does not contain a first order derivative — in fact. J. The subdivisions (8.102) where D is a second order differential operator and E a parameter. Muller [210]. J.V0)<f> = 0 (8. and the solutions then have to be matched in their regions of overlap.
V0)4>m = 0.E0\ (the latter in a restricted domain).m + s)" in order to relate them to "steps" from m to m + s.V0)<f>^ .Em+s)(/)m+s. v = {D+EmVo)<t>m.m + s)4>m+s s (8. Then if ]e < \E\ and \v . TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.108) with coefficients (m.e\ < \V0 . the equation Dcf>m+S + (Em+S . (8.) This result is obtained as follows. (8. the function 4>(0) = <t>m (8.110) If we write the next contribution <f>W to <p(°> in the form m+st (8. s an integer: (ve)(pm = 'Y^{fn. We write the coefficients "(m.156 CHAPTER 8.113) .V0)<f>m+a = 0. m + s) which are determined by the recursion formulae. Considering (frm+s we have.109) D(j)m+s + (Em . m + s)0m+s. since D<f>m + (Em .e ) ^ = 2 ( m .Vo)(j)m+s = (Em .102) with E ~ Em. and so v (8.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. where m is an additional integral or nearintegral parameter (depending on boundary conditions).+s ' (8. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .111) s then (see below) 0(1 = E (m.
e. Thus (m.Q (m.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. /01_. (8. (8.m)4>m in (8.8.m) 1^1^ IW . (8. The factors (Em — Em+S) in the denominators result from the right hand side of Eq. (8.116) m ^ v (m. . .. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8. V ^ V ^ {rn. (8.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.119) is effectively the secular equation). (8.m + s)(m + s.m) — 0 (first approximation). .m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .n\ ^119) + . Eq.7 DingleMiiller Perturbation Method and so (m. We then find altogether ^  l^m  Em+S) + y^y^ SJ.117) „ .m + s)(m + s. . and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.113) must vanish. (8. .119) follow a definite pattern and can therefore be constructed in a systematic way.m) + > ^ .^ (Em .m + s)(m + s.m + r)(rn + r.„N This procedure can be repeated indefinitely.m) — —.110) and are therefore related to the inverse of a .118).m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m. We observe that the coeffcients of expansions (8.102) to order (1) of the perturbation (v — e).m) E J^ ( m ~ Em+S) ' • fa.Em+S) = 0 (second approximation).TP \(T? . the coefficient of the term (m.T? S The first expansion determines the solution 4> (apart from an overall normalization constant).m + s)(m + s.
120) n4 .423) + Z(2720 9 3 .e. 1 . J. TimeIndependent Perturbation Theory Green's function. . The result is the following expansion.38). 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. .I2q + 64) + 256Z 3 (41 g . i. for instance. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. S. .9) + 4096J4] + 0 ( l / g 3 ) .2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential.2). Imposing normalization later one obtains the normalization constant as an asymptotic expansion.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . W.6q + 1)1 + 24(5? .158 CHAPTER 8. since it will be used extensively in later chapters. 2 . as will be seen later.[q(llq2 + 1) + 2(33g 2 .* Example 8. (8.1)( 2 + 64J3] — . Other applications can be found in the literature. as already mentioned after Eq. this is — so to speak — the price paid for obtaining a clear systematics which is essential. in which q = in + 3. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. and is typical of a large number of cases. as will be seen in later chapters. . The following Example 8.[ 4 ( 8 5 g 4 + 2q2 . for the investigation of the behaviour of the late terms of the perturbation series. ^The potential A r 2 / ( 1 + gr2). (8. (f>m in the above. Kaushal [146]. The eigenvalue expansion obtained is an asymptotic expansion. An explicit application of the method (not including matching) seems suitable at this point. This can in fact be done and will be dealt with later.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.e. has been treated by R. for instance. § H. does not occur in any of the later contributions. n = 0 . An additional aspect of the method is the full exploitation of the symmetries of the differential equation. i. Miiller [211]. This means the solution is not normalized.
2) where 1/ . 2 The solution XQ(S) = $ ( a . 11.. \2 . Therefore in the general case we may set k2 + g2 = ga(2l + q) .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. Setting q = in + 3 implies k2 + g2 ga(2l + q).z2 is a normalizable function if a = —n for n = 0 . a)ip(a) + (a.l)ip(a . (8._ „!2 ^ ( " « 2 r r 2 > .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . i/> with Dq = d2 .2a2 A.V ! 4 I z S *! V2 ipq(z).b. (8. (fc2 + <?2) 4ga and b = I \—.123) the contribution fl<°> Ah + . S) is a confluent hypergeometric function. multiply the equation by onehalf and set h = —a/g. (. .2 '(l + l) . .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. 3\ 2 V ' 2.1). a . „2 J„2JL. 1 . .123) The first approximation (note the last expression as convenient notation) ^ . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. 2 . Sees. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ .b. . (8.a + l)if>(a + 1) + (a. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.121).0(z) = zl+1ez2^fa. The solution i. . b. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf.8. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.6 and 16. We now insert this equation into Eq. 1 1(1 + 1) 1 .122) where A is of order 1/g. .
^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously.e.j^0 ' + jU+1 J ^(a + j) with h[a. Now proceeding as explained in the text one obtains finally the expansion ip = ^(°) + v ( 1 ) +y>(2) H — along with the expansion from which A is determined.a]i = . „ . a + j)i/>(a + m). a]i + h fa. If we now evaluate the various coefficients in the last expansion we obtain the result (8. a + r) satisfy the following recurrence relation: Smi(a. a ] i + 0(/i3) [a. a + r) = 0 for r > m. (a.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps".a + j}i+1^(a + j). a + 1) = a = In general CHAPTER 8. a + i) = 0 for i ^ 0.a]2+L ^ J1[a + 2 .S m (a.1) = a . In fact. a + 111 r [a. ( 5 m ( a .a — 211 .3).2a = I + q. one could now use the RayleighSchrodinger method and rederive the result (8.a—lli. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . o + r) + S m _ i ( a . of course. a ] i + L ' ^ J 1 [ a + l . a + r — l)(a + r — l . This equation shows that a term fiip(a + j) on the right hand side of Eq. a ] x + L _ 2 J 1 [ « . a + r) with the boundary conditions Sn(a. TimeIndependent Perturbation Theory (g . The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T (i+2) [a.a + 2]i P . Hence the next order contribution to tp^ becomes oo i+2 r [a a . one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. .2.i ^(i)=£V+i i=0 J2 j = (i+2). So (a. a) = 1. As a check on the usefulness of the perturbation method employed here. fa. a .l .Z. (a. fa. a) = b .b = (g + 3) . Now we make the following important observation that Dqip(a + j) = jip{a + j). a + r + l)(a + r + l . (8. a + i)(a + r. . when j = 0. i.120).120).a]1=0.160 where (a. _ J [a .123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. the latter determining to that order the eigenvalue. The coefficients S m ( a . 0 = h[a. a + r) + S m _ i ( a . where [a.
and we shall then see that these lead to analogous results as in Chapter 2.Chapter 9 The Density Matrix and Polarization Phenomena 9. and this means electrodynamics. 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. as we shall see in Chapter 27. 9. We therefore begin with the appropriate classical considerations with a view to generalization. light. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. Regrettably this has to be left out here in view of lack of space. 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. 161 . the generalization to many particles.e. i. proceeds along similar lines.2 Reconsideration of Electrodynamics In this chapter we consider some aspects of classical electrodynamics with a view to their generalization in the direction of quantum mechanics.1 Introductory Remarks We saw in the foregoing chapters that in considering a particle system which is classically described by the solution of Newton's equation* is quantum mechanically described by states which in the position or configuration space representation are given by the solutions of the Schrodinger wave equation. although not exactly straightforward.
6) . It follows therefore that A(r.162 CHAPTER 9. This is the equation of an ellipse as indicated in Fig. The Density Matrix and Polarization Phenomena potential A(r. (9.1) Fig. i.4) (9. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. 9. (9.e.3) (9. A 0 = A0R + iA0I. 9.1. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. (9. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.i). t) = A 0J R cos(cji) + A 0 / sin(u. t) given by A(r. In general the constant Ao is complex.1 The vector potential ellipse.t) = 3ftA. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.2) (9.
the direction of polarization.A(°)). I=<l2x2>:=E40)*l*40)i. 17. (9.e. From the vectors A(°). i.e. i. as e.AW).k (910) We give the initial wave the label "0" and subject this to an experiment.A2 is the sum of the moduli. a calcite crystal or a film of nitrocellulose. AW = £ fc=l.t in which k and ui remain unchanged.8). p. we recall. the vector potential A.11) 'This means we consider the polarization. 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.2 *i*4 0) (98) with coefficients F^. represented by a filter "F".g. (AW. which leads to an outgoing wave which we give the label "1". ie.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9. Observable quantities can only be those which are invariant under translations and rotations.2) the vector potential A is completely determined by its two (say (x. A2. determined with the help of a polarizer. We set therefore A 0 = Ai + A 2 . the connection must be linear.9. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. for instance. (9. See for instance A. . with k . Rae [234]. (AW.e. the fields E and B are observables. however. Then their connection is given by Eq. not. i.y)) components A i . Here we are interested in observable properties of light waves — like.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. In classical electrodynamics. (A(°).A«).A i = 0 = k .A 2 .7) We consider now measurements with polarized light. the intensity of a wave. (9. The intensity 7 of a light wave with amplitude coefficients A\' .A« we can construct only the following scalar products with this invariance: (A(°\A(°)). (9.
(9. (9. if (F) is an observable quantity.17) . (9.13) the quantity F must be an hermitian matrix and thus representative of an observable. We use definition (9.e.164 CHAPTER 9. (9.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%.e.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik).12) i. i. i.13): j2^FikAk = Y^AiF*kAi = Y.13).13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.k l f c 4 ° \ (9.15) in order to rewrite Eq. To this end we observe that as a consequence of Eq. (9. so that (F) can be looked at as the expectation value of an observable F. the connection between the initial polarization and the final polarization. Thus. (F)=Tr(Fp).k i (9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. Thus (F) = J2FikPki = ^(Fp)a. (9. pki = AkA*. (A(°).k (9.A( 1 )) = ^ ^ F i. i.16) i. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.e. it must be real and hence (F) = <F>*.A*kF£A* = E and hence F = Fl A p i tkAk.A*FikAk. Then according to Eq. We write the observable quantity as (F):=Y.
165 (9. i.* For a lightwave travelling in the direction of z.(AAt) = ( £ £ ) . "incoherently. For a 45° polarized wave we have 1 . that of a pure state. i. .21a) follows from Eq. The density matrix for the pure state (9. (9. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions.e.21b) This state is still a pure state.18) The vector A describes a particular ray of light.( J o ) . But light. we have (9 19) ' In this case there is no preferential direction.15) with A\ = a. <»^> (9. In the present case of a polarized beam of light we have two possible. This admixture is described by introducing matrices p. A2 = b as </*) . 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip.9.19).2 Reconsideration of Electrodynamics In particular we have / = Tr(p).22) For a — 1. 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. The socalled pure case.e.d "»=(o ! ) " In the case of an admixture with equal portions of 50%. like that of a bulb. is in general unpolarized.e. is given by a single polarized wave. b = 1 we obtain the expressions (9. b = 0 and o = 0. i. mutually orthogonal polarization directions x and y described by the matrices (see also below) * . (9.
166 and CHAPTER 9. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics. We set in analogy to Eq. in which the intensity I = Tr(p) is not conserved (e. b = —=. 1 = . Equation (9. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . Then according to Eq. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.23) 1 . as a consequence of absorption).Z^ 3. (9.g. l l\ " i °\ ) ' ^925^ 9. (9.3 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.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.™ V 0 n kmAm / ^^jPim^mk' . (9.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F.15) Pik — A i A k .8) A(1)=E^40)> k R = R(<*). Pi35° = \ \ 2 ? )• (9. (926) where a is a parameter whose variation describes the continuous variation of the polarization from its initial state labeled with "0" and hence parameter value OQ (analogous to t or j3 in our earlier considerations of the density matrix) to the final state with label "1".
the dependence on a is contained in F' = R^(a)FR{a). here this contains the dependence on a. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations.4 The Liouville Equation 167 pW = RpMRl (9. (9. (9. In the case of Eq. (9.33) i. and p(0' remains unchanged. the observable F remaining unchanged.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9.34) .29) where (i.28) (9.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. (9. In the case of Eq. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.e. the observable is transformed. The corresponding condition here is that no absorption of the wave takes place. i. in other words without creation or annihilation of systems. in the state functional) in analogy to the time dependence of W(p. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .32) (9. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).30) Ti{FRp{®I$) = Tr(F. (9.q.t) in Chapter 2 in what we described as the Schrodinger picture there. i. that the initial intensity Jo is equal to the final state intensity I\.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.e.9.pW).e. (9. on the other hand.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.e. 9.31) (9.31).
In view of Eq. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. CHAPTER 9.e.40) can be considered as quasiequations of motion. the matrix M is antiHermitian: Mf = .F}. (9.p]. = Tr(R^Rp^). (9. We set therefore M^:=iH.M .32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. (9.41) We see therefore that the 2 x 2 matrices entering the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics in Chapter 2.36) (9.B}:=i[A.e.p(°\ao)]da. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. Eq. (2. j£ = i[H.40) Equations (9. (9.35) Thus the matrix R has to be unitary.21) and (2. (9. Comparison of these equations with Eqs. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H. i.37) (9.46) suggests to define Poisson brackets of matrices by the correspondence {A. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.2) has to be taken into account.B]. . (9.39) If the dependence on a is not contained in p^1' but in F' (cf.39) and (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H. (9.168 i.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.
Thus in particular we establish the uncertainty relation for observables in general.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. and consider timedependent perturbation theory.Chapter 10 Quantum Theory: The General Formalism 10. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. we establish their Heisenberg equations of motion. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u). (10.2 States and Observables Every state of a physical system is represented by a certain ketvector \u) which is an element of the corresponding Hilbert space "K. To insure that the 169 . Heisenberg and interaction pictures. the Schrodinger.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. 10.
3 represent the three Cartesian coordinate directions of x. E. [qi. q). classically qiPi = PiQi.Qj] = 0. \pi. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. The phase space coordinates qi. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. Pi • Pi. we assume this now.Pi are called fundamental observables. (10. and assume that A = A(p.q).Pj] = 0. Such a correspondence with classical mechanics is not always possible.2. .* a conspicuous exception being for instance a spin system.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.170 CHAPTER 10. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A.y. however. Naturally one can also write (10. where i = 1.1 U n c e r t a i n t y relation for observables A. Here.g.Pj] = ihSlj. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A.4) *Even in those cases where this is possible the operator correspondence may not be unique. 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. (10.2) (F(A)) = {u\u) <"lf. y.B] = ih.2. 10. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space. we must have (u\u) = 1. such as for instance of the angular momentum operators Li. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. which clearly does not hold in the case of operators.
C = A.2A(A) + (A)2. (10.9a). (10.8) and.10.{A)2)1'2 {{A2) .7b). so that (A) = (A) .(A)2)1'2 (1 °= a) ((A2 + (A)2 . (3.9) ((A2) + (A)2 .. (4. (A{A))(B{B)){B{B))(A(A)) (10.7b) = = [A.6) B:=B(B). AB defined by the mean square deviation AC=((C2)(C)2)1/2. Applying to the expression (10..12) where \u) 6 "K is a ketvector representing the state of the system. (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).5)) (for A\u).11) (A2) = (A2) . (10..2A(A)}  {A)2)1'2 (10.13) AB = AB = (B2)1/2. B}= ih. Then [A. Ai = = = ((A2) .0)1/2 = AA. we have A2 = A2 . Eqs.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). (10.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.(A) = 0. using Eq. 111 (10. = \\A\u)\\2\\B\u)\\2 (10.2(A)2 .B\u) G IK as vectors) we obtain (for an application see Example 10.2 States and Observables then their uncertainties AA. Thus AA = AA = ( i 2 ) V 2 .B] = ABBA > ^h.7a) (B) = 0.12) the Schwarz inequality (cf.(A)2 = (AA)2. Since A = A— (A). (10.B.10) .. (10.
(10. (10. AAAB>I=h.e. (10. since A and B are hermitian. for a complex constant c. and hence for condition (b) 0 = = tSee Sec.16) I = h.18) >R2 + I2. We can therefore rewrite Eq. Zi + BA)\ + ^ih. and with Eq.3. The condition (a) is satisfied when.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u).17) as had to be shown. The equaltosign applies in the case when (a) it applies in the Schwarz inequality. (10. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih.14) Hence (u\AB\u) = /^(AB where. From this we obtain (uiJBu) 2 = i22 + / 2 .4).15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. 4. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. (10. (10. (AB + BA) This means (AB + BA) is real.172 CHAPTER 10. R= (AB + BA). (10. Zi Zi (10. .13) (AA)2{AB)2 i. Zi (10.15) as (u\AB\u) = R + iI. and (b) when (AB + BA) = 0. A\u)=cB\u).
we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. (ALZ)2(ALX)2 > \h2 = (Ly)2. = 0. 9ftc = 0.e. (Lx±iLy) ^ 0. Example 10.i relation holds: (ALi) 2 (AL.e. ALy = (u\L±\u) ^ 0. in view of Eq.) 2 > = x. . 9 = h{u\eiikLk\u). i. (Ly)2=0.y. implying {Lx)=0.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. the following uncertainty h2{Lk)2.13) to operators Li — Li.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. = 5R2 + G 2 > 9 2 = hi2(Lk)2. For these we have in the onedimensional case the relation [p. m\lm)). for which one component of L is "sharp"? Solution: Applying Eq. {u\LiLj\u) = R + iQ. What does the relation imply for states u) = \lm).3 OneDimensional Systems 173 Thus c = — c*. (10. q) = ih.10. i. i. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. Ly are not "sharp". such that Lz is "sharp". h2(Lz)2. Thus in this case Lx. (A{A))\u)=i(Zc)(B(B))\u). For states \u). we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li.z.e. i. (Lz)2^0.e. (Ly)=0. ALX 10. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.Lj] = (LiLj + LjLi) + ih£ijkLk. + LjLi\u). (ALy)2{ALz)2 > ~h2(Lx)2.20) or.7a). The observables like angular momentum are functions of p and q. Separating the product into symmetric and antisymmetric parts. (10.
qn] = ih^(qn). Similarly we obtain \p.p2] = [q.B]C + B[A.A(q.p}p + p[q. (3.p) can be expanded as a power series in q and p. Eq.174 CHAPTER 10.21a) Equations (10. i. Using [q. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). and (c) the spectrum is not degenerate. (10. i. Next we have a closer look at the operator q.p)]=ih^^and \p.34d)) [A. (10.C\. . (b) its spectrum is necessarily continuous.e. Then (see below) we can derive for A the following commutator equations [q.pn]=ih^(pn). We have (cf. We obtain these equations as follows.23) = ihg(q2) (10.A(q. they do not have a common system of eigenvectors.p] = ^ihp ~ Proceeding in this way we obtain [q.i h ^ ^ .q). as remarked earlier.p] = ih we obtain therefore [q. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators). q]q + # > v] = ~2ihq and \p. Q2] = [p. A — A(p. This is why the operator p requires a representation in the space of eigenvectors of q.21b) (10.e.21b) now follow with the assumption that A(q.p)] = .BC] = [A.21a) and (10.22) d ih—(p2). for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable.
We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). 10.27) (10. i. .1 how the delta function arises in the limit K —> oo.25) It is shown in Fig. U is unitary.e.28) i. = eipa/h = Uia)'1 = U{a). an operator. (10. (10. we obtain [q. > We begin with (a).10.e.26) it Here p is an observable. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. For the cases (b) and (c) we consider the operator U(a — pipa/h (10.24) lim SK(x — x') — S(x — x').f^ ^ J—oo dke ikx —ikx' (10.21a) A by U. and a a cnumber.3 OneDimensional Systems 6K(x) 175 Fig. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.1 The delta function for K — oo. Replacing in Eq. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. 10. U] = ih^= aU.
which means that the spectrum of this operator is a continuum. (10.29). ^Observe that {iji\A{p. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). 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. Thus all these values belong to the spectrum of the operator q. and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p. q) can be constructed representing a dynamical variable.e. q any other observable F(p.00).00).1 T h e translation operator U(a) We saw above. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). that with U(a) = e~ipa/h = U.q)\i>) = (i>\U^UA(p.3. the spectrum is not degenerate. CHAPTER 10. Eq. Since this holds for any arbitrary value of a in (—00. b a cnumber). cf.30) (10.31) . The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').q)U^U\ilj). or qU(a)\x') = (x' + a)U(a)\x'). i. Evidently every eigenvalue has only one eigenvector. 10. we can see that also p possesses only a continuous spectrum (from —00 to 00). In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.176 i.e. WU = 1. (10.29) This means: U\x) is eigenvector of q with eigenvalue x + a. or qU\x) = U{q + a)\x) = (x + a)U\x). (10.
U(a)\x') = c\x' + a).hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x").39) . \c{a. (10. it follows that 5{x" .e. (x'\p\x").38) h 6(x' . \x') = U(x')\0).5(x> . (10. The operator is therefore called translation operator. offdiagonal elelemts of the {x}representation of the momentum operator. In the expression of Eq. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). since q\x') — x'\x').x")c(a. We choose the phase such that U(a)\0) = \a). (10.34) {x"\U\a)U{a)\x') c*(a. Comparing Eqs.31) and (10.e.e.32) we see that \x' + a)ocU(a)\x').37) With the help of the latter expression we can calculate./ (10. (10.a). i.x") . (10.36) i.x')\ = l. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' .10.3 OneDimensional Systems 177 Moreover.e.x" .x')5(x" = (x" + a\c*c\x' + a) x'). we have q\x' + a) = (x' + a)\x' + a). (10.x" .e.32) Since U is unitary.e) ~ 5{x' .35) (10.x" . i.x') = = i. the operator U(a) acts to shift the value x' by the amount a.e) (x \p\x ) = . i.x") U{x'\p\x" % pz. (10.33) (10. for instance.
§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.L fdpeipx.35).° ° d(p) (ip) e"« = S'(x). 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)). (io.t0). The operator U(t.44) (10. it follows that 5'(x) is an odd function of x. (10. H\uE(to))=E\uE(to)). . a conservative system. See Eqs.t0) = HU(t. . Quantum Theory: The General Formalism We mention in passing that. (10. since 5(x) = . In this case the classical Hamilton function does not depend explicitly on t.4.to)m0)).45) (10.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .27) and (5.178 CHAPTER 10. 5.f dpipeipx.to) exists which is such that^ m)) = u(t. 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. This is in conformity with our earlier postulate in Sec.e. (5.4i) We assume first of all. i.e. to) with respect to t we obtain ihjtU(t.to) is called the time development operator.e. 6'{x) = !.tQ) = eiH(tt<Mh.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i.40) 10. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E. so that U{t.x") = (x'\p\x")*. i. Differentiating U(t. 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).43) (10.
to). The operator H is d efined by this relation.t) and therefore [U(t.t).t 0 )^(*0)> U(t2. (10.t0)U(t0.48)).4 Equations of Motion 179 We now demonstrate that Eq. But obviously (cf. Hence U(t.t0) [U(t + e.to) = 1.tiM*i.e.to) (differentiating we obtain immediately Eq.t)U(t. From Eq. t0) (10. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.47) e.*l)^(*l.45) has very general validity.t)l]U(t.t0)]1 Let us set for small values of e: U(t + e.t) = 1. (10. (10.t0) or ih—U{t.41) we obtain ^(t 2 )> = = J7(t2. Evidently U(t.48) with U(to. Eq. (10. Ufa.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. (10.t) = l^eH{t).49) . t0) = E/(t 2 .t0)\^(t0)).10.to) can be expressed as an integral as the solution of Eq.*l)^(*l)> = I7(*2. = U(t0.t) = l. (10.48): [/(Mo) = 1 . (10.44)) we have U(t. Then with U(t + e. i.z / dt'H(t')U(t'. t0) = H(t)U{t.
t) = l % ^Hdt) \il>(t)) Hdt (10.(£ + dt)) = U(t + dt. t)\t/>(t)) =U~ and since H is hermitian.5o) ihjtm)) = Hm)). 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. >(*)> = i.180 CHAPTER 10.53) .e. and these remain unchanged under unitary transformations. the instant at which a measurement is performed on the system. 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. (io. (10. (5. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. Now let \tp(t)) be the state of the sytem at time t.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.t0)\ii>s{to)) (10. Since the only measurable quantities are moduli of scalar products.51) Since </. the Schrodinger equation ju(t. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities. The eigenvectors of these observables are also constant in time. the predictions of different descriptions are identical. Eq. Then the probability to find the system in a state \x) is \(xm2 = (xm(ip\x)With unitary transformations we can pass over to equivalent descriptions.e. Then \Mt)) = u(t. i. Comparison with the equation postulated earlier. (10.41).30).t0) m0)). The operator U(t. Another such description is the Heisenberg picture. it follows that U(t + dt. requires the identification H(t) = H= Hamilton operator.
58) HHAH (U*ASU)(U^HU) [AH. {AH. As(q.HH]. (10.59) This equation is called Heisenberg equation of motion. Thus.t0)\ips(t)) = \ips(to)) = constant in time.54) Observables transform correspondingly with a similarity transformation. but here. we have to take into account the contribution dAs/dt.With explicit time dependence.i). t0)AsU(t.57) and rf[As. 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.H]U + iW]^U.HH]. . Differentiating Eq. (10. It does not always have to be the case that As does not contain an explicit time dependence.56) U^[As. for instance in cases of nonequilibrium or if part of a system disappears through absorption. a scalar product) remains unchanged: AH(t) = U\t. if ^ = 0. (10.e.48). i.55) and using Eq. (10.55) Thus AH is explicitly timedependent.p.10. subject to a continuous change. Here H is the Hamilton operator in the Schrodinger picture. (10. In the Heisenberg picture it is HH = U]HU.HH]+ihU^U.4 Equations of Motion and \^H) = U\t. 181 (10.p) in the Schrodinger picture dAs/dt = 0. even if As is not timedependent. as a rule. = so that ihjtAH or ihjtAH = = [AH. t0).(tf HU)(rfASU) (10. However the associated observables are timedependent.e. so that a matrix element (i. in order to obtain the Heisenberg picture (description).H]U = = = U]ASHU AHHH — U^HASU . we consider only cases without explicit time dependence of As. i. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10. For As(q.e.
In many cases the Schrodinger picture is more amenable for explicit calculations. We also observe that the momentum pi is conserved when \pi. 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. In particular.H] = 0. i. 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.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. We end with a word of caution. Example 10. Solution: We have for an observable A (A) = (i>H\AH\TpH). in the case of the fundamental dynamical variables qi.Pi): and where the expressions on the right follow from Eqs.13) to derive the uncertainty relation for energy and time. An observable CH is called a conserved quantity.Pi we have in the Heisenberg picture (replacing in the above AH by qi. if ffCH = 0. Finally we add a comment concerning conserved quantities. For an extensive discussion see B. since in the transition from one to the other the commutator remains unchanged. we used in essence unitary transformations of matrices. (10. [CH. In our treatment of representations. A. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture.HH}=0. .182 CHAPTER 10.21b). (10. Orfanopoulos [224].e. This holds also in the case of the Schrodinger picture. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms.61) Thus conserved quantities commute with the Hamilton operator. We see that formally one obtains Hamilton's equations of classical mechanics. provided the Poisson brackets in the latter are replaced by commutators. In the treatment of the pictures we used unitary transformations of operators (and vectors).21a) and (10.
62) The Hamiltonian H does not depend on time.\HAM\2) \(nA. in a different form: The eigenvalues of an observable are called "good" if [A. we can replace ([AH. This means that for a physical variable A(q. 2 d{A)/dt " and.68) To put it The eigenvalues a of A (i.63) so that with Eqs. (10. dt in.66) This means. 183 dV ' \ dt >"•"»» + ( £ ) • (10. the "centre of mass" of the statistical distribution of A.e. i. H] = 0.4 Equations of Motion Then.H}\f)\2. Let \ip) = \"4>H) represent the state of the system. then if a measurement is made at time t'.62) and dAs/dt > \WAH\i.e. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. > (10.65) and rA AA d{A)/dt' . H\tl>) = {H(H))\f). this is practically the same as at time t for times t with \t — t'\ < T. since \IPH) is timeindependent.7a). We construct the following vectors using Eq. and TA O C (10.10. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .15): (AAf(AH)2 Using Eq. a position or momentum observation is independent of t.2 > h. if we set AH = AE we obtain TAAE (10. at ^±AH>h. is displaced by A A in the interval A T = TA.67) 1 d(A)/dt is oo and AE = 0. AW = {A(A»V>.e.e.13) and (10. (A). of A\<t>) = a<£)) are then called "good quantum numbers". . (10. (10.)\2 ( = \{i.p) we then have 0=^M) = U[A. (10. i. (10.Such a characteristic time interval can be defined for any dynamical variable. (10.H]) = 0. If r is the smallest such characteristic interval of time.HH]) Hence we obtain by ih(A). i.64) = 0.
Eo(x)eiEot/he'^2h9(t). The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x . (10.70) by From Eqs.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . i. Quantum Theory: The General Formalism 10..70) we can determine the function fEo{E) integration with respect to t.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. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. the particle density diminishes in the course of time (t > 0).(x)fEo(E')e«EE'Wh 2irh f dE'^E. i ) = J dE'^El^)eiEltlhfEo{E'). the wave function ipEo(x. (10. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10.)exp(iEt/h)). But the number of radioactive (thus excited) nuclei decreases exponentially.H ) ( £ . This means that in the case of decaying particles. First we obtain from Eq.69) and (10.E1) 2irhfEo{E)4>E(x).184 CHAPTER 10. Recall for instance radioactive decay. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".72) = V'EoW 7/2h + 0i(EE )/h  oo . (10. states with sharp energy (AE = 0). (10. /oo (10. for instance.t) — ipE(x. (10.69): oo poo p / oo *PEo(x. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.e. This means.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. the wave function ^ ( x . (x)/ E o (E')5(E .tyEt/hdt = = = / J—oo dt J dEJrl>E. have around EQ the form ipEofct) = i. as is observed. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei.71) From Eq.
75) (10.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. They are called "resonance states".e.to) = HU(t. where U is the solution of Eq. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ.74) Thus the solution of this equation is the main problem. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. and is called "interaction picture" or "Dime picture". (10. i.11). ^(*o))> then \xp(t)) for t > t0 follows from Eq. of course. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. We saw previously that if the state of a system is known at time to. to) be an approximate but unitary solution of this equation. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. This description contains effectively parts of the other two pictures and is particularly useful when Eq. that** EQ. . i. Considered as a function of E. £/ = £/W(Mo)^'(Mo). (10. i.e.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). 7 > 0.105) and (20.41). (10.76) **For specific applications see e. The state with lifetime r — h/'y is not a discrete state.45). VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). 185 i. (C/(0) unitary). 10. Eqs.e.45) can be solved only approximately (in actual fact.toM{to)). i. (14. (10.e.e. every unitary transformation of states and operators defines a possible "picture").t0).10. Let U^°\t.g. We obtained the result (10. (10. W)) = U(t. ihU(t.
[/' has only a weak ^dependence (i. £/(°)£/(°)t = 1.e.e. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. H(°Ht) = ih^uW. with Hamiltonian H = H^ + H'.e.t0) 0 = l. i.79) ^'~°i. i. = HU<®U'. Now let H^(t).e. (10.e.e.78) (10. (10.h^U> = at ( io.186 Then CHAPTER 10. (10. i.81) But [/(°) is unitary. In addition we have (10.82) (10.83) ih^U^ ishermitian.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. with Eq.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. . Quantum Theory: The General Formalism ih^(uWu') i.e. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. varies only slowly with t). dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).
10.78) at (10 85) (10. (10. We also set H^U^H'UW.89) where ^s(i)) and As are state and observable in the Schrodinger picture. H = H{®{t) + H'.85) = where we used Eq.80).86) at = U^HU^U'. (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.84) U^ where H' is also hermitian.92) . lfe(*)}. (10. (10. U<®\Mt)) = \Mt)). 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^.90) = U^H'\^s(t)). (10.78).U^HU^U' H'fi'. (10.e. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. (10. (10.85) multiplied by U') on the right of Eq. and obtain from Eq. (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). U^H'U^U' (10.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). (10.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. i.
(with ihjtAI(t) = [AI. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. (10. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)).89)^0. (10. (10.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. comments after Eq. A^t) = U^AsUi0). Since H'j is assumed to be small (i. we have i. In order to obtain the equation of motion of an interaction picture observable Ai.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation.e. i. we differentiate with respect to time the operator defined by Eq. (10. (10. i. so that with Eq.H\0)}.94) _Hf)Al 1 + lhdAl+AlHf).90) and replacing the left hand side of this equation by the right hand side of Eq. H'j = UWH'UW.e. i. cf.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .60b) with iff0) instead of H.188 CHAPTER 10.e.e.e. (10.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. a perturbation so that H ~ H(°)). the vector varies with time (different from the Heisenberg picture). near zero and \tpi(t)) is almost constant in time. ot i. (10.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). dAs/dt=0) .89).60a) or (10.55)) at + imm<Ms_u(o) at dt _umH(. ^ (0) l^(*)>.e.94) Then we have (normally with dAs/dt dt (io=8o) — 0. (10.
\Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. H0\ipn) = En\(fn). H' = H'(t).i0) = e<Ho(tto)/ft.80) as E/(°)(t.102) We assume that before the perturbation H' is switched on at time to. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. ih±\Mt))=Hi{t)\Mt)). ^ \<Pn){Vn\ = 1n (10. We have the Hamiltonian H = H0 + H'. (10.10. (10. (10.99) (1098) Instead of the original Schrodinger equation we now solve Eq.t0).98). We assume again that the spectrum and the eigenvectors of Ho are known.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. (10. {ipn\<Pm) = Snm. We obtain the operator U^ from Eq. i.e.100) (10.93). (10. (10. i. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture.e.e. i. Since .t0)H'uW(t.87) and (10.94) Hi = uW(t.97) The equation to solve is Eq. where according to (10.7 TimeDependent Perturbation Theory 189 10.
Hence we set ^> = £°«(*)l4 0 ) >n.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„. (10. (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (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 .107) Jto With this we obtain from Eq. n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon of the perturbation H'. the time dependence of \ips } can be separated as for stationary states.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). 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. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). (10. (10. (10. i.104 ) Here we insert Eq.103) for m — n and use Eq. Quantum Theory: The General Formalism HQ is independent of time. we can write l 4 0 ) ) m = elHot/hWm) =e .105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. (10.190 CHAPTER 10.e.106) We insert this expression into Eq.102) and truncate the series after the first perturbation contribution.108) .^ V ™ ) .
< KO?t . if in the scattering of a particle off some target its momentum p changes to p'. for instance.Em)t' (<Pn\H'(t')\pm) = Wnm(t).112) for a^0.110) With this we obtain from Eq. (10. A further example is provided by adecay. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.En (10.109) the transition probability Wmn(t) = jp Jo .111) Different from Eq.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.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. This is the case.109) 10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. » — < (10.10.Em)t} \{tfn\H'\ipv h 2h(En .25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . We assume that the perturbation is switched on at some time to. t o = 0. Thus we set H'(t) = H'0(t).Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . but that otherwise the perturbation is timeindependent. or if by emission of a photon with continuously variable momentum an atom passes from one state into another. where the momenta of the aparticles form a continuum. (10. (10.
2. and no other state would be nearby. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.io. 10. Hence lim St(a) = 5(a). we would have En — Em ^ 0.192 CHAPTER 10. so t h a t Wnm would be zero. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. . Otherwise. 10.Em)t} r i 7 TTT. (10.2 The delta function for £ — oo.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. Fig.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.E7m)t} = (10. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .— = o (En 2h En 2h5{En .113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . though sufficiently small. > 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.Em). Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. if the energies belonged to the discrete spectrum.
142.116) The transition rate V is the transition probability per unit time. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . _ 2irh AE > > Se. if p(En) varies only weakly with En. Pauli [227]./ t>oc TZ J_00 dx . Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). sin 2 (tx) ax i—( * According to F.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. Fermi gave this formula the name "golden rule" . 266. for instance from a potential. i. (10. . r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. pp. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. p.8 Transitions into the Continuum 193 as equal. so that initial and final states belong to the continuum. F Jo . then Eq. Schwabl [246].r Z x 1 ^ ) = 4>{Q).113) Use the method of test functions to verify in the sense of distribution theory the result (10. If we are concerned with pure scattering.117) in many applications. (10.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation .4). Example 10. We evaluate the integral 0(0) = 1. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.e. Then.* In view of the usefulness of Eq.113) this is* hm . f See E.117) provides the outgoing particle current (see Example 10. In the case of the functional (10.3: Application of test functions to formula (10. (10. Fermi [92].117) were derived by Pauli in 1928. (10.115) and (10.W . in the article of W.113). 75. Let p(En)dEn be the number of states in the interval dEn.10.
and hence . Gradshteyn and I. dEk = h2kdk//j.? 2TT / L \3k2dkdUn 1 /V(x). formula 3. . Quantum Theory: The General Formalism with the help of Tables of Integrals. Jin § m = See I. 1 . 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). E = h2k2/2fi.a 2 ' i that for a = b = t .tan p 2 + a 2 .^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . .( ^ V V • < ih 2 _ i k n .k. .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). M. p = 1 we obtain I = .. . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. V +(ab)2 a _x 2p6 7 m —^ '. The ingoing particle flux is therefore v / L 3 .„I. Hint: Use for ip„(x) a plane wave ansatz with box normalization (volume V = L 3 ) and obtain da from the ratio of outgoing and incoming particle fluxes. Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . 2 . x . with n . = ( — ) dk= 2ir (—  2irJ k2dkdfl.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. L \3k2dkdUn It follows that T.. . ^o n .194 CHAPTER 10. K = k 0 .. iknx . S. and is obtained as ih. . with v the velocity of the particles. E x a m p l e 10. p. * ^ = . .<p). 77^ + .. Ryzhik.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. . 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. . . 491. . 2 T / ^ M. Then p(Ek)dEk Since Ek — h2k2/2[i.e " 1 * 1 = 2 lim — . § We have r Jo \p\x dx sin ax sin bx  V. Then ^ . = 0 .117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n ./ cfc a n p 2 +fc 2 .947.
(10.121) . The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>).x) = S(EE').10. 148. .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.^°>(E. For a possible application of this result.122a) (10.^ We start from the Hamiltonian H = HQ + H'. by which we mean without leaving the Schrodinger picture.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. 10.x)^(E'.t)+ / n a{E. (10. 1 rfx^°)*(x)^)(x) = 8mn. (21. (10.t). (10. 193. Thus we set ^(x.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).t)^°\E^t)dE (10. we present another derivation using the usual method of timedependent perturbation theory. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.120) In the case of Eq.122b) S e e also L./continuum = <San(£)</4°)(x.t) = ^On(t)^(x. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known. see the discussion after Eq. pp.69). We expand the wave function V( x . t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. Schiff [243].
. since f rJ>(°>(E. Observe that these conditions do not contradict Eqs. The second term on the left of this equation and the first term on the right cancel out. f 6kn in case of discrete tp's.t) = eiEtlhy(E.x. ^0)(E. (10.x. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). (10.„.196 CHAPTER 10. ....iJ5 »*/Vn(x).126) + H')<pn^)eiE^h}.x.128) . = e^EE'^h5(E Thus with Eq. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. V>(x. Next we use the orthogonality relations .„„.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. We multiply this equation from the left by </?£(x) and integrate over all space.t)rJ><®(E'..E'). . (10. into dip (10.125) where an(t) — dan(t)/dt.123) We now insert these expressions into Eq.127) i n c a g e Qf c o n t i n u o u g £.e.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. .E') = 5{E . (10.. dnptMrnW = { ^ _ En) (10.120).122b).x). (10. (10. (10.127) we obtain from Eq. t) = San(t)<pn(x)eiE^h.t)dx. .129) . / *. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.122a) and (10.i) = e. i.
e. In order to solve Eq. (10.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.134) .120). (10. SH'kn(t)e^~E^ha^(t). SH'kn(ty^E^l^\t) (10. Comparing coefficients of the same powers of A on both sides. (10. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. (10. Thus A serves as a perturbation parameter which permits us to equate coefficients of the same power of A on both sides of an equation. We restrict ourselves here to the two lowest order equations. i. let's say am / 0. to perturbation theory in the lowest order. they are fixed by the initial conditions. i.130) we develop a perturbation theory.10. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. determine the state of the system at time t — 0.130) along with the parameter A in front of H'kn.132) These equations can be integrated successively. since it is equivalent to the latter.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . (10. In the first place we replace in Eq.131) and insert this into Eq.133) The numbers ak = const.130) H'kn by \H'kn and at the end we allow A to tend to 1. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. (10. 0. Thus first we have ak0) = const. and we set ak°}=5km or ak0) = 8(EkEm).e.130) This result should be compared with the Schrodinger equation (10.
198 CHAPTER 10. i. (10.136) we see that both "amplitudes" are Fourier transforms and look similar. i.Emy in agreement with Eq.135) and (10. (10. (10. What is the physical significance of Eq.ml2sin2{^frn (En . Equation (10.135) / 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.132). H'^y^"^^/hd£.135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0. The amplitude / B o r n is a Fourier transform in spatial coordinates.111). (10. so that ihakl\t = oo) = 0.137) .e.e. whereas ak is a Fourier transform in time (a distinction which is easy to remember!).135) We put the additional constant of integration equal to zero. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. (10.135)? 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. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . ) ' X V(x')dx'. Integrating the second of the equations (10. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9. ipm) into the state k. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. ip) in the context of timeindependent perturbation theory. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'.e.136) Comparing Eqs. ifm{kl) = f J —oo (10. In this case we obtain from Eq.k . (10.
m.2. meaning the inclusion of the creation and annihilation of field quanta). Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory.2 Separation of Variables. r=  r . Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.i be respectively the momentum.y. 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.Ti. ( 1L1 ) We let pi.Chapter 11 The Coulomb Interaction 11. Let r : = (x.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. Z<i = Z) y(r) = _ M ^ . The Hamilton operator is then given by w= jt + ji_^M. 199 (1L2) . such as the electromagnetic field (the quantization in these contexts is often called "second quantization".1 Introductory Remarks The quasiparticle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields. 11. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. the position coordinate and the mass of particle i with i — 1. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.
y. so that as expected according to Eq. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. /'Pi p = m0 P2 A v. Z) and r = (x. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. p are such that as operators in the position space representation or {R. r} representation they have the differential operator forms Setting R = (X. so that as expected we obtain the expression for p in Eq. Y._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. mi + m 2 iTi'2 r2 = R ^ —r.200 CHAPTER 11.5) The momenta P .3) and centre of mass momentum P and the relative velocity p/mo by setting r» . (H4) \mi m. (11. mi + m 2 va\ (11. where m 0 = mim 2 . P = pi + p 2 . . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc.2/ mi+m2 The mass mo is usually described as the reduced mass. (11. r = rir2 mi + m 2 (11. (11.U7&.4) P = p i + p 2 . Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' . The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . Solving Eqs.3) for the individual position coordinates we obtain ri=R+ •^ —r.4). 2Pip2 m\m2 pip2 mi + m 2 .
etc.2 It follows that p2 2 P2 n2 n2 n2 (H.Ri. (11.Ri. (11.16) Although we write ri.11.= ^. (i.pj. (11. .r2) to (R.pj. (1114) m0 We observe that the equations of motion in R. 2(mi + m<i) 2mo 2m\ 2m.y.e.z).13) * and = M ' 1V1 P = 0.12) (11.il) ""Op' i.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. r = —. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii.10) we set H = Hcm + H. Pj over to operators ?i. Angular so that Momentum 201 ^ ^ .+ ^2.+ ^(r) 2(mi + m 2 ) 2m 0 (H. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. p on the other are completely separated.15) In accordance with Eq.pj]=iMij. (11.8) and so W =777 v + 7 T . P on the one hand and those 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).Pj} = ih6ij. Hence Hamilton's equations apply and we obtain. p = W. for which we postulate the commutator relations* [ri. Ri.Pj). (11. with M := mi + vn^R = ^ P. Pj (which in the following we write again ri.j = x. we mean n = rx s x. The canonical quantization must always be performed in Cartesian coordinates. pj.+ P. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.2 Separation of Variables. r) is a canonical transformation.10) One can convince oneself that the transformation from (ri. [Ri.
V(r) = V(r).20) r2MAR_ $(R) = ER*(R). The Coulomb Interaction nc H = 2m. (11.Q 2M' + V(r).r) = *(R)V(r). —n < tp < ir. 2mo (11. 0 < r < oo. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). where Hcm$(R) = and Hip(r) = with ^total — ER (11.24) .23) y — r sin 9 sin tp. 0 < 9 < ir. (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R.22) If. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11.e. 9. (11. Thus the twoparticle problem is practically reduced to an effective oneparticle problem.21b) + E.202 CHAPTER 11. it is reasonable to go to spherical coordinates r.r). = rcos9. (11. ip: x z = r sin 9 cos ip. r ) = J E t o t a i*(R.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R .17) (11. i.21a) L*+v{r)] ip(r) = Ei/j(r) (11. as in the case of the Coulomb potential.
2 Separation of Variables.2. = (p2 + ^ V (r + 0). i. (11.26) However. otherwise. = 0. with tijk — + 1 .pr]=ih. i( i.j.)In Example 11. 2 1 9 / 1 9 .3.28) . is the operator of the relative angular momentum.5r = .31) . . (11.27) We can now write (see verification below) p V = h2Ar^ where 2. k an anticlockwise permutation of 1.29) . n ( . We have I2 = (r x p) • (r x p). (11. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«.2.e. i—>0 (1L25) (11.2.11.12 — I2.3. = —1.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq.33) i. j .24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11. (11. (l + . _ h2 \ .1 it is shown that pr is hermitian provided lim n/>(r) = 0. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r.i ' . where (r x p)j = ^eijkrjpk. k = 1.3. (11.k a clockwise permutation of 1.j. that the above expression follows from l:=rxp=ift(rxV). if i._0di})\ x We now demonstrate that 1.32) (11.
(r • p ) 2 + 3ih(r • p) .k ~ rjpkrkPj) (11.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'. (11.35) j.j.ih)pr and x 2 2 9 9 2 2 Q O (11. Eq.43) r2(p2p2). This means that l2 = so that 9 9 1 (11. * ' S Since r =xz we have (cf.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .44) . (11. (11. ih6jkrjpk (11.34) W i t h this we obtain — keeping in mind the ordering of r and p — I 2 = 5 Z eijkrjPkeij'k>rj>Pk' = ^{rjPkrjPk i. (11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .ihr • p r2p2(rp)2 < + ih(rp). (11.37) . r— = r • — or or (11.40) (11.ih5jk)pk (11.27) we obtain r{prr)pr = r(rpr .42) (11.ih) — rpr(rpr With Eq.ihr • p ) .38) + y* + z . (11.41) o = r pr.39) + ih).k Using Eq.
<p))ril>(r. 6.<p)drdna " r r=0 J \ll§r~(r4'(r.d. 0. (11.d_ r dr the operator pr is hermitian. pr = . 9. r—>0 il>(r.48) . Then {<j>.ip) = (<£.e. <p)*Pri>(r. 11.e. 9 € [0.<p) (11. (pr4>.<p G [0.46) (11.2.1 Separation of variables We deduce from the expressions of H and l 2 that [H. S o l u t i o n : Set d£2s = sin.<p)*?(ri.47) Example 11. prip) = = I <j>{r. (11.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).2 Separation of Variables. or J . reR3. <p)r2dfls = .. 8. /'(r. (11.e lim rip(r) = 0.<p))drdns.6d0dip. <p)*rip(r} 0.e'iP^} .30) I2 is to be identified with the square of the angular momentum operator. i. h f l d —(r<l>*(r. Angular Momentum 205 Comparison of Eqs.(r.[ 4>(r. The Schrodinger equation of the relative motion can therefore be written V1 V 2m. Pri/Oi V0.— (riP(r.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. <p) d °° . H 1 2mo pi + K) +V(r).2TT].28) and (11.e.7r]. <p)* .e.. ¥ p)r 2 drdn s = (p r 0.e. 6.45) oe Thus in Eq.6.[ i J With partial integration with respect to r this becomes (<l>.6l.11. i.! /'). r := r : lim rip{r) = o i . ip 6 X>Pr C 7i.l2} = 0.Q ' 2mor2 with the condition (11.26). <p))r2drdns % J r or r<t>(r.<P) = Etl>(r.Pril>) J / r= dQs rcj>(r.
51) (11. g i = x. (11. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l . zpx] + [zpy.lz]=ihlx.lx] = ihly.e.q2 = V.ly]=ihlz. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. These operators here play a role analogous to that of the quasiparticle operators .pz]x = ih(xpy . for any two of the components there is no common complete system of eigenfunctions. so that [lz.l2] = 0. we have [h. zpx . 11. (11.53) Either of these operators is the hermitian conjugate of the other. i. y.l2} and correspondingly [lx.50) Thus the components of 1 do not commute pairwise. However.e.49) By cyclic permutation of x. z]px + py[z.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0.ypx) = ihlz. H and l 2 possess a common system of basis eigenvectors. z we have altogether [lx. ?3 = z). We therefore first search for a complete system of eigenfunctions of l 2 . 3 .xpz] = [ypz.206 CHAPTER 11. 2 . ly] = [yPz ~ zpy. Since 1 = r x p .lz (or l . (11. [lz. (11.3 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. Example 11.lx or l . xpz] = y\pz. [ly. [lyA = 0. they are incompatible variables.e. The Coulomb Interaction i.52) This means that l . 2 2 2 = 0. /z> *xJ = l <i\}ylx + tx'j/Jj [h.2: Verify that [h. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. i.
in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.e. i + ] = [l2.i] = m.i+] = m+. Similarly we have lg{l)2\l) and more generally lz(l.lh)\l) = {l0 . Hence we must have Z+I0=0. (11. In the present context the operators are called shift operators for reasons which will be seen below. .+ ll.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq.Y\l) = (lQr)h(lY\l).11. (n. (11.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). We write or define: / . (11. With the help of Eq.3 Representation of Rotation Group 207 A.54) On the other hand we obtain from Eqs. (11. A^ in the case of the linear harmonic oscillator. since lz is hermitian. (11. [i+M = 2%iz.).52): 0 = [ l 2 .l)hl\l).57) i.49) we obtain the following relations: [iz.54): lzl\l) = {llz . [ig. (11.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real.h~lz(1156) (11. y . Therefore (lo + l)h cannot be an eigenvalue of lz. (11. (11.61) But lo is by definition the largest eigenvalue of lz. l\l) is eigenvector of lz with eigenvalue (IQ — l)h.58) {l0 : max. (11. assuming a finite dimensional representation space.51).Z_] = [ l 2 . (11.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.r) (11 = {l02)h{l)2\l) (11. and with Eq.59) = 9) h(l0 r)\lr).54) l 2 = \{l+l~ + 11+) + ll = l+l. (11.60) Similarly for l+ with the help of Eq.r) := (Z_)rZ) so that lz\l .
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> (11.55)) we have 12Z_Z> = l.66) It then follows from the second of relations (11.Z_] = 0.e. (11.. (11.h)\ln) {(lQnf(lQ.y\i) l)h2\lr). Eq. l\l .e.l2\l) = l0(l0 + l)h2l\l).56) CHAPTER 11.54) (n = 57) (11. [12.63) + l ) ^ 2 .n). i. (11.\ll). (cf.60) and (11. (11.\lr) terminates at some number (let us say) r = n (n a positive integer).+ 11+) + l2z\l) = i+i.n) (1 = 60) {l)n+1\l) = 0.e. IQ(IQ 1 0 (11.64) i.67) i (11 60) = But according to Eqs.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.e. (11.208 Thus for l 2 we obtain: V\l) ' = 11. \l).68) . (11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. (11. The Coulomb Interaction (l+l.From Eq.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i)..65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.. i.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .. (11.n)}h2\ln). l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .
17. We therefore write the 2/o + 1 orthonormal vectors: \l..m\lz\l.. The normalization factor in Eq. *D.m) (l.  U o . II.l)h. (11.m) = mh\l.. \l — n) into one other..e. The dimension of the representation space is therefore n + 1 = 21Q + 1. p. p. where lz\l. (11.n .. lQ = .l. (IQ — l)h. m + l)h. i.m) with — lo<m<lo2 (11..63) and (11.m).71) can be established in analogy to the method used in the case of the harmonic oscillator. 24..70) According to Eq.70) they satisfy the eigenvalue equations (with m = l0Jo l. The operators l+. (11.1). lo = l0n. The nonvanishing matrix elements of 1 = (l+..m) = = = = y/lo(lo + 1) — m(m +l)\l.lz transform the vectors \l). .m) — mh\l.. (11.m + 1)(Z0 + m)\l.. Brink and G.m). we skip this here.n) 2 .m) = = mh.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .m) = IQ(IO + l)h2\l. R..m) : Mo)..m)..l0): l2Z..65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. As in the case of the harmonic oscillator we introduce normalized basis vectors. y/(l0±m + l)(l0^m)h. The phases are chosen such that§ l+\l. lz\l. M.l_. y (lo .m / + l)h . (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.g..69) Prom this it follows that IQ must be either half integral or integral. Messiah [195]. According to Eqs. \l — n) are eigenvectors of lz with eigenvalues loh. Satchler [38].{lo .m) l\l. We write these* for integral values of IQ: \l.n){l0 . —loh.11. In the following we consider primarily the case of integral values of IQ.2.Z 0 ).1).m)\l. lz) are then: (l. The basis vectors \l).m .3 Representation of Rotation Group so that with Eq. Vol..n)} = (l0 . A.m±l\l±\l. § See e. (11..l)h \/(lo + m + l)(l0 .m y/lo(lo + 1) — (m — l)m\l.
l0)=0. —lo + 1 .74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).yp.<p) = i)n2Y^{e^).. 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.. for the simple "rotor". d2 sin2 9 dp2 _' (11. In these coordinates we obtain: h l± so that h2 = xpv . mhY£(9. CHAPTER 11. We now perform the analogous procedure for the angular momentum or.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. i.<p) = i0{io + lzY£(0.73) 1 d d sm0 06 \Sm9dd) + (11. . ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ .. $ oc eim*. as remarked earlier.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. The Coulomb Interaction Z+Mo> = 0. 11. 9.. ip. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). where IQ = 0. l_\l. .2.<p).75) the variables contained in the expression (11.1.76) d2 (11.With the ansatz Y(9. .4 Angular Momentum:Angular Representation In the case of the harmonic oscillator we arrived at the position or configuration space dependent wave functions by considering the position space representation of states.210 Then. . (11. and m = —IQ.77) $((p) = m2<b(y).e. (11. as this is sometimes described. with i2Y£{e. ip) = e(9)$((p). /o. (11. We . as above.
* ? ^ .ypx = tfrK(11. orthonormalized system of eigenfunctions.11. This is the case if one demands that l 2 and lz possess a common complete. (11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.^ = *(n .n>).82) .79) .(p) are called spherical harmonics. m > 0.(5cos 03cos0). (11. These functions are defined in such a way that they are normalized to unity on the unit sphere. so that im(ip±2ir) The functions Ylm(8.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. is given by (2Z + l ) ( Z .i». The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs. (11.80) y? Furthermore Y°l = V^fcosO).\ = W . and that their phases are such that Yt0(0.V>)Yir'(8. (11.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}. We mention in passing that in spherical coordinates d lz = xpy . y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.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.t»'\Y7n((l'.r o ) ! YT{e.<p) Jo = 6mm'5w.76) and (11. / 3 / 5 J — casO.0) is real and positive. Uniqueness implies immediately that m has to be integral. (11.
the designation s. Consider the equation for Pi(x).e. (11.212 CHAPTER 11. (1 ..88) we obtain for every value of x ^2 ai(K + 0( K + * .e..88) y = ^aiXi+K.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function. we would now have to search for those values of I for which the solutions are square integrable. in the replacement ip —> if + 2K.m(m + l ) ] i f = 0.x2)P[' . Instead of the values I = 0. It is instructive to pursue the appropriate arguments.x2)y" .x2)Pf .x2)y" .l)xK+i~2 i . oo (11.. i.2.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 . 22 (11. Setting y = (lx2)m>2Pr(x) m . The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.2xP[ + 1(1 + l)Pi = 0. exp(imip).87) we obtain the equation (1 .2xy' + 1(1 + \)y = 0.g.85) (11.86) y = 0.90) . i=0 (re lowest power > 0). (11.89) Inserting this expansion into Eq. i. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . (11. (11.e. respectively. i.1.m = 0. is also in use..h.2(m + l ) x i f + [1(1 + 1) .d. i. of Ytm.1(1 + l)]xK+i = 0. f.. For the solution we use the ansatz of a power series.p. (11..e.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) . If we did not know yet that I = IQ = an integer > 0.
l(l + l)h2 _2mo + 2m. Schmidt).94) ^For further details see H. (11. Margenau and G. . 93. they have to be polynomials (recall the orthogonalization procedure of E. Eq.1(1 + 1)].5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. We see that this happens precisely when I is a positive or negative integer or zero. ai(re + !)« = 0. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . p. Eq.94) from the left by r) h2 d2 h2 2 + H! + \)IL.46)) p.E Xl(r) =0 (11.78) to be square integrable.11. > 1. This is precisely the case. (11.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.92) This is a recurrence relation. We have therefore (multiplying Eq. For the solution in accordance with (11.Qr2 with (cf.91) from which we deduce that K = 0. first line of Eq. Murphy [190]. (11. (11. However. (11.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. when the series terminates after a finite number of terms.96) + V(r) . but also those for a different series for a. (11. l 2 and lz are the solutions of the Schrodinger equation of the form (11.*" 11. these are of little interest to us here.26)) yi(r = 0) = 0.5 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. M.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.93) where Xl(r) is the solution of the radial equation (cf. For \x\ < 1 we can obtain from this the coefficients of a convergent series.1. .
101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr.1)).99) e ip(r) = Eip(r).98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. L i + + .1.1)+1(1+ 1) = 0. The equation also has an irregular solution. Equating the coefficient of 1/r2 to zero.e. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). and substitute this into Eq.102) . since it satisfies neither the condition yi(0) = 0.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. r m0 = meMp me + Mv (11. and others. The Coulomb Interaction (11. He + . Obviously the irregular solution has to be rejected. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. (11.96). The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11.100) with (11. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. We now consider the hydrogen atom with V(r) — —e2/r (i. nor the condition of normalizability for l ^ 0. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. (11. oo) of a onedimensional space. i. Z\ = Z2 = 1 in Eq. s = 1 + 1. an analogous consideration applies to hydrogenlike atoms like. we obtain the socalled "indicial equation!'' s(s .e. (11. In the first place we investigate the behaviour of yi in the neighbourhood of the origin. for instance.214 with CHAPTER 11. which behaves like 1/r1 in approaching r = 0.
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.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.103) °y moe2 ( = 0. (11.v) vi(x) = 0.108) This equation has the form of a confluent x$" + (6 . (11.g. *The factor 47reo appears in SI units. We set therefore Vi X J.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0.105) 11. It remains to determine the function vi(x) for a complete determination of the solution. hypergeometric (11.11. see e. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . W.e. i. p.105) we obtain the equation for V[(x). (11. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. (11. (11.6 11. .e. (11.104) W i t h these substitutions we can rewrite Eq. (11. Magnus and F. We can see from Eq.88.a * = 0.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.x ) $ ' . i. Oberhettinger [181].107) equation.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). T This equation is known as Whittaker's equation.+if>x/2 vi(x).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. for E < 0 they > decrease exponentially. Inserting the ansatz into Eq.6. behaves there like xl+1) extended to an exponentially decreasing branch at infinity.e.
and hence the *Some writers. the factorial for an arbitrary argument.e. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + .106). more precisely.! ) ! = imlFI' c a l led inversion or reflection formula.z) = ^f^.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. However. (z)!(z . . (2) r ( z ) r ( l . particularly pure mathematicians. In general the function $ is an infinite series and behaves at x — oo like » x{l+l+u)ex_ Obviously such a series is useless in our case. The function T(z) is defined by the integral /co T(z) := / eHzldt.112) The gamma function is a nonvanishing analytic. (11. and for both cases. with many of these around in some calculation.J. ^ ( 2 ^ ) ! = 2 2z z!(z . of its argument z with simple poles at z — —n and residues there given by (—l)"/n! as may be deduced from the property (2) above. it is convenient to use only the factorial notation. (11. meromorphic function. called dupZication formula. (11. i. (5)r(n + l ) = n ! .(0_i)!(6 + n l)!n!' (11. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .z(zl)\ = z\.1/2)!.216 CHAPTER 11.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (4) r ( i ) = V5F. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z.
2.V))T(v . This is achieved.. n' = 0 . every n.1) + n = n2.110) as follows: T(l + 1 + p . which are also described as its nodes. in our case: a = 0. One defines as the principal quantum number the number n.e.e.» • » • » .1.113) This is a quantization condition. . .. i. and so to the number of different values of m. The quantum number n' is called radial quantum number.I) r(vl p)sm{n(i/ I . 1 .p ) } n (l)Pr(^Q n'\ K T{vlp) ' (n'p)!' Note that 11 + 1 is the number of possible orientations of the angular momentum vector 1 with respect to a preferred direction. for z. d = 1. i. which is thus seen to be a consequence of demanding the square integrability of the wave function. 2 . i. the orbital or azimuthal quantum number I can assume the values 0. 1=0 (11.101) to (11.1) + n = n(n .. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression..104)) 1 n =v = K. must be a polynomial.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. = n = Z + l + n'. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). Thus the infinite series must break off somewhere. (11. Eqs.v) T(l + lv) _ ~ K sin{7rQ/ . 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 . (11.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. For every value En. if the series terminates for certain values of u. . With (cf. . .a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ .6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. (11.e.11.
I = 0. hence d).. nf .'.. and hence the fine structure of the energy levels is excluded.§ One also writes n2l+1l. 2 .l = 2 to n = 2.1 = 1 (called diffuse series.. stand for I = where n denotes the principal quantum number and s..218 CHAPTER 11. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11.1.1 = 1 to n = 2. ne. (c) The nucleus was treated as pointlike and not as something with structure. The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig. . np.d.1 Hydrogen atom energy levels. from an initial state n > 2. from n. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. .. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy.1. The following spectroscopic description of states is customary: nl —> ns. nd.\ I = 0 or 1. . 11. hence s).4.'. / = 1. respectively. . hence p). 0. . I = 0 to n = 2.1 = 0 (called principal series.p.1 = 0 (called sharp series. from an initial state n > 2. 11. The spectrum of the hydrogen atom thus has the form shown schematically in Fig... (b) The spin of the electron was not taken into account. ' O n e describes as Lyman series the final state with n = 1. One describes as Balmer series the final state n = 2.. and from n.
(11. 2 . and I. .2 (11. • (11. .6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'.118a) where Lk. Unfortunately. 84. l .117) The function *Lk(z) is a polynomial of degree r.414(3). We distinguish between different definitions in use.r f . Magnus and F. (u.z) = (r + k)\ Lkr(z). so that one always has to check the definition. Oberhettinger [181]. Appendix B. Apart from factors. however. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). (11. for instance. the polynomial contained in y\ of Eq. • fc. . • ( .1. S. Ryzhik [122]. . . p. 3 .2l { + 2. For clarity and later use we denote the function used there by its normal form L^. several different definitions are in use. p. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. formula 7.x). 11. see. Vol. Unfortunately the definition mostly used in mathematical literature differs.r = 0 . (11.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.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. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. 844.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). Messiah [195].= t . I. 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. Each can be recognized by reference to the generating function.115) is effectively the associated Laguerre polynomial normally written Lk(z). Gradshteyn and I.6. M. "1'f„+11" P] „. a + 2:*) = t v .115) ' =0 (ri .114b) yields the binding energy of the states. in fact.2.p)\(2l + 1 + p)\ The spectrum (11. With the help of Eq. k + l. We mention here three different definitions used in the literature.11.
Example 11. ^ F o r this definition and the following equations we refer to I. It is said that this is the definition preferred in applied mathematics.. Appendix B.3): e2t/(lt) (1 . N. (11.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. . 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. is defined by^ dk (—'\\k(r^\2 (11. . N. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 .122) "See A.220 CHAPTER 11.** The associated Laguerre polynomial here.d z—^z + {k + l~z)—+r Lkr{z) = 0.164. There at some points n + 1 is misprinted n + 1. and Ll(z) = l. 162 . Vol. Sneddon [255]. p.„\*Kk.121) In Example 11..120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.t) + k l r)! E E (zY (i +(fc +h)\(ri)\ i\ r=0 L OO p OO i=0 E (k + r)\ *L (z) = J2L ( y. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. I. Messiah [195]. Sneddon [255]. L\(z) = 2z. **I. r r Z r=0 v tr k k (11. L2(z) = l2z + z2.2.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function.1. 163. Ll(z) = 33z + ±z2. L1(z) = l .119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. pp.z . . which we write ^Lk.„M„ z"*L.
Example 11.124c) Example 11. and obtain with this the normalization condition (11. and partial integration.1 . .118a) and (11.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq. (11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.1 8 + 18z 3z2.4 + 2z. ^L2(z) = 2 . (11.124a) (i*) pi The orthonormality condition is (cf. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11. and *L\(z) = .119) derive the generating function of associated Laguerre polynomials (11. Then. N d rk 0.z .4z + .11.8): (11. Jfpd>2s{p))eptdp.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp. (11. *L\(z) = . The limits will be determined later. (11. i L1(z) = l . (11..119): ty"(t) + (a + 1 ..120)) it)k& zt/{lt) ~*P p=k fc (11.124b) Comparing Eqs. ^L\(z) = ..121). '"''Comparison with Eq. Solution: We consider the following equation of the form of Eq.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n .120).. (11.107) implies k = 11 + 1 and r = n + I.
q) and P~ t ( . agreeing with Eq. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp.^ . since for n > b : (b — n)\ —• oo. . and (4) around p = 0 if b > 0 and integral.. s bK> (lq)°+i (11. the only possibility is the last. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. (11. e ^ t '?/( 1 '?) = coefficient of qb in /"OO / = = / Jo Ll{t)Lac{t)tae~tdt taet{P/(lp)+q/(lq)+l)dL l t I * * [" (2TT." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt.] from the above equation: C(p) := p(p + l)s(p)ept = 0.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. We demand (to be considered later for integration contours) the following condition. this is the rejected irregular solution. Hence the expression in square brackets vanishes.119) with a and b as there. j p = (P 0 "V?. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. with requirement of a finite solution. p and one has (as polynomial of degree 6.. case (3) similarly. Then dp = dq/(lq)2. (2) from —1 to oo.p ) « + i q c + i ( i . t > 0.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. (11. Hence one defines now as associated Laguerre polynomial L£ and as the generating function g(p) of these (Cauchy's residue theorem requires the coefficient of 1/p): L b(t) •= — <£ 27T.) 2 J J p 6 + i ( l ./ ( l — <?) —— = G(t). It follows that 1 d i I _i. Thus in the case of Eq.222 CHAPTER 11. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . (3) around p = — 1 if a + b is a negative integer.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} .a = 21 + 2. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). which removes [.p ) o + i 7 t = 0 .p + 1 = 1/(1 . Case (2) implies the exponentially increasing solution ~ e + t . 2TT_ Jq=0qb+l(lq)a+l 1 .( a + l ) p s ( p ) + 6s(p) dp 0.
—: <f h_nJ_[ dp = (a + c)\ T^ if c = o.—i. (11. 11. Jo (11. where Fnl(x) = yi(x) (11.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.112).11.6 The Discrete Spectrum of the Coulomb Potential 223 We evaluate the integral on the right with the help of the integral representation of the gamma function.(x). so that I = (c + a)! 1 f .125b) The normalization condition {J \tp\2dr = 1) determines Nni (which we leave as an exercise!): w "< = Ul^TW F (1L126) It is useful to know the following integral for certain cases: InmiP) ••= l^ exx^k^Lkn{x)^Lkm{x)dx. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . Clearly only the second form is sensible. and 0 if cj=b. . m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.6.127) . 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. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.125a) and the factor a~ 3 ' 2 with a = Bohr radius. is introduced for dimensional reasons.?). The wave function of the state (n.111). I. (11. Eq.
thus on the average the electron is the farther away from the proton or nucleus the larger n is.6(n + Z)[(2Z + ! ) . (11. This behaviour is indicated in Fig. The Coulomb Interaction One finds in particular (see also Example 11. For the ground state (n = 1.3(2Z + 1) + 2} 2 {2(n + 0 .! ] + (21 + l ) 2 .2. 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 + /„..128) These expressions permit us to obtain the mean or expectation values (1/r) and (r). 11.2Z .n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .7. + .8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .1 + 1) (11.224 CHAPTER 11.129) 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 .3fc + 2). (11.131) .130) We see that the mean value of r is the larger the larger n is.(l) n+l. and for the explicit derivation of the second case below Example 11. a (r)is = ~a. = 0) we have ) r /u =.v (6n2  6nk + A:2 + 6n .(2Z + 1) + 1} = ^[Zn21(1 + 1)]. 2 (11.
11.l ) n ] =  [ 2 n 2 + n] an[ n + (11.J n 1s 2s 3s *r Fig. S.134) *These are taken from I. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . 173/174. Sneddon [255].j . Problem 5.. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 . pp. / ! J — . i.20. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.132) (11.2 Hydrogen atom wave functions.2 I 3/2 Kl r 225 \2 1 R„.2 1 A r = V (^2) . In the case in which I assumes its maximal value n — 1.( n .11.133) for large r. v ^ V2n + 1 (11. or (r ) ~ n a 2 2 ^ [ 3 n 2 .(r) = x W 2 n + 1 =.e. \f{r ) ~ n a. with n +1 replaced by n and 21 + 1 by k. = l a nl n N'nll rFnl.6 The Discrete Spectrum of the Coulomb Potential r.^ We obtain therefore for the quadratic deviation (r) . we have and tp becomes particularly simple.
= ngh. The Coulomb Interaction or TT = ^ = (11. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler .135) This expression becomes very small for large values of n. for these values of I it is not distributed with uniform probability over the spherical surface. the electron remains practically localized within a spherical surface of radius (r).— > where ampler is the length of the semimajor axis of the elliptic orbit.e. This means. in which periods are obtained from the BohrSommerfeldWilson quantization condition. In order to obtain an impression of relativistic corrections in a simple context. (c) The nucleus was treated as pointlike and not as something with structure. 6. Thus. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. Example 11.4: 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).r.6.3.226 CHAPTER 11. r. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". p2 = p2 + — pg • eV. . relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). i. but rather like classically in a plane (i.e. we consider in Example 11. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. i = 0. pJ = —— and j> Pidqi = nth.4 the relativistic equation in cylindrical coordinates. \/v vT 7 v^ See also Example 14. and separating this in cylindrical coordinates r. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. Since 6 is a cyclic coordinate. the states Ytl are predominantly concentrated in the xyplane). However. we have pg = const. Solution: We have H = yjm%c4 + pIn cylindrical coordinates.
Whenever there is degeneracy. Schiff [243]. r\ = r ( l — cos#) = r — z.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. 11. (p). .11. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. y=y/%qsanp. 11. ip are the spherical coordinates.9. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 .136b) £ = r ( l + cos#) = r + z.(p . It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'.2. which are related to Cartesian coordinates in the following way. as also indicated in Fig.* See L. <p = <p. where the constants are identified by comparison.. 86. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines.3. where r = ^/x1 + y2 + z2. z=^rj). which correspond to a new choice of polar axis. 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. where ip = 0. y — 0: V ^ c o s tp. The deeper reason for this are symmetry properties of the problem.136a) (11. (11. 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.V .T). ( v c + i ) . p. it is possible to separate the Schrodinger equation in other coordinates.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».6.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. 2 2 Therefore the energy is no longer a function of n = nr + ng.
.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. 11.228 CHAPTER 11. In terms of the coordinates (11. We rewrite the first of these equations with the help of the relation M (w ^  m + e) . Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A.22) becomes — as we show in Example 11. . 1 . and r\ = const.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. so t h a t the variables can be separated. = 0. with zaxis as axis of rotation. there is really only one basic equation to solve here. The Coulomb Interaction Fig. (11. a separation constant. 2 .3 Parabolas £ = const.136b) the Schrodinger equation of relative motion (11. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . for uniqueness of the wave function. .
(11. " . . ds2 = dx2 + dy2 + dz2. . Eq. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. 2 . To obtain a closer analogy with the separation in the spherical polar case.105)) Hence we write I = (m — l ) / 2 .106) and (11. . (11.e.125b) (dividing by p 1 / 2 ) £ oc e .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. p = /?£ and cr = /^r?. .i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . . .± ( m + 1) = 0 . i. n 2 = n" . Adding now / . .' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).6.o. from Cartesian to parabolic coordinates. as well as the Laplacian A = V and the Schrodinger equation. . E x a m p l e 11. T h e n by comparison with the spherical polar separation the answer is as in Eqs. 1 . .11. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . In the spherical polar case we had (cf. . We leave the calculation of the degree of degeneracy to Example 11. we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". m0Z1Z2e2 ^ 2 = n (say).5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. 2 . 1 . .114b).
\— < / — cos ipdr] — \/£r] sin ipdtp.114c). when m = 0 = 77. the total number of wave functions we have is the number n with m = 0 plus twice the number with m ^ 0 starting from m = 1.2 .l)a + (n . i.e. We consider first the case m = 0. as t h e two signs of t h e exponential indicate.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. Observing that the highest possible value of m is n — 1 (i. {g^Y. for instance. see Eq. We then sum all possibilities and select the coefficient of ojn~1 in the result. etc.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . n the double sum ]T]2 is TI — m.2 = n — 1. . dx = d{ y £77 cos ip) = . N2 where g^dS.2) and summing over m. Thus we want to obtain the coefficient of wn~1 in £2 . The Coulomb Interaction Solution: We begin with z of Eq. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). i. (11. 712 and m which give n. In the case m ^ O w e have n\ + 77. We use the method of selector variables. i. are elements of length. .[df + dr]2 .. 2 _ = (9id02 + (gvdvy + /„ . n 2 (1UJ)2 The coefficient of a .136a). (n _ !) n = n2 §See e. Thus dz2 = . Oberhettinger [181]. 0 there are two. Similarly we proceed with dx and dy. / — cos ipdl. we have to find the number of values of the quantum numbers n\. (11. 2 2V? V^ One now performs the square of this and obtains similarly dy2.2)d} = (n..g.1) ~ .E E n l „ . </.1 in this expansion is seen to be n.230 CHAPTER 11. + l 4 {^^)drl f i + v \ .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. / J„^2 .2)d/2. last chapter. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .e. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. W. (11. . = e±im^e("+^'2{par'2L^{p)L^ (a). all cross terms cancel in the sum of 7 the squares. and so dz = (d£ — drj)/2.l ) ( n .136a). Thus. Solution: For the wave functions ip = £(£)J\f(r])&(ip). Here we have (see preceding text) 711+77. for m / .e. 2 + ^dV . Beware! For m = 0 there is only one mstate. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. Eq. > .2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 .2d? dr)}. z = (£ — rf)/2. " .tx2 . and attach a factor to to each term introduced in the wave function.e.g<p 9 \ . Magnus and F.
128) if one makes there the substitutions n —> n\ (or 712) + m. where F is the electric field.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. Thus e. I. With the help of the wave functions we derived above.e.. if we expressed the perturbation in spherical polar coordinates. with Eq. which will then give the change in energy of that state to the first order (i. .eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. these expressions agree with those of Eq.124c).•* n (l) = / Jo It follows that . and determine the average value of v for a given state. (11.) = ^eF(p . (11. Z2 —> e. 8.r.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk. eF{z). . In fact. 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. we can relate the expressions Iq to those of Eq. . Ikr+k.6).a).e.6 The Discrete Spectrum of the Coulomb Potential Example 11. i.g. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny.™(p)] 2 By contour integration (see Examples 11. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£.—r~n = 3(mn 2 ).1 (cf.drid<fi oc (p + o)dpda. = j (2ni+m+l). Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. However. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .11. with z = r cos 9.127) and setting n = r + k. L ni\ ny. s 0/ .128). with Z\ —> Ze.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. . {p — a) — = dxe~*xkvLkn (z)] 2 . Sec.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. . with the connection » (11.8 and 11. (11. fc — m. i.e. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. with RayleighSchrodinger perturbation theory). Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).(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.
the integral arising here becomes with Eq.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11. c positive integers.l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .3 by one power of t.128).9 we consider integrals with an arbitrary power of t.1)! n\ This verifies the expression I™t in Example 11.a. 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 . (11.9 )« Solution: Using the generating function for both associated Laguerre polynomials. tt„:_ (1 .n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 . s = at.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. (n + r . dt = ds/a.( 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)!.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .3).9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.124d) of the alternative integral representation obtained in Example 11. E x a m p l e 11. the integral / is the coefficient of pnqn in the expression _ / .111) /*oo /*oo „ — sam+l / e~attm+1dt= (m + 1)! ds = am+2 I ' Jo For use in the following recall that Jo «m+2 {i + X)n = ~ ^2(ir .p ) r + 2 [(lp)(lg)]"l+1(lpg). (11.b..8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. and is a special case of Example 11.p ) ( l .7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.232 CHAPTER 11. . ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) . _ „ . In Example 11.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .3: /"OO I := Jo et[L^{t)]2tm+1dt.g )(l.^~T^——^ ) = coefficient of (pn .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.pK — — —L .x ( n + m)! ( n .p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n .1 ) ! (n .p ) ( l . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . The Coulomb Interaction E x a m p l e 11.
6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.124d) of an associated Laguerre polynomial. c + b even) and/or j / = 0. .1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. so that = (q+fc±£) .11.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 «.e.f ].r ) ! a ' . c + 6 — 2r = 0 (i. (a) j / = 1. . we have to have y + (c + b — 2r) = 1. c + b — 2r = l (i.e.p) .e. i.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01. ° ^ + a ( l . — (cr)!r!(6r)! Kv.p ) ] .™( r > (1 + a)P r £^0y^w Jc+6_2r = . we have 2r = 6 + c = even. (lp)(lq) + a(l q)(l . 7£.?(t)t a e.111) of the gamma or factorial function. Using the contour integral representation (11.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .l ) * x coefficient of a* in the expansion of 7 ( a ) . c + b odd).^ ( l . i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .q)(l .a _ 1 = [1 + a ( l .a t [Lg(i).(i) = 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 . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.L. where (a+**£)! c j n Jo = .w(r) = and / a ) = r\(P ~ r .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. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.e. i.
^)(.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11.~* Hence (recall that 7£ c (i) = i!(—l) l x coefficient of a% in / ( a ) ) (a+^)!(/3a) r r in (£f^)!(^)!(^±£)! t J = ( a + b)\((3a) > 6! _ (a + ft)! = [a + lb + l ) a . cannot really — and this means in a rigorous sense — be applied in this case. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. R. also with regard to earlier literature. Since such a rigorous treatment is beyond the scope of this text. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). Taylor [267]. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.)!' f^)(. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. These results are seen to agree with the expressions 1^.1%^ 11.7. the standard principles of nonrelativistic scattering theory. . t We have the Schrodinger equation of relative motion of particles with charges Z\e. is the work of J. b\ V cited in Example 11. + (!±<)«. . however parallel to the Coulomb case. ^A very readable source to consult.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. we consider the traditional treatment here and refer the interested reader to literature in which it is demonstrated that the derivation given below finds its rigorous justification on a basis of distribution theory.7 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. In view of the slow falloff of the potential at infinity.
142) This equation is of the form of a hypergeometric equation. (11. .146) See. (11.. (11. b.. (11. .z).11. Magnus and F. for instance. A = const. . In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). with the asymptotic expansions (11. (11.^ W ) = 0.109) we can therefore write the solution ipr = Aeikz$(i>y.e. that we encountered previously.z)..+ ^  + . z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. for r — oo) • ipr = Aeikzf(r .M) = l + .138) becomes 235 ^A + k2 . what is the > behaviour of 3>(a.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr).141) Inserting this expression into Eq. W. Our question is now: How does ipr behave for r — oo. i. where ^.143) and v = ikrj.b. (11.87 under Kummer functions.136b).z) = Wi(a.b. a(a +1) z2 ] (11.z)).e.140) The simple Coulomb potential permits particularly simple solutions.144) *(a.140) and using (11. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.108) and (11.z) + W2(a.. According to Eqs. i. v = ik(r . (11.7 The Continuous Spectrum of the Coulomb Potential Then Eq. (11. pp.b. Oberhettinger [181]. 86 . z = rcos<9. „ az 1.
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) . Without t h e logarithmic phase factors.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.b. The Coulomb Interaction W2(a. J\") ci(kr~/\n2kr) (11.149) (11.151) with Eq.e.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo. > A = Aeikz[W1(i1. (11. if tbr were „ikr ibr ex el"z + f{6) r (11. = r cos 0.ik(rz))} =^ + ^d.150) ^(rz)]1"^ 1 + 0 T(«7) i.154) .z)T(l .^l*7 l + O (11.z) + la)T(n + b.153) The asymptotic solution represents t h e stationary scattering state (E = const. (11.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.i 7 ) (11.236 n\Zzab CHAPTER ^T(n '£ fo 11. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.152) Since z = r c o s # .l.kz+l^ktrz)) C .148) . with 2. i.7 ln*(rz)) k(r .151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].e.l. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.(11. a comparison of Eq. (11.a) (z) na)"~ r(la) T(ba) n\ ' (11.
The quantity /(#) 2 would then be a measure for the scattering in the direction 9. 11.11. (11.155) Since the gradient is proportional to the configuration space representation of the momentum. (11. also called phase anomaly.4 Variation of the observed differential cross section with 0. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. the current density is * h . do dQ K 12 K Fig. In other words. 2im.7.156) .n . The problem of the logarithmic phase. As a consequence of the masslessness of the photon.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity.Q hk mo v0. We define as current density the quantity h J := 2imo [^(v^)v(v^n. For the incoming wave ipi = exp(ikz). we see that this expression is the exact analogue of a current density in classical considerations.7 The Continuous Spectrum of the Coulomb Potential 237 the wave could be looked at as the sum of an incoming plane wave exp(ikz) and an outgoing scattered wave. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. z — — oo. that we encounter here is a characteristic of the Coulomb potential. 11. the effective range of the Coulomb potential is infinite..
i.e. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . for 0 —> 0. (11.154) we chose the prefactor of the incoming wave in (11. This implies that scattering takes place for the attractive as well as the repulsive potential.238 CHAPTER 11.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. That this formula retains its validity here in quantum mechanics is something of a coincidence. since in actual fact the Coulomb potential occurs only in a screened form. 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.159) as the Rutherford formula which can also be derived purely classically. is called scattering amplitude. in the current this would in any case contribute only something of order 1/r.153): 4fc2 s i r  The expression /(#). In Eq.152) as 1. We recognize the result (11. In reality this divergence does not occur. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). which yields the scattering cross section. 'screened ZlZ<2e2 cr/r0 ° .e. (11. We make a few more important observations. the square 7 2 ). We see that a depends only on the modulus of the potential (i.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. 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. thereby screening itself off from the surroundings. Correspondingly the other part of (11.
t). i. The wave packet is a superposition of waves ip(x.4. t 0 )^k(x.8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig. The particle velocity v is v = — . _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . t) = U(t. whose maximum moves with the particle velocity. the wave packet in the absence of a scattering potential. t 0 ) . which in the case of free motion (zero potential) are simply plane waves.yJ ^ j A k o W 727)3/2 e 0. (11163) (11. / dk 7^340(k)4(xxo. t0) = e.^ ^ % ( x ) . v0 = . (11. of continuum wave functions.i i f (''°>/Vk(x.x° ' r j .e. 2 (11162) where Vk(x) = ^ ( x .166) Let the incoming free wave packet at time t > t$. t 0 ). m0 m0 and V.164) Ek~Eko + {kk0)v0h. i.2 fr2 .8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.11.t) = e .t). 11. be given by ^ f dk AL. and if H is timeindependent one has the stationary wave function ^(x. (k).165) ^ ( x ) = £ fe Vk(x). A wave packet with momentum maximum at k = ko (with momentum p = hk) is therefore given by the following expression in which xo can be considered to be an impact parameter.161) where Ak0(k) ~ A(k — ko) and ^ k (x.e. i. (11. 11.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. and which provides the uncertainties arising in quantum mechanics.e. 2mo V + F(x) k (11. of the equation r fi.
ip) could be carried along but we ignore it. Then with k = ^(ko + k .165) ko • (k .171) The phase of fk(0. this solution > has the following asymptotic behaviour in which f(6. k = kez. From Eqs. i. (11.ko) k0 (11. (11.iB *(**°)/ R °'tj + 1 pikr e * .172) h h k0 {rv0(tt0)} (11.e. the stationary scattering wave.t o ) / V k ( x . ^k(xxo) ^k(x) ~ = e.k 0 ) 2 ~ Jk20 + 2k0 • (k .173) .161) and (11. The Coulomb Interaction In Example 11.154)) that for x — oo.163) we obtain </>ko(x .169) and (11.240 CHAPTER 11. (11. With Eqs.10 we show that the solution V'k(x) of Eq.^ ( * .ik .x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.<p) + Ol Akr l (11.x 0 ). can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.170) A ko (k) e .t) = J ^ A k o ( k ) e .e. Eq.167) by replacing the plane wave exp(zkx) by this scattered wave.x °Vk(x).xo.ip) is the scattering amplitude with respect to the incoming plane wave.k0) ~ k0 + and the expansion of E^ in Eq. i.166).i) dk (2TT (11.169) The wave packet in the presence of the potential is obtained from the free form (11.167) we have ^k0(xx0. (11. (11. (2TT)3/2 eik* + —fv(P.
y)eiko'Xoe'[fcor£.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.r')U(r')ip{r')dr' /4T7 (as we saw earlier) .^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.r ' ) . T h e scattering amplitude is t h e same as t h a t given by Eq.154). (11. with the energy of the maximum momentum of t h e wave packet.171).8 Scattering of a Wave Packet so that with Eqs. t) . (11.fco(tto)/ft]^0o)(x/xo.z Fig. 11.r'). Here ^ (x' — xo. .(«)/ ipko(x . A particular solution of the Schrodinger equation is — J G(r. however. U(T) = ^ V ( r ) . Solution: We define the Green's function G ( r . Adding a suitably normalized solution of the free equation.x 0 .5 Scattering of a wave packet with scattering centre O. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.5. r ' ) = 47r<5(r . E x a m p l e 11.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. '. i.r')l/(r>fc(r'). (A + fc2)t/.^y /k0^.11. r ' ) by the equation (A + fc2)G(r.167).174) r /k0(^.to) has t h e form of t h e wave packet in radial direction as indicated in Fig. (11.e.to).r')V(rW). and determine the Green's function G(r.x 0 . t) f. the solution can be written iMr) : (2TT)3/2 ^dr'G(r.*0ko (x . *. 11.
6 The contour C+ in the upper half of the k'p\ane.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. Thus we consider the contour integral taken around the contour C+. which is such that the contribution along the infinite semicircle vanishes. dk' ik'*r 2TT2 fc'2 .fc ) oo dfc' 1 d nik r 7rr dr / .242 CHAPTER 11.6. The Coulomb Interaction We find a set of Green's functions G(r. These are the socalled outgoing and ingoing Green's functions. Hence we write G(r)G+(r). e > 0 small. 11. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' .=k+ie/2k 1 d f. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r .O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. 11.Rek' kfc/2K Fig. one obtains another Green's function given by G_(r)= r eikr. G+(r) 1 d dk' lim e>0 TTT dr Jc+ fc'2 — (fc2 ~ik'r 1 d 2ni nr dr \k' + k+ ie/2kj ik r J__d_ 7rr dr 2ni N ^ residues k.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . . Imk . displaced slightly away from the real axis as indicated in Fig. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k)./2 dk'd(—cos 6)dip iki.
Before we introduce the general form of the socalled partial wave expansion.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. If we proceed from the radial Schrodinger equation. Then yW (11.176) " 0—A ~ (I + 1 + iy)<t>i = 0. (11.2Z + 2. i.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. (11. Then £^<fc + (2l + 2 £ = ~2ikr.178) (11. e. whose continuation to infinity is again given by (cf. under the stated conditions 2/07 = Z\Zie2 = —Mo.g.0 + W2(l + 1 + iy.140) (previously we solved this only for the bound state problem) § . i. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. 2ikr)] [Wi{l + 1+11.2l + 2.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi).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.9 Scattering Phase and Partial Waves 243 11.£). (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. z). Thus we proceed as above.z2 27fc 1(1 + 1)' yi = 0. z) = W^a. (11. b. . Eq. b. 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. (11.e. Eq. where Mo is a parameter in Chapter 16. 21 + 2.e.11. b. (11.21 + 2. z) + W2(a.145)) F(a.
+ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.183) r(Z + 1 + 17) := Reie. T(2l + 2) (2fcr)*T« (Alliy ikr .j syyll iy il( _ .180) ^ ' (11. The Coulomb Interaction Using Eqs. (2kryr{i" V 7 .181) so that eldlT{l + 1 .n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .«7) „ .l e*"" + T(l + 1 + H ) T(l + 1 .' .ry) = e"^r(Z + 1 + ry). (11.146) and (11. Then e2i5< = Reie Re~ J2i9 9 = Si.ii)' (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 .244 CHAPTER 11.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5. .
(11. i._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). For instance the poles of S are given by Z + l + z7 = . as we saw. with Eq.187) contains almost the entire physical information with regard to the Coulomb potential. . Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .e.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. where the continuum of the spectrum starts. considering I as a function of E extended into the complex plane.e. g = e«* = rSi + 1+ *V*ri (11. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type. (11. a positive integer. i. i.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7).11.e. i.. kr) the regular solution with the following asymptotic behaviour: r 0 (11.n ' .185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function)..114a) reveals).188) in the form y/E. considering I —• an(E).u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. One defines as regular Coulomb wave Fi(j.e. In the present quantum mechanical considerations I is. However. That the energy E appears in Eq. (11.1..139) n' = 0. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.2. 245 (11. .
(11.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. we obtain trajectories which are known as Regge trajectories.191) . 11. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. 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. Singh [252]. In these cases.7. we obtain the partial wave expansion of the scattering amplitude as follows." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. (11. 11. We shall return to these in Chapter 16 in the consideration of screened Coulomb potentials or Yukawa potentials and indicate other important aspects of these. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. The Coulomb Interaction and plotting these for different values of n. For the Coulomb potential a typical such trajectory is indicated in Fig.7 A Regge trajectory of the Coulomb potential. and thus in general.246 CHAPTER 11.
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.196) "See e. The potential disturbs this phase relationship (by delay or absorption or redirection).195) with the scattering matrix element S(l. r e.22. (11.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—.r ^ V ^ Wostf). The mathematical expression can be found in Tables of Special Functions (there cf.^2l + l^S^k) lik (11. It follows therefore that pikr ^scatt ^ f{0) = Aegeikz = ~ ^(21 + l)Lilneikr  rHe2iS<eikr\pl(cos9) eikr\pi(coa9) eikr " 2 ^ E ( 2 Z + l){e i ' .eikr\P^cos9). .~ J > Z + l)Lil*e~ikr . (11. 21 . 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.k)=me2iS^k\ in which r\i^\ indicates absorption.192) Thus the incoming and outgoing spherical waves have a definite relative phase. pp.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9. (11.g.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg . Oberhettinger [181]. 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„ ' . Magnus and F. W.i f c r .11.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.
This page is intentionally left blank .
that quantum mechanically a particle may have a small probability of being in a spatial domain. was developed only in the last two to three decades. Razavy [236]. in texts on quantum mechanics is rare. For various general aspects and applications of tunneling we refer to other literature. namely the instanton (or more generally pseudoparticle) method.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. M. can be found in any traditional text on quantum mechanics. however. i. the computation of finite lifetimes of a quantum mechanical state. These newer methods will be presented in detail in later chapters. It is not surprising. which in one form or another.g.Chapter 12 Q u a n t u m Mechanical Tunneling 12. In fact. which a particle in classical mechanics would never be able to intrude. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. i. Similarly difficult is. 249 . in general. It is very instructive. Such a vital quantum mechanical phenomenon is socalled tunneling. as — for instance — for trigonometric or double well potentials. states of a finite lifetime. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts.* *See e. that the explicit derivation of such quantities in nontrivial contexts. therefore.e. to explore first much simpler models in order to acquire an impression of the type of results to be expected.e. one of the main and now wellknown methods to perform such calculations.
Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.. i.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).t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. According to Eq. M^ Then dt + Vj 1=0. Quantum Mechanical Tunneling 12.250 CHAPTER 12. t) was developed by Born in 1926.t)\2 =1 (12. (12. r (12.t) = HiP(x.t) that the one particle concerned is somewhere in space at time t. ihr(x. .t) (12.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. so that p = T/>2 is the probability density for a spatial measurement at time t.1) as the probability described by ijj(x. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.4) (12.e.
2 .12. the equation can be integrated at every point. 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. This effect is known as "tunnelincp. 'Maybe the reader dislikes being confronted again with the onedimensional case. Delta potentials in more than one dimension do not permit bound states and scattering.6) . Thus it is possible that the probability for the particle to be in such a domain is nonzero. 29 . as shown by P. as we mentioned above. pp. 2m E 2 V0 > 0. If there is no absorption. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. t For the singular delta function potential a modification of these demands is necessary.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. However.33.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. Nonetheless. Gosdzinsky and R. If ^>(x) and V^>(x) are known at every point x. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. An important consequence of the wavelike nature of the wave function ip(x) is that it can differ from zero also in domains which are classically not accessible to the particle. (12. From this follows that both p and j have to be finite and continuous everywhere. 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. This case plays a special role. Tarrach [119]. 12. we can do this also here. the problem is analogous to that of the scattering process from a potential in a onedimensional case. which is the case we consider here. higher dimensional delta potentials do not have properties like that in one dimension. 1 0 ~x o a= f See L. where „ mpVo V(x) =+V0S(x).2a6(x)]il> = 0. ~hr> k = n^ (12 7)  . although in general small. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. Schiff [243]. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). The Schrodinger equation for this case is* ip" + [A. Such a region is for instance the domain where V is larger than the total energy E. In much the same way that we define there reflection and transmission coefficients.
e.: (i2iQ) Ce for x > 0.8) (12.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo. The continuity conditions at x = 0 are therefore for V : and according to Eq.ik(C . 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. (12. except in domains of the potential V. i.252 CHAPTER 12. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .11b) 2aC.9) for ij. a — ik £Qr < { eikx a eikx •t" . (12.= 2a^(0).2 a J 5{x)4>(x)dx = 0.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.e.e.' : It follows that ik .11a) ikC .[^']o. i. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). (12. a 2 — a 2 a — ik a2 + k2 ' .1) .e 1 + B = C.13) ~A R= for x > °(12. 2 (12.(ik .ikB) = 2aC or 2a(l + B). that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .ikC = i. .12a) (12. [<//]0+ . e Q (12.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. (12. i.
e.e «x = J \ CeKX for for x > 0. or h2 E = —a2.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. when normalized to 1. (12.ip'(0) = .12. as coefficient of the outgoing wave. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. = . and C follows from the normalization: OO A. 2m 0 (12. See e. possesses simple poles at a ± ik = 0. Proissart [223] or other books on scattering theory like R. the quantity T.i 8 ) .15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential.2 = .e. R.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. . Omnes and M.19) Thus the delta function potential supports exactly one bound state.2 o ^ ( 0 ) .) and V>'(0+) . i. at k2 = —a2. ( 1 2 .2 a C .(K)C = . Newton [219]. i.K 2 < 0. z<0. We write the solution ^ so that ^(0+) = ^ ( 0 . f Ce~KX c. 'Note that the physical pole has negative imaginary part. Thus ^ = Ce.K C . .17) da#(s) 2 = 1.g. K = a. oo (12.a l s l.
The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0. x G [a. 58. Quantum Mechanical Tunneling 12. Schwabl [246].254 CHAPTER 12. Fermi [92]. .a].22) + De Feikx + Ge~ikx *We follow here largely F. The latter contains also specific applications in nuclear physics.1 The square well potential.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.20) for for \x\ > a.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. p. V(x) V=E a v=y Fig. 12. \x\ < a. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.1. (12. p. (12. We want to consider the case of stationary states for E > 0. 55. or in the lectures of E. x€[a. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. 12. —a]. oo].
) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. (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.ikBeika = inCeiKa + DiKe~iKa.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 \ . .12.27) v ' Similarly Eq. M(a) .iKa ? ikAe~ika and .25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12. I (12. 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . (12. (12. fc Af(a) = .f )e" t\ K\0—tKa—ika .26) K^gifoa—i/ca fc. (12.ikGe~ika.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.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca. Fikeika .24) yields F ^ G.e. ( V C n vD )..
256 CHAPTER 12.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. Quantum Mechanical Tunneling Hence with the complex quantities IK (12.(1 + e 2 ) sin2 2na 1 .35) with with with x G [oo. xe[a.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.oo].—2ika [cos 2/ta + \e sin 2«a]e~ (12. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.e.31) we have A AT(a) = )MFo B ^r)sin2na . we obtain A = F( cos 2na — .a].e.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 . = < Ce~lKX + DeiKX .36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave. (12. (12.a . (12.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.4T(£)e ifci. i. \R\2 = T(E)^sm2Ka (12.a ] . (12.^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.34) into Eq.e sin 2na IJlika e B = r)F sm2na.32) [cos2«. .37) . x € [a. i.
44) . (12.12. and k real. i. (12. Dwight [81].e. B. 2m0\E\ h? ' (12.e.31): 1\T(E)\2 257 1 .e. an equation of the type f~1f = g19. which is satisfied if either / = g or / = — g'1. in the domain of A imaginary and K real. We see in particular from Eq. (12. p.e. (12.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. for VQ < E < 0. 2\k K. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. i. (12. (12. (12.43) Thus we set k — —i ' H .40) we obtain cot(na) — tan(/«j) — i K IK . Li ( — H— I sin 2na = 0.39) 2 i.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. (12. they have solutions.38) as expected. 83. However.4 Scattering from a Potential Well On the other hand according to Eq. formula 406. cot(fca) = i— or tan(fta) = K K .12. nor when both parameters are purely imaginary.42) These equations have no solution for K.35) and using Eq.41) i. .
t2 2 (12.42) which then leads to bound states with associated even and odd wave functions. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. T(E)e2ika = — r .l .+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .' We observe that T(E) has poles at those values of E which supply binding energies.258 and obtain the equations CHAPTER 12.fl = ( . . The amplitude of the transmitted wave assumes its maximum value where (cf. ± l . Quantum Mechanical Tunneling .e. (12. (12. 64 . t&n(2Ka)ER = 0.48) We expand T{E) around these energy values.46). Then according to Eq. (12.47) These states in the continuum are called resonances. pp. Also. F.20). and is itself infinite there. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . „ . .e. (12. (12. i. The remaining wave part becomes exponentially > decreasing. Schwabl [246]. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.50) See e.e. 2Ka = nn. 2m0\E\ and Kcot(Ka) = y—^—L. (12. i. (12. n = 0 .45) These are the equations that one obtains from the connection relations for the case of Eq. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12.g. At their positions cos(2/«z)£. (12.65. from Eq. This can be understood. ± 2 .c o . . = 2/rn (EER).34). i. We now return to the scattering states.35)) for E > 0 sin2Ka = 0.49) Taylor expansion of the denominator in powers of (E — ER) yields . 2m0\E\ Ktan(«a) = — y——^— .l ) n . Eq.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. (12. A = 0. . around the values given by Eq.
^ l / . . We restrict ourselves to the period — la < x < 2a.12.52) Here ER is the energy of the resonance and 2/F the lifetime of the state.' ''' 5a 3a \ a ~~~. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.e.2.51) i. 12. This result is physically very plausible. 12. a 3 a. . V(x) x=2a x= =2a . A potential of this type is known as a PenneyKronig potential. T(E) has poles at E= ERi: (12. = 0. s \ V0 /• ~\ 'X .' ' 5a \V / Fig. 2a<x<2a. 12.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. These are precisely the resonance states.2 The PenneyKronig potential. V(x) = 9(x + a)VQ6(ax).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.
. E^En = ^ ^ t n = 0. We set . with V as parity operator.2 = W h 2 _„2 = 2mo(E .. HV^(x) = H. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r.53) *w«w^( « j)'*^{i i ^ . In view of the periodicity we have ^(x) = ^(x + 4a). Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo.a ) = 0 = ^(a). and we choose (12. the differential operator is even in x..56) Thus the problem has a twofold degeneracy: Associated with every eigenvalue E there is an even and an odd eigenfunction and h2 n 2 7T 2 E — En > 2m0 4a 2 (b) Next we consider the case VQ > E.2 < a . . and hence for = Eip(x).260 CHAPTER 12.. i.e There is the additional boundary condition (for Vo = oo) ^ ( .2.Vo) n.e. . x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. (12. = EVip(x). Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.1. (12. i. The particle of mass TUQ whose quantum mechanics is described by the Schrodinger equation is free to move in the domains under consideration to the left and to the right of the central barrier.e. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x).55) The Hamilton operator of the problem..
58) D sin k(3a — x) with xE[a. (12. — a].2a]. 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. = Ka ±Asin(k2a). We expect. (12.. (12. ±Asink(3a — x) with x€[a. 12. xe[a.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka]. This wave function is no longer restricted to the interior of a box. V'(O) = const. we shall not exploit here.59) Asink(x + 3a) with x & [~2a. V+(0) = const.12. We > obtain even and odd functions ip± by setting ^ = ±1.60) The boundary conditions that ip± have to satisfy.2a. We therefore have different boundary conditions.il>]?0.63) = KBe~ F = . At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. BeKX±Be~KX with xe[a. — a]. At x = ± a we now have to connect ip and ip' of neighbouring domains. of course. The central region of the solution is the classically forbidden domain.Be Ka (12. (12.2).5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . (12. so that (W meaning Wronskian) W[tl>+. VV(°) = °. This means that i>±(x) = <  = ±1. which. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . however. that these are such that for VQ — oo they approach those of the previous case. are ^_(0) = 0.a]. ^kAcos{k2a).2a]..a].
F ^odd (12. (12.+ • • • 1! cos^(n7r) 2ak .mr)2}.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. i.nir + 0[{2ak . the odd solution with.67) The value of k belonging to the upper sign.— (1 ± 2e~2A + O (—). we set tanhx = l . 12.e. Taking the ratio of the equations in both cases..262 CHAPTER 12. say.e. (12.even. + 0 ( e . say A.3 Level degeneracy removed by tunneling. (12.68) . i. i. (with reinsertion of nir/2a for k) 2ak ~ nir .e. where Akoce'2™.66) Inserting these expansions into Eq. It is plausible therefore to use appropriate expansions. for VQ — oo. i. In general one now rewrites the equations in matrix form.nir)2}} = 1 ± 2e~2Ka.e.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . For large values of Vb.65) and we expand tan(2afc) about nir.2 e . we obtain k v ' eKa^eKa  tanh(«a) v ' These are transcendental equations for the determination of the eigenvalues of the even and odd eigenfunctions. = AA. k = fc0dd. is the one belonging to the even solution.2 a . However. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. (12. Quantum Mechanical Tunneling Fig.64) we obtain ~[(2ak k . the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—. and is — as we see — less than that belonging to the lower sign case. in the present case it is easier to continue with the above equations.nir) + OU2ak .
Classically the car is unable to overcome the bump.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. and is a reformulated version of a problem set in lectures of Fermi. . Solution: The wave functions i/>o>Vv in regions V = 0. 2mn. '.. as indicated in Fig. 0.70) xo = (a + b) 2' .v = 0.1: 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. 57.4 A car meeting a bump in the road.V{x)\i. 2m0E „ 2mnE .e. b > a. Example 12. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. V ^ 0 are given by the equations ^o + " J . 12. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig.E)da amplitude divided Thus with E ~ 0 the probability P is given by the square of the transmission by the square of the incident amplitude. . ^v + ^[E. Fermi [92].. but quantum mechanically there is a finite probability for this to succeed.3. Compute this probability. 12. i.** " W ~^7~ Fig. y(x) = sin and y(xo) = 1. .^ 0 = °' Hence in dominant order •00 — exp . exp . p.12. (12. w(x0) •• = = mogy(x).2 dxy/V(x) (12. .4. 1pV 2ro 0 4>o '.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. exp r rb rb / •J a 2mo (V(x) . 12.
71) The integral can be looked up in Tables of Integrals. one evaluates an approximate probability of exp[6. 14.4 x 10 3 9 ]. Gradshteyn and I. p.9191 ~ 1.2 .**. .72) Assuming a mass of 1 ton of the car (i.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.621.y) is the beta function (see Eq. 1000 kg) and using the following values of the natural constants: g = 9. S. formulas 3. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. (15.807 m s . n I . p. We can now evaluate the probability. Emde [143].1.17)). we obtain ( . ^ ) = 2 ^ .l .055 X 1 0 _ 3 4 J s .2.1 / 2 ) ! = v^r) TT/2 /. 950. Ryzhik [122]. Jahnke and F.e. : 2 ^ B ( ^ . which give /•TT/2 dx sin'M . h = 1.1. **For instance E. p.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. M. (12. 369 and 8.384. We find™ (with ( .264 we have ro / dxyfV(x) Ja CHAPTER 12. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. dx sin ' x •• Looking up the value of the factorial in Tables. and assuming a length of b — a = 100 m of the base of the bump.
the last case allows only a discrete spectrum.* In Chapter 14.Chapter 13 Linear Potentials 13. each of which propagates with the classical momentum and energy of a Galileian particle. we consider the linear potential in three space dimensions that forbids the particle to escape. in the book of G. 144. Finally. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.2 The Freely Falling Particle: Quantization Under quantization of the freely falling particle we understand here the solution of the Schrodinger equation for the linear potential. 13. for instance. Then considering the probability distribution determined by the squared modulus of the wave function. Our treatment in the present chapter aims predominantly at an exploration of the quantum mechanics of the freely falling particle in a domain close to its classical behaviour. we consider the corresponding case of a particle above the flat surface of the Earth. This is the quantum mechanical version of the problem of Galilei in one space dimension. This is the problem of Galilei in quantum mechanics. * A highly abridged treatment of this problem is given. whereas the first case permits only a continuous spectrum. Thus. An additional reason to consider the linear potential at this stage is its appearance in the following chapter in connection with the matching of WKB solutions across a turning point. Siissmann [266].1 Introductory Remarks We distinguish here between three different types of linear potentials. p. 265 . Example 14.3. 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. in Chapter 15.
13.t) = m0g with potential V(x. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle. (13. from d_ m x2 0 dx + V(x) 0.= l V 27T Joo dxelkxip(x. Recall the classical treatment of the freely falling particle. The equation of motion follows for instance from the derivative of the constant energy.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. We see " that the case g < 0 can be related to the reflection x — —x. x > 0. (13.1). In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x. A different > "linear potential' is the socalled "confinement potentiaF V(x. and yields rriQX = mog.3) In general the momentum representation is of little importance.= l V 27T J_oo 1 dkeikxijj{k. (13.t). (13.t) = .t) = . Therefore we use the Fourier transforms il){x.2) which is the case with a discrete spectrum.1) where g is the acceleration due to gravity and we can put Vo = 0. V(x) = —mogx. c> 0. (13.e.266 CHAPTER 13.4b) . (13. We consider here the case of Eq. The present case is an exception and a good example of its applicability.2. i. g > 0.t).4a) f°° ${k.t).t) = mogx + V0.t) = c\x\ + V0.
(13.t) the expression (13.< (13.t) h2k2 d ~J(k.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13. We rewrite this equation in the following form th (13.t). we obtain the differential operator on the left of Eq.7) applied to every quantity in Eq.13. • fd ~(' m0gt %h .t). i.t) m0gx /2TT m0gxip(x.3) yields us now by inserting for ip(x.t) dk nikx k=oo '2K (im0g)il)(k. Applying the total derivative d/dt to this function. Equation (13.9) . The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.t)im0g^k.t) dketkxiP(k.6) and wish to solve it.8) With this result and the substitution (13. (13. multiplication by a test function and subsequent integration). To this end we consider the following substitution in Eq.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.6). (13. (13. (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13.6) the argument k is then replaced by this.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k. d \ ~f1 dkr[ k + m0gt ^r't (13.7) In the wave function on the left hand side of Eq.e.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.e.d m +imo9 .2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq. 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.
i) .11) in a neater form. (13.6). On the other hand the second operator expression implies dtj>o imog^—ip(k.6).t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13. we have to reverse the substitutions and replace in (13. Solution: We apply the two operator expressions on the left of Eq.t).t ) ' U ( f c . E x a m p l e 13.6). t ) . the following result h 2moi tp[k\ ^—. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.11) solves the partial differential equation (13.1 we verify t h a t this solution indeed solves Eq. i.gh [ d t ' j f c + ^ ( t ' .11) *[k?f. (13.t In Example 13.6) is then $(k.10) Note the amplitude function ipo(k) in front.6).1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.= k rmt.10) k by m0gt back : k —> k h The solution of Eq. Thus addition of both indeed leaves the expression on the right hand side of Eq. Our next step is to reexpress the solution (13. (13.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.12) . We have thus obtained the solution of the transformed equation.11).t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. (13.6) to the solution (13.e.268 CHAPTER 13. To obtain the solution of the original equation (13. (13. We define the wave number kt as kt. Thus we have first (the second term arising from the upper integration limit.
(13.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.o o 2 —  exp iktx h mZgH' .t) = .11) can be written j'A ^.t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . 27T ^ .16) Next we evaluate the Fourier transform of ip(k. Inserting ip(k.t) defined by Eq. (13.+^f (13.19a) . we obtain rp(x.t) into this integral and recalling the property (13.18) h =k mogt p h S' h .4a). .gfci + 2 g2m0 2/i 2 ' (13. m0g (13. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .17) = J_/ C X ) h2 [It(k) .13) rn0_t+2 + mlg2t3 .13. m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.} fA ' _ _tkt 2 k+ t) 2 k .2„2+2 mfaH 3^3+3 .o ut = 2mo —kf 2mo A.14) (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential. _ k t  h2 3 Inserting here the explicit expression for fct.11). .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.15) we obtain / = . we have ip(x.— [It(k) ..I0(k)].
These wave vectors vary in accordance with Eq. To this end we convert Eq. Such a decomposition is possible only for homogeneous force fields.o o Jo This result can be interpreted as follows. (13.270 and CHAPTER 13.2.19a) exactly as for a falling particle.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .20) (13.20) expresses the wave function tp(x.ut> = Jo t2 g2m0 i 3 k t .21) In other words.t) as a decomposition into de Broglie waves.19b) h3 3 we can rewrite ip(x.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x. 13.v = kh/mo.t) as 1 I00 ip(x. (13. (13. With the help of the de Broglie relation this wave vector corresponds to the particle momentum Pt = P — rnogt or to its velocity vt = v — gt. 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.23) .The associated kinetic energy of the particle is mv+ = huJf.18) into a • different form in order to be able to perform the integration with respect to k. 7 u>t'dt' (13.t)\2 for t — oo (distant future or past). the relation (13. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.t) =y= dkip0(kt)) exp i< fctx V27T J .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 . fm0gt\ \ 2h ) (13. Linear Potentials f dt'.
/oo dkip0(kt) exp l x ixI exp oo .t) = 1 1 oo 271 27T /.13. (13. (13.26) d£e ie £ kt^ m0' 0.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ . ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. /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.QX 2 ^ 0 7 (m. m0xcl ht m0gt 2h gt O v: °° 2hti (13.18) thus becomes ip(x. fi > 0. Then ip(x. being now independent of £.28) (for a verification replace a by a + ifi.29) ./m 0 a.24) We now set fit / 2m 0 V" 0 V m0a.t) 1 exp r . It follows that ip(x.e. (13.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. rnogt m0g2t3 tX^r~. 2h h 24ift ' mogt^ r m.26) with the formula 1/2 (13. We evaluate the Fresnel integral in Eq.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i. may be put in front of the integral. integrate and then take the limit fi —> 0).2 The Freely Falling Particle: Quantization The integral (13. TOO#A ft and.
this expression can be brought into the following form: if *\ 0 ip{x.31) and obtain from Eq.e. These are incoming and outgoing waves.t) ~ W —exp i V irlt m ktxd J cot. kt and ut 13.dA 4>o(h (13. Linear Potentials Replacing here x by the large time value x c i.t) = eiEtlhct>{x) (13. V(x)=m 0 gx V=E Fig.3 Stationary States In this section we show that the Schrodinger equation with the potential (13.1) has only a continuous spectrum.272 CHAPTER 13. 13. (13. (13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . xc\).33) oo for E <C \mogx\. 1/2  ±^(2mlg/h^x^ / 2i (13. For x exponentially increasing and decreasing solutions are possible.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. For stationary states we set xl){x.1 A turning point at x = —XQ. .30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.
e.t) = dt (13. t) = — = / V27T J .36) m0g \ 6m0 with c = const. 1 f01 dketlcx^{k. cf.4b). exponentially decreasing. i.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. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation.t). 13. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. i.e. However quantum mechanically this is possible with a small. but also d 00 dt Eijj(k. Fig. is reflected. E)dk. The point — XQ. The Fourier transform is given by the relations (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. and hence is continuous and extends from —00 to +00. and we can establish for the resulting continuum solutions the orthonormality and completeness relations.0 0 dkelkxE^(k.t). t). t) = Eil)[x. (13. H2k3 4>{k) — cexp Ek (13.34a) Then d &k 1 f°° ih—il>(x. i. Akx (13.e. probability. at which the classical particle bounces back.1. J \ „ (13.e.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. we .4a) and (13. which is not possible for real particles. i.34b) It therefore follows from Eq.35) It follows that /W tm0g \ 2m0 i. i.13.3 The TimeIndependent Schrodinger Equation 273 Classically the particle coming in from the right with energy E cannot penetrate into the potential barrier since (where V > E) with conservation of energy E — V = p2/2m. so that ih dtp{k.e.e.00]. at which E — V = 0. its momentum p would be imaginary there.
M *K Joo Setting dt cos ( ts n Jo st).38) we have for <j)E{k): OO /"OO / oo (13.e.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# . (1340) i. In a similar way we can verify the validity of the completeness relation.41) J—oo Next we insert (13.43) y=2^x+2^E. h? h2 . (13.L dxeikx(f>E(x). the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13. (13.. </>E(x) = . apart from a constant phase which we choose to be zero.e.a : ' ) .39) dk4>*E{k)4>E.36) into (13. i. (13.40)) into (13.L [ dkeikxMk). Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf).38).£') = <*(#£')• 1 27rm0g' = /o . so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). 13.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.. / dE(j)E(x)(t>E(x') = ^ s .37) (eikxdk.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 . / <>~ikx VZ7T J (13. V 2lT J Mk) = .36) (with c replaced by (13.
3 The TimeIndependent we can write <J>E{%)'4>E{X) = Schrodinger Equation 275 2ir^/m0g J dkexp h2 yk — k '2m20gV~ 3 W i t h the substitution k = [it we change the variable to t. (13. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.45) One can now derive normalization and completeness integrals for the Airy functions..^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= ^ . the function 4>E{X) i and setting s y = — = 1 ./ 2m 0 E (13.J * . / s M .46) This follows since with Eq.^ . For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. (13. whereas the other is of exponential type. and we shall see t h a t one branch has periodic behaviour. .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.44) is then expressed in terms of the Airy function: .13.
TT < 6 < 0 for r > oo. Thus we must have sin30<O. (13.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0. (13. (13. i.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . (13. Linear Potentials 13.43) contains the phase (with x — z): > *(*) := sz ..2 The contour C for s < 0 with ends in the angular domains (a) and (b). 13.f dzJ*W 2TT JC with z = reie.276 CHAPTER 13.. . i. (13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.r 3 sin 30 ). or n <6 <TT o for r > oo.TT < 30 < 0.e.51b) . .e. We want the integral to exist and so desire that lim e**^ = 0.\zz.49) (13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 .4 The Saddle Point or Stationary Phase Method The integrand of Eq.r 3 cos 30 J + i ( sr sin 9 . o (13.
Then the ends of the contour lie in the domains specified by Eqs.55) s < 0. and the main contribution to the integral comes from the a region around the saddle point.2. (13. —i ${zs) = szs .53) Since we shall have the contour C in the lower half plane. Such a piece can be found in the neighbourhood of the socalled saddle point zs. i.51a) and (13.13. i. where the derivative of the phase function vanishes. we choose zs = iy/^s. Inserting (13. (13. Then 6S = ±7r/2.57) $"(z s ) = 2% s. provided its ends at infinity satisfy these conditions.54) i ¥~s)3/2' (13.48). (13. (13.51b). and we set with a real: z := zs + a (S)V4' (13.45) exists for the contour C. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.52) = exp(±i7r) Let s be real and s < 0. 0 = &(z8) = s~zt s = za „2„2i6>. oo] as indicated in Fig.e. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.e.56) 2n(s)^Jc But I daefr"?*"^. at which the phase becomes stationary.z3s and V^s + . 13.S )V2 (13. (13.( \ / ^ ) 3 2z.4 The Method of Stationary Phase 277 The integral (13.58) . we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. so that exp(±2i9s) — 1 and — s — rl. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo. Then Ze TeC : V^se±i7r/2 = ±i\fzrs.54) where a is the new variable with —oo < a < oo.55) into (13.
45) 4>E(X) ~ Ai(z).44) and (13. (13._ 1 .2/32u = 0 (13. Gradshteyn and I.32) therefore becomes dz2 + zd) = 0. S.42) into an equation with a more appealing form. V 2h2 (13.60) The reason for describing the point zs as a saddle point is that the integral in Eq. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.. Ryzhik [122]. For (—s) — oo the integral converges towards T/TC.44) we can convert Eq. cos . We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. 8. 2 s 3/2 . . Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here.278 CHAPTER 13.32) with the solution (13.59) In Example 13. (13.56) decreases exponentially around a — 0 for real values of a.64) tSee I. Sec. This is an important aspect in the WKB method to be considered in Chapter 14. M.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). > so that Ai(s) 2 v ^ ( . where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. (13. With the help of the substitutions (13. as one can see from the following equation^ ri' + Ffz •2^2.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). (13.491. V^s / 4 V3 (13. According to Eqs. (13.2 we show that for s — oo: > Ai(s) 1 .s ) V 4 exp \(s) 3/2 (13.
It follows that for s —> oo: Ai(s) ~ . * " ( z ± ) = =F2Vi.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.59) and (13.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).51b) are satisfied.z ± ) 2 * " ( z ± ) .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._ . we obtain f° Joo dp=—[a+ i\a\]da. (13.. Thus there are two symmetrically placed real saddle points on the real axis. changing to the variable p. Example 13.51a) and (13. with *"(z) = 2z. 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 .s 3 / 2  . cos  .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. 27T Ic Jc 3 This time s is real and positive. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = . Then. Eq. the conditions (13.13. Choosing the contour C to lie along the real axis.^ e ^ / V2 4 [°° Jo dae'2^. P Introducing a new variable a by setting p = ai\a\. . Hence the saddle points are given by 0 = * ' ( z s ) = s — z2.
. i. since f'(zo) = 0 and (z — ZQ)2 = —x2. and hence integrating as described above through the saddle point indicated in Fig. 13. Then.Be 2 Fig.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix.280 CHAPTER 13. ~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq. 13.3.111).e.3 The saddle point at z = n + ix. n!= / e~zzndz= / e~z+nlnzdz. / " ( z o ) = —n/z2 = —1/n.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. z o = n . ^/2KJW1. Thus here / ( z ) = — z + n l n z . (11. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en.*. Solution: Briefly. Linear Potentials E x a m p l e 13. Jo Jo \ .
Dingle [70]. 281 . i. of misunderstandings and other aspects may be found in the monograph of Dingle.e. Jeffreys [144]. Brillouin [37]. H.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. It goes without saying that such comparisons are very instructive. Kramers and Brillouin.* The method had. N. ^ The central problem of the method is the topic of "connection formulas". Froman [99]. 316 . Kramers [153] and L. Wentzel [282].317. The WKB result is therefore frequently described as the semiclassical approximation.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. 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. been developed previously by others in different contexts. Froman and P. the linkage of solutions above a turning point to those below. will there be that of a coupling parameter. O. A. named after its first promulgators in quantum mechanics: Wentzel. B.Chapter 14 Classical Limit and W K B Method 14. where *G. *R.§ 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. § See e. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i.g. for misunderstandings pp. 292 . and the role played by h in our considerations here.e.295. For the history see pp. A critical overview of the history of the method. however. ' T h e most prominent earlier expounder is H.
Classical Limit and WKB Method E — V changes sign). (14. 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. For stationary states ^/.V(r)]^(r) = 0.t) = eiEt/h7P(r) and \V(r. In this equation we set.t)\2 = \iP(r)\2. A(r) and S(r) being real functions of r. V>(r) = A(r)eiSWn.3) .7. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . for which the observation of position and momentum is independent of the time at which the observation takes place.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. the energy has a definite value E with V(r. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.g.282 CHAPTER 14. the derivation in Example 18. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition. 14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. the function S being called phase function.1) (14.
4) (14.1) and separating real and imaginary parts. we obtain the following two equations: E .+ VAVS + AAS = 0.t)}2.Q A 2i 8A 1 (14.6) 2mo dt now with A also a function of t. h2 ih— = H$ A + V(r) * .10) m 0 ^ . (14. (14. We set in this equation. (14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.13) .14.* VAVS+AAS + V(r)A.12) VA A (14.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.V(Y) + and 2VA • VS + AAS = 0. (14.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *. (14. These equations are equivalent to the Schrodinger equation. (14.t)eiS{r't)/h. V(r. We can identify Eq. h h (14.10) with a continuity equation by defining as probability density p:=*** = [A(r. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.
^ + V • J = 0. TTIQ (14. (14.10). mo <™> (14.21) . (14.17) With Eqs. (14. (14.20) With this we can rewrite Eq. we have mo—(A2) + V(A2)VS + A2AS = 0.16) BA 1 moA—. .9) for the phase function S.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS.. ie.18) ot ot This is the equation of continuity that we encountered already earlier. Next we investigate the significance of Eq. (14. the implications of this equation are also those of the equation of continuity. (14. (14.n. + h2 AA V = — ^ . Since we obtained this equation from Eq.284 CHAPTER 14.9) as 9S 1 mo^ 2 + T. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.01. m0 But now (14.15) and (14.16) this becomes m 0 1 rr + m o V .J = 0.10) we obtain = — [V(A2) • VS + A2AS}. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. Since J = — A2VS = PVS. (14.14) A2VS.+ AVA • VS + A2AS = 0. (14.
(14.e.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.20). where. 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. Such series are typical for expansions in the neighbourhood of an essential singularity. r o t v = ( l / m o ) r o t g r a d S = 0.2) for the probability amplitude. a singularity different from that of a pole). This result is very general.V ) v + V 7 = 0. We deduced this equation solely from the phase S in the limit h — 0. . with the help of Eq. dt 2 Constructing the gradient of this equation. V • J = 0. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. that the quantum theory implies an essential singularity at H = 0 (i. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). The particles have no interaction with each other. However.14. they are asymptotic expansions. This expression with h in the denominator shows. This means. In the first • place we require the ansatz (14.e.1/h. and hence''' ^(m 0 v) + m 0 (v. which for very small values of h yield the dominant contributions.23) (14. the wave function \I/ describes effectively a fluid of particles of mass mo. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.22) (14. We see therefore that in the limit h — 0 the particle behaves > like one moving according to Newton's law in the force field of the potential V. The equation of continuity then describes the stationary flow of a fluid.24) d 1 dr d •— = d Lv _ . (14.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. so that V ( v • v) = 2(v • V ) v + 2v x rotv. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. d — = (14. i. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. we obtain ^ V 5 +  v ( v v ) + V V = 0.
(14. Dingle [70]. 318. Kramers and Brillouin. i = eiW(x^h. We consider the onedimensional timeindependent Schrodinger equation y" + ^. 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. 6.[EV(x)}y = 0. B.5)) and 2 ^ f dx ax + All=0. . (14.4) and (14.28) The calculations we performed at the beginning imply the following equations (compare with Eqs.1 The W K B Method T h e a p p r o x i m a t e W K B solutions Without exaggeration one can say that (apart from numerical methods) the method we now describe is the only one that permits us to solve a differential equation of the second order with coefficients which are nontrivial functions of the independent variable over a considerable interval of the variable.26) We observe already here that an expansion in rising powers of k. We set. The expansion in powers of h2.27) (14. such as "LiouvilleGreen method' and "phase integral method'. is an asymptotic expansion.30) ax *For details see R.3. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V). Classical Limit and WKB Method 14. Messiah [195].3 14. similar to our earlier procedure with A and S real functions. i.286 CHAPTER 14. Sec. There are also other more specific descriptions.In A(x).2. (14. z (14. As remarked at the beginning. the method is widely familar under the abbreviations of the names of Wentzel. p.§ y = A(x)eiS^'h where W(x) = S(x) + . §We follow here partially A. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large.e.
36) Equation (14. . 3 / 5 .35) .4 (S>)2 1 S'" 2 S> (14. (14.14. (14.33) 3 (S")2 .\ In S'.3 The WKB Method 287 Equation (14.V) and (14.2rn0(E V) = h2 lS'A 2S'0 (14. Li This expression can be substituted into Eq. (14.28).2mQ(E . Eq. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .31) \a.A = const. + (14.35) can be integrated.32) s" = sz + h2s'( + •••.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .10)). Since the expression (14.30) corresponds to the equation of continuity (i. (14. 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.e.32) is therefore (S'0)2 + 2H2S'QS[ . (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.29) becomes (note that this is still exact!) (S')2 .V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .
and instead is asymptotic which we shall not establish in detail here.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .19).40) rx dx v "7 W)\.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.27) we obtain now with Eq. (14. The solution is seen to be periodic provided k(x) is real.41)? We note first that the expansion in powers of h2 does not converge. E > V(x). (14. Eq.37) For Eq. Classical Limit and WKB Method has the form of a momentum. we expect that in any case So » ft2Si. (14. (14.e.and So = ±h / — — A J k(x (14.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. Chapter XIII. . y/2mo(V(x) .E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.288 CHAPTER 14.39) and (14. it is reasonable to set (cf. 1 S e e R. p = h/X) k{x) :Then S'n +. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). (14. i. Dingle [70].' For the expansion (14.39) where a and /3 are constants.33) to make sense as an asymptotic series. (1.38) h y/2mo(EV)' (14. B.
Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .3/2 s&T + 3 {So Il« we have (cf.36).36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.1 / 2 ]" [§0SS). (14.45a) .3 The WKB Method We have 289 S0 = ±h and Si follows from (14. E > V(x).42) [(^). Since k(xy 1/52\S'0 (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.V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.14.
(14. Classical Limit and WKB Method we can rewrite the inequality (14. In the neighbourhood of E = V(x).3. Roughly speaking one therefore requires E to be large in both cases.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh.46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. and hence (with UIQ = 1/2.48) behave in the neighbourhood of such a point called "turning point'. around the classical turning point. *2 A(z) = V / 2m 0 (£ .V(x) = V(a) . V(x)) (14. h = 1) .290 CHAPTER 14.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l .V{x) ~ Or . (14. (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. i.a)V'{a).e.48) The (dominant) WKB approximations are exact solutions of this equation. In terms of X the original Schrodinger equation is y" + = y = 0. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. the WKB approximation is in general invalid (see below). How do the solutions of Eq. at which E — V{x) changes sign? We note first that in this case near x = a: E . In a similar way we have for E < V(x): l'(x) <C 1.2 Turning p o i n t s a n d m a t c h i n g of W K B solutions Let x = a be the value of x at which E = V(a). 14.
14. which connect one domain with the other. (14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a.3.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq.1.48). 14. in the domain E > V the WKB solutions are oscillatory. in the domain E < V the WKB solutions are exponentially increasing or decreasing. 14. On one side. and on the other side. We shall not derive these conditions here in detail but rather make them plausible in Sec. V=E Fig.1 there are WKB solutions on either side of the turning point and some distance away from it. We assume we have a turning point as indicated in Fig.1. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. (14. 14.e.3 after stating them first here. 14. As indicated in Fig.1 The regions around the turning point at x = a. i. since this requires different approximations there. 14. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for .3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. 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. The transition is provided by matching relations.
and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. in order to make the formulas independent of whether the potential is rising or falling.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 / —.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . 14. (14.52) 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. This means the asymptotic part suffices only in t h e exponentially decreasing case. Note also the extra factor 2 in yi. has the same asymptotic behaviour in the domain i « a a s Ay\.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a .50) x S> a.H— J =^ v/exp + m (14. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. (14. however with the limits of integration x. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right.292 CHAPTER 14. 14. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. as in Fig. a interchanged. Thus. one could simply replace everywhere /•a / pa • • • b y / Jx J x .
14. i. whose exact solutions are the dominant WKB approximations. If higher order contributions are to be taken into account.a).3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. (14. (14. 2 1 X (14. the equation becomes approximately y" = (xa)xiy.14. 14.3 The WKB Method 293 In that case Eq.3.46).Airy overlap* Fig. the equation V approximately linear around turning point V=E Airy solution WKB . the matching relations have to be altered accordingly.51) — always with the decreasing exponential on the left — is valid in both directions.48). is singular at a turning point x = a. as we saw. y" + Toy = o.2 The overlap regions around the turning point at x = o. (14. 1 ~ (x .e. Therefore we consider now this original equation in the domain around x = a. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. One should remember that the matching relations above are those for the dominant terms in the WKB expansion.53) Since at x ~ a.54) .
— a ^> 0. With the substitution z = (x . A second solution is written Bi. A.Bi(—z) for \z\ — oo.60). p. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~.294 CHAPTER 14.291.57b) To the same degree of approximation as the equation y" = (x — a)xiy.59). formulas 10. R. as > may be seen by referring back to Eqs. 448.^ 4 I 7T x — a <C 0.5960. p.a)X\/3. (13. Cf.1 / 4 by \^. J (X (i4 58)  Replacing in Eqs. (14. Abramowitz and I. In the following we require the asymptotic behaviour of the solutions Ai(—z). (14. . B.56) We have discussed some aspects of solutions of this equation in Chapter 13. (13.4. we have r x * f x ' /2(x a)i/2d*=^(x . (14. Dingle [70]. In particular we encountered a solution written Ai.57b) UV2 3 11 by J dx .55) the equation becomes the Airy differential equation d2y dz2 zy. This is given by the following expressions.57a) and (14.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. (14. i I 772 \3 + > for ia>0. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. Stegun [1].a)3/2 ^ /2 =^ 3/2 o Q. > for a. (14. or M.
e. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. we enter the domain of validity of a WKB solution.2.3 The anharmonic potential well. by Ai(z).3 The WKB Method and 295 •. Example 14.. 14. however.51) and (14.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = . As an illustration of the use of the WKB solutions we apply these in Example 14.52). are only valid in a small interval around the turning point at x = a as indicated in Fig.(zf2 by J we obtain the matching relationsX (14.l v/2m0(EV(x))dx 7r = (2n + l). In this sense we have now verified these relations.l. In this way different asymptotic branches of one and the same solution are matched across a turning point.59a) . (14. These solutions.2. 14. As we approach a limit of this interval in going away from the turning point.1 to the quartic oscillator. n = 0.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.g. In the direction to the left as indicated in Fig. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). and hence must be proportional (in view of the uniqueness of the solution there). We summarize what we have achieved.14. 14. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. The latter domain is a small region where the solutions from either direction overlap.
l. . 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. Then around these points the potential would be well approximated by a linear potential (as we saw above). . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator.(x) = c ' v Acos I / — I for a < x < b. (14.44) this implies large values of n. 2 n = 0.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). (14. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. (14. (14. .. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.. (14.59b) Solution: We consider an anharmonic potential well as depicted in Fig. Sometimes this distinction becomes imprecise.2.3. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b.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 . b are the two turning points. 14. extends only over a length of very few wavelengths and is therefore valid for n small. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. According to Eq.e. The condition for this is ya{x)=yb{x) for xe(a. 1 . Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator.296 CHAPTER 14. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. i. as in the case of the harmonic oscillator itself. 2 . Here we are interested in the discrete states (the only ones here). Let E = V(x) at x = a and x = b.^ .3. n = 0 . 14. Classical Limit and WKB Method where x = a. 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 ( . 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. .b). j/i. We argued earlier that the WKB method is suitable in the case of large values of E. /n = s u ^ TJcos^ya TJ+cos^a T J . ^ A = (2n + l ) . . In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.60) Using the matching relation (14. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .y o dx TT T provided 6 _ 4 /.51). regions 1 and 3 in Fig. Hence we search for solutions which are exponentially decreasing in the far regions.
. B. before Heisenberg's discovery of the canonical algebra and the formulation of the Schrodinger equation) this condition was always given as §pdq = (n + l)h. and knows of no published form or script. .14.4 Three cases with different pairs of zeros. (14. 2. This relation is sometimes wrong.1. In this socalled "old quantum theortf (i. . Bohr. with n — 0.63a) or. 14.63b) *The author learned this in lectures (1956) of R.1. pdq = n + 1 — . (c) V Fig. The corrected form* is.4 BohrSommerfeldWilson Condition 297 14.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N.x number of turning points TTH. equivalently. (14.x number of turning points h. Wilson by supplementing classical mechanics by Planck's discretization. 2 . ri2 I Jqi :pdq n + 1 — .64) (14. Dingle. . . Sommerfeld and W. A. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0.e. n = 0.
if /(g) is negative. (14. together with q — — oo.298 CHAPTER 14. Case (a): The case of two trigonometrical zeros. In quantum mechanics we have dq2 + Ki/j(q) = 0. as illustrated in Fig. (14.65) We use the knowledge of these solutions to find the eigenvalues of the equation. Jqi ^ fpdq J =2 pdq = (n + l)h. i. where /(g) is positive. this is an exponential zero. where n —. i. this means dqyf—f = (n + l)7r. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. where /(g) is negative. 14. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h.e.4(a).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 (14.66) J<1\ Thus in this case the socalled old quantum theory is correct. Since sine and cosine have zeros spaced at intervals of TT.e. The WKB Method One pair of solutions is (as we know from the earlier sections of this chapter) tpT(q) oc 7T7Iexp 1 sin (_J)l/4 Cos 1 = JdqVJiq) F if f(l) is positive. 0(g) ex .51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. (14. this is a trigonometrical zero. There are three cases. the case of tp(q) — 0 at q = gi. Before we continue we recall from Eq. We have the case of trigonometric solutions.67) Case (b): Next we consider. / • n an integer > 0. In this > case the wave function for g > go is .4(b). 14.
^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. Since both cases refer to the same region qo < q < q'0.63b) to obtain the eigenenergies of the quantized harmonic oscillator. i.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1).2 and 14. Solution: As in the case of the quartic potential. i. 14.e.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . (14. For the wave function to be exponentially decreasing to the left of qo and using Eq.63b). Hence we have pdq= i n +  .63a) or (14.69) pdq = (n + l ) " 4 h. or Pdq (n + l ) " 2 h.70) We can therefore summarize the results in the form of Eq. Therefore their arguments differ only by (n + l)ir. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.4(c).3. Thus in this case the socalled old quantum theory is wrong. As illustrations we consider Examples 14.4 BohrSommerfeldWilson This will vanish at q = q\ if rii / V^f{q)dq+j^ Jan It follows t h a t i 4 Condition 299 = {n+l)TT.e.2: The harmonic oscillator Use the relation (14. and • • for the wave function to be exponentially decreasing to the right of q'0. this is the case of two turning points. Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. E x a m p l e 14. (14. the sines must be proportional. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . (14.14.68) (14. ri\ / J do y/f(q)dq (n + 1) 1" (14.
t Hence we have the case of the integral J> pdq (n + 1 )  Inserting the potential this becomes fqo=E/™0g 2 / dq^2mo(E Jq: Jq=0 Evaluation of the integral yields the energy 1/3 2/3 — mo m) • (n + 1) En = ( — m0g2h2 n+ • (14. . so that E = hu{n + 1/2). permit this although with rapidly diminishing probability.28 x 1 0 . Example 14.63 x 10~ 3 4 J s . and h = 6. as illustrated in Fig.807 m s . in principle. 4 / x CHAPTER 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. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B .5. where g is the acceleration due to gravity.V(q))dq The result is E/u.3: A particle in the gravitational field A particle of mass mo is to be considered at a height q above the flat surface of the Earth.300 With potential V(q) = 2ix2mov2q2. g = 9.2 . 14. Calculate its quantized energy.2 3 J x (n + 3 / 4 ) 2 / 3 . one obtains E ~ 2. whereas a turning point does. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. 14. :: V=m0gq Fig. Thus there is one turning point at E = mogq.71) For m 0 = l x 1 0 _ 3 k g .5 A particle above a flat Earth.
AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.f V ft h JXl h 2TT (14.14.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.73) Jx! rp2a dx I^<E 2 v) = ^ n — = . (14. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .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 . Chapter 16) one obtains the quantization relation (for Mo = const. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V).72) \\2dxp=Qdx^{Eh Hence.7. and u> is the oscillator frequency in either well.4: W K B level splitting formula Derive from the WKB solutions of the Schrodinger equation with symmetric double well potential the WKB level splitting formula A f B £ = . with p = modx/dt.= — <*— = h(n+\.2.75) which verifies the formula immediately. 1 . 9 .5 Further Examples E x a m p l e 14. E x a m p l e 14. d dE V)={n+)n. Solution: The proof is contained in Example 18.l. (14. ra 2/ = 0.( 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 (14.5 Further Examples 301 14. In the case of screened Coulomb potentials (cf.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 ( — = .
Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.<?2i<?3 for E > 0 given by E — V = 0. which determine the imaginary part of the energy. 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 .76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf.78) (a) We determine first the three turning points at qi = <?i. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.302 and T CHAPTER 14. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.6 The cubic potential.79) .6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). The q term in this equation can be removed by transforming the equation to a cubic in y. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.133)). 14.y3) = 0. q = y+ a . where 1 2 • 1 2 y = Q.~ \ (14. (11.77) Determine (a) the turning points. i.y2)(y . (14. (14. Example 14. where . 14. Eq.a .6.e. Fig.
ui=K(k).T h o m s o n [196].9 3 — = = 2 " 3 " [ . 78: sin 120° = v ^ A c o s 120° = . tan60°=%/3.84) 92 .9 l ) f f / du s n 2 u en2 d n 2 u . i. H e r e cos(<9 ± 120°) = . 6 = 0 i m p l i e s E m a x of E q . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 .2 c o s 0 ] . gi ~ y {V3e} < 0.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . M . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . 91 < 93 < 9 < 92(14.9 2 [ .79) a n d t h e n s e t t i n g m o = 1.08.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0.78)) c o s 30 •• 4P3 ( ' \ aeb2} . B y r d a n d M .0^6O°. (14. 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. 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. ' P .cos 0 + cos(6> + 120°)] = a2 92 . p .' : ± 6 0 ° w h e n E = 0. 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 = 1.81) m a 91 .c o s O + cos(6» . f o r m u l a s 236.< 1.120°)] = . we h a v e qx = .g i < 1. a n d t h e a n g l e 0 = 0° t o E = £ m a x . 37. (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0.2/3 i n t o q = y + a 2 / 3 .1 / 2 .5 Further Examples 303 T h e r o o t s of t h i s e q u a t i o n a r e k n o w n * a n d for 4 P 3 > 27Q2 a r e r e a l a n d e x p r e s s e d in t e r m s of a n a n g l e 6 given b y (observe t h a t w i t h t h e choice of t h e m i n u s sign. F . 14. (14.a 2 [ l . q3 = a2[\ .gi = 1.q)(q .80) I n s e r t i n g t h e r o o t s yi.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. (14. B . q3 ~  . e > 0. V3 + t a n 0 2 . 0<6O°. 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 .= s i n t 3 v 3 (14.93 2a2 [ . 93. (l_fc' 2 a + fe'4)l/4. 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 . 80 a n d 361. . p .c o s ( 0 + 12O°) + c o s ( 0 .k 2 2tan# \/3 + tan! .04. M i l n e .{ V 3 e } > 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 . p .6. ui=sn 1 \l— — = l. we set 0 = 60° — e.93).9 3 a n d t h e elliptic m o d u l u s k a n d t h e p a r a m e t e r g a r e given b y k2 k'z = 92 — 93 _ 92 .2 c o s ( 0 . D w i g h t [81]. 92. F r i e d m a n [40]. p . (14. D .g i ) ( g 2 . 92 = a2[l .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ .14.\tan0^ v/3. g 2 ~y{3}>0.2 c o s ( 0 + 1 2 0 ° ) ] . _ \ / 3 — tan(9 2 V92 .e. 212. § T o h e l p we cite H. V cose' + sin 8 (14.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.3/2.
^ + ^ ) (l + fc'2)2 .2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. 91 < q < q3 < 92 bJc.) .32)). Next we evaluate the required integral across the well from 91 to 53 at energy E. the integrand of the integral across the well is effectively the inverse of this momentum. Friedman [40]. The WKB Method elliptic integral of the first kind. with K(k = 1) = oo. Eq. as in (14. p. . (14. We can evaluate this integral again with the use of Tables of Integrals. Whereas the integrand of the above barrier integral is effectively a momentum. V 93 . 72. formula 233. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .86) 15 tan. F.85). V?2 . G(fc) =86aS (ITS£ li[fe.sm t %/3 £\.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — .?)(?3 .= —^ = k a2[cos0+^] v ^ + tanfl V=". E(u = 0) = 0. (24.~ y ^ ) a< [ COS 5 + . sinV = q3 qi ~ = 1.00. . g= 2 . 2 2 .89) This expression will again be obtained later with the help of configurations called "bounces" (cf.91 . 0 (14. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .84) we obtain Carrier = 2a 2 sin6> 2 b( . (14.304 CHAPTER 14.91 F(TT/2. E(u = K(k)) = E(k).k). V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .qz) (14./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. A:') = # ( * ' ) • (14. D.92) ''P.2(fc2 .fc'2 + fe'4)"1/4.87) Inserting G(fc) together with the expressions for the prefactors into (14.S . (14.91) a2[cos0^] y/3tanfl ? L . E(u) the incomplete elliptic integral of the second kind. 1+72 = and J o s 0 + ^ = (1 . we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . Thus Iwen=lgF(ip.^ V3 / V V3 y a / c o s e J. Byrd and M.
(14.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.80).93) where w is the harmonic oscillator frequency in the well. we obtain E a6b2k'4(l + k'2). .88) in rising powers of fc' . Ryzhik [122]. S. to represent a physical decay rate.14.4 2n + 1 fc' = 4 — ^ — .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. we obtain > . 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. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. (14.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. and hence at E = 0. a°b n = 0. With q = 1/z the integral (14.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).113. 906. M.85) (14.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . (14. 32 (14.96) Expanding the coefficients of the elliptic integrals in Eq.l.. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1.114. Gradshteyn and I.3 and 8.2.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . formulas 8.63b). Thus one obtains from (14.** However. pp.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . .3. (14. so that even in the case of the ground state the zero point energy will contribute to the prefactor. (14.97) 4 32 4 32 Here we insert the corresponding expansions of the complete elliptic integrals'^ (note the argument of K and E is fc): K(k) E(k) fc'2 = = < $ ) fc'2 1+ f l n\k'l f H + 1 4 5 3fc' 1T Hfc^J ~2_ 1^ \k'J 12 (14. 2a 5 b K(k) + 2 E(k). 905. we obtain 15 .
also Eq. (18. 2n + 1 — / 26a5b \ . . 7 ~ —.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results. _ exp[2/barrier] >. The WKB Method For k' —• 0 the first and the third terms dominate.ba / 25a5b\n+2 ±l_l eT5 a f >.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.306 CHAPTER 14. i.Wigner formula (10. /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . this has not yet been done.J + 0(VoJb).e. V (14. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s... so that we obtain 4 .96) for the imaginary part of the energy as in the Breit. (20.> tf = = 2/wfiii = ±tv8a°b——P. _ T T • (14. (14. i. 8 Bfc .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq.„. the result agrees with the ground state path integral result (24.101) (14.j . We observe that with \j2je as 1. 15 a . n ~ V27T . It follows that (for one complete orbit back to the original turning point) .— (14.7 .e x p [ ./^~^l^ab ^aub . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .73).fc'2 + fc'4)1/4 * ± t ^ .178)) call the Furry factor set equal to one.100) 'barrier = ~0°b In [ ^ — j .. (0) i hw En = E\> . .( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) ..e.102) .2 7 b a r r i e r ] Z Z7T where (cf.
307 . Lane and T. B. along with inclusion of angular momentum and spin effects and their interactions. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. Thacker. i. There are numerous. Quigg and J.* 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. The study of this potential. T. the Coulomb potential. [232]. K. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. L. in fact. after the realization that quarks as their constituents might not exist as free particles and. L. the classification of the various states of nucleons and mesons and other particles.g.M. which permits only bound states. D. describing free fall under gravity. This potential became widely popular in the spectroscopy of elementary particles. 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. See particularly E. Rosner [269]. Rosner [230].e. C. [231].Chapter 15 Power Potentials 15. [233] and H. now known as quantum chromodynamics. C. at least indirect. Quigg and J. Gottfried. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. indications from the nonabelian generalization of quantized Maxwell theory. Yan [82]. Kinoshita. K. Eichten. that this is indeed the case.1 Introductory Remarks The particular linear potential we considered in Chapter 13. led to a spectrum which contains only scattering states.
~ (I + l)rl (15. R(r) = ^ .V 7 = const.3) with the boundary conditions''' u(0) = 0.2 The force The Power Potential F = . h2 2n V 2 * ( r ) + [V(r) .m2. In order to have a clear starting point.1. i. and so u'(r) . where \x is the reduced mass of the two particles of masses mi. We leave consideration of the logarithmic potential to some remarks and Example 15. we return to the Schrodinger equation in three dimensions.4>)ru{r) = Ylm(9. These particles therefore cannot be separated by any finite amount of energy. 15. </>)R(r).5) u'(0) u[r) r=0 = R(0). we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r).6) = (I + l)u(r)/r = R ~ rl. A > 0. m\m<i M= : .1) i.4) Instead of considering the linear potential. (15. f dr{u(r)}'2 1. Power Potentials general power potentials.u ~ r i + 1 (/ + l)R(r).£ ] * ( r ) = 0. v > 1. and the normalization [ dr\y(r)\2 = 1. mi + m2 and r is the relative coordinate.2) tt(r) = Ylm{9.e. For a central potential V(r) we write (15. This force maintains this value irrespective of how far apart the particles are separated.r acting between two particles is constant and directed towards the origin (of the relative coordinate). and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. (15.308 CHAPTER 15. and u'(0) > [u(r)/r]0.e. (15.
. Then k(x) = k(x).V(r)) = (2nl) +1 N.9) into the threedimensional case in the following form: / Jo C dry/2n(E .8b) This condition implies that we have only one pair of turning points. (15. (15. odd integer ^=(nll)vh. This means that we transcribe Eq. We therefore proceed as follows. . We now want to find a quantization condition for the threedimensional problem. (15.1. and the turning points are at a — — b := — xc. or f l H " ^ ) ! . .15.9) Threedimensional case: We transcribe the above considerations of the onedimensional case into that of the threedimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!).7) k(x) = h/y/2m0(EV(x)).63b) for the onedimensional Schrodinger equation. and there identify the Swave states with those states of the onedimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq. so that with Eq. which has been reduced to an effective onedimensional problem in the polar coordinate r. (15.2. .10) Here rc is the turning point and n — 1.63a). 3 . (15. x > 0. (14. We also assume that V'(x) > 0. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality. (15."MA. d5.. is to be interpreted as the principal quantum number in three dimensions.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. N = 0.4) in the threedimensional case.8a) ^ JO = (2Ar + l ) £ .63b): 2 and V(0) = 0 (15.2.8a) we can write the quantization condition (14. In order to to obtain the ..
_ —vr" *dr. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2. (15. s.17) The integral on the right is the integral representation of the beta function B(x. c (nj\irh.310 CHAPTER 15.y) defined by (xl)\(yl)\ y ^ (x + y . E A fv. (15.13) dt .12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. Power Potentials eigenvalues E —• En for the radial potential (15. .'(1*)1/2= J1F\lt)y1dt.13) we obtain _ X_ „ _ so that A£_ (15. Now. dr = . (1516) (15.11) rc = (15.15) I ' = ^ 1 (( 1) B{x.12) and (15.\rvc = 0 .6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E . (15.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/.14) Prom Eqs.
18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15.15. and n — — ]7r/i. T(z + 1) = z\.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.1.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. i r ( i ) = r ( l + i ) = ( i ) ! . One can convince oneself now. 15.1 Approach to square well with v —> oo. . $T(z) = {z.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. ^ '"Compare with Eq. Weinstein [281]. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. The case v = 4 is that of the pure anharmonic oscillator also discussed by M.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) .1)!. (12. for instance simply graphically as indicated in Fig.55). (15. 15.
oj2/2.61) h2 = I. eigenvalues (JV + h)tkj — (2n .1 + \)hx> = {An • l)£fiw. (6. 15. discussion at the beginning of this section).2) implies A s fj.2.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones. x > 0. we obtain E„. In the case v = 1. the case of the linear potential.21) (4nl). 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). since these are the ones we are interested in (cf.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). The potential V(x) = X\x\ (15. (15.oj = ^/2X/JX. Physically it does not make sense for the probability amplitude to have a discontinuity there.> r ( 2 ) A V 2 in \)*\W \it^T& (15.41) for the .mo = \. the harmonic oscillator. Power Potentials In the case of v = 2. we obtain En = (n . = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. (6.g — —2. For the following we set in Eq.25) "Comparison with Eq.312 CHAPTER 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. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. > and with Eq. (13. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig.
57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . this wave function has the required exponentially decreasing behaviour at infinity. (15.26) For z = s := E — x —> —oo.2 Behaviour of an odd wave function at the origin.60) applies. dz2 According to Eqs.29) . and for s —> oo the trigonometric behaviour of Eq.e. Ai(s) 1 2v^(s)V4 exp x —> +oo.28) In particular at x — 0. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E .15.x). i. E > V or s > 0.e.e.x)\x=0 = 0. i.56) and (14.27) i.e. (15. s = E.59) applies. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. Ai(s) "v^ G ' cos s 3/2 _ t (15.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. In the domain —a < x < + a . 15. (14. Eq. (13. (13.e. i. i. \(s?» (15.
Quantum Mechanics with Applications to Quarkonium. (15.. [231]. (15.02137 10.1 demonstrates the quality of the WKB approximation by comparison with the exact values. copyright of NorthHolland Publ.08181 5.2.02265 10. (2/3)E3/2 or 3 / 1 2/3 E — En — .93602 12.93528 12.94249 9.78445 7. Power Potentials Table 15. to obtain the zeros of the Airy function numerically. Co. Table 2.^ = !(2nl).e.00852 11.04017 11. i. 201. (15.82814 i. with permission from Elsevier.3. Table 15.51716 6. n = l. p.94413 9. but for large > values of E Eq.. of course. so that n — 1/4 ~ n.29) implies that cos (  ^ .30) ! ^ / 2 .7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large.33811 4. It is possible.31) (an odd function has an odd number of zeros!). Rosner. It is interesting to note that in the dominant approximation both methods agree.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . the eigenvalues En are determined by the zeros of the Airy function.32025 4.e.52056 6.00767 11.78671 7.e. Quigg and J. . The expression (15. *This Table is reprinted from C.03914 11.08795 5..En from Ai(£„) = 0 2. i.= ) = «.28) is really only valid for s large and s — oo.314 CHAPTER 15..82878 1 2 3 4 5 6 7 8 9 10 2. = im — (7r/4). L.
.3 The ThreeDimensional Wave Function 315 15. r <53) 1 3  dr cos 2 () (15.34a) **The most frequently quoted reference for this result is R. The decay widths T. We indicate briefly here — without entering into further details — how these quantities can be calculated. of course. (14.g. See also J. Jackson [140]. can be shown to be proportional to the modulussquared of the particle wave function at the origin.e. which are inversely proportional to the lifetimes. We saw previously (see for instance Eq.• •) by a mean value. e.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (.e. have only a finite lifetime and decay into other particles such as electronpositron pairs. in Chapters 24 and 26. i. *Note that this makes N a W K B normalization constant. D. In order to be comparable with experimental data these investigations. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. ee.g. cos2()^— / 1 fr° 2 > w . van Royen and V.** r ( # ^ e e ) o c *(0) 2 .e. F.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. i. P. Weisskopf [239]. e. The resulting masses agree in most cases very well with those extracted from experimental observations.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. the charmanticharm meson ^ . In general these quarkantiquark pairs.15. We shall encounter WKB normalization constants at numerous points in later chapters.
.1 )n .Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .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.\ \ * n = l.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .. (15.40) VWY . & (15.l N n . (15. dE a—— / dn J0 so t h a t with Eq. (15.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.39) in / . From Eq.e. (15.rcu w X{r)dr = Kh . so t h a t approximately l N2 / Jo k(r)dr .2.34b) VMEvy Variation or differentiation of Eq. Power Potentials this averaging implies a numerical factor like 1/2. i. A = (15.36) \i dEn irh2 dn According to Eqs.2) and (15.316 In accordance with t h e relation CHAPTER 15. (15.37) (15.35) N2 ..38) N : sin V ^ Jr dr' 7T (15.
.g. W. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. J. An enormous literature exists on the subject in view of its relevance to the spectroscopy of mesons and baryons made up of quark constituents and the necessity of checks with results derived from experiments.2.* Using the result (15.23) for the energy levels of the linear potential. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15.l ) n .1 ^ ( 2 ^ „ ) 1 / 4 . W.15.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. Hence Eq. G. K. E. we can investigate the dependence of the decay rates on n (note that these would be decays into particles different from those bound by the linear potential — a consideration we cannot enter into more detail here). S. MiillerKirsten [28]. J. W. 3 . . (15.41) I*(0)2=_L^)^. (15.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. MiillerKirsten [147]. for various potentials: (a) For a confinement potential of power v. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. MiillerKirsten [29]. the expression (15. . Bose. . The case of arbitrary positive power potentials as above has been considered by R. for instance. K. we have ^'(0) = ( .3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. § S.37) we obtain: 1 (2M^)1/2 47r ft. Kaushal and H. 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. and hence with Eq.38) implies 317 (15.44) (15. Hite and H. Bose and H. Itt^pocn2*"1^"^. J. S.
K. i. (15.L2 h4 = 4/3exp[(2a — /3)/f3].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.L 2 + U{z)]<f>. D. 7 With the substitutions r = exp(z — c). ip = exp[(z — c)/2].^ +•••. (15. as well as Expanding U(z) about its extremum at ZQ = —1/2 and using the perturbation method of Dingle and Muller derive the expansions N)2 = ^^ 3 .. = 1(1 + 1). n = 0. . and there Eq. W. E' = E + gln(r0). in agreement with Eq. . .2.46) and the classical Kepler period of Eq. {S = ^ > 0. S.5. Solution: Details can be found in the literature.^ ( 3 ^ + l)55p^«(3^l) + .j . Example 15.1. q = 2n + l. MiillerKirsten [29]. Bose and H." 1 11 L . Landau and E.g ^£+(<*Pl»riy = 0.— oo < z < oo.133).c = —a//3. rewrite the equation as ^ z + [ . Power Potentials *(0) 2 oc .318 (c) and for the Coulomb potential^ CHAPTER 15. Eq = i q . We observed there in the case of the Coulomb potential that the period is proportional to n 3 . U(z) =  dz zf3e2a/?e2*. (11.75).I l n f ^ p . (14.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). = (I + 1/2) 2 . M. *=^f. Lifshitz [157].. J...e.
Historically this potential arose with the realization 319 pVr .Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro .1 Introductory Remarks We observed previously that the infinite range of the Coulomb potential — the concept of range being defined more precisely below — leads to a scattering phase with a logarithmr contribution. The exponential insures a rapid falloff of the potential at large distances and for this reason the parameter ro can be looked at as a measure of the range of the potential. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. that no matter how far the particle is scattered away from the source of the Coulomb potential. The result is a screening of the Coulomb potential which thus attributes it a finite range ro.Chapter 16 Screened Coulomb Potentials 16. and in fact. it is. so that like charges are repelled and unlike charges are attracted by each other. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. This implies. it will always notice the latter's presence. A very similar potential is the Yukawa potential V{r) = g2 — . This effect sounds unphysical. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. In reality. It is clear that for ro — oo the > potential becomes of Coulomb type.
in fact the parameter [i represents the mass of such a spinless meson. however.1) We encountered Green's functions earlier. and that. the mathematical expression with an exponential is not so easy to handle analytically. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.e. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. from the relation Pi + P 2 + M2c2 = 0. all coefficients Mj are real and independent of the energy E = k2. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. Heisenberg [133]. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. .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.ii k .* The relativistic equation of motion of such a meson when free is the KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x. one frequently resorts in calculations to the expansion of the potential in rising powers of r. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. (16. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq.320 CHAPTER 16. 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. i. (16. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). What is meant is that the fourdimensional Minkowskian Fourier transform of the propagator is given by the relation dk (2TT) J 4 e x ^r—2 = 7Z^u o)^n k + fi2 4TT v y x 2 ikx ^ e^W ( 16 .3) where for the real potentials we consider here. Since.2 ) or (p = hk) 1 /" .
to be of the order of 1/k. They are usually written I = an(E). "These were first considered by C.e. Masson [180].r)=0. 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. i. Potentials expandable as in Eq. Squires [258]. C.4) where E = k2 and h = c = 1 = 2m. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. 2 . Basically this argument amounts to an argument similar to that used in the case of the Coulomb potential where the integer n arose from the requirement that the wave function be normalizable. (16. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.e. . Wu [47]. (16. S S.§ Our intention here is to consider these potentials as a generalization of the Coulomb potential and hence to proceed along similar lines in the derivation of Regge trajectories. Kinoshita [21].Q. only for the cases of n = 0. (16. . i.3) have been considered by various authors. Mandelstam [185]. Cheng and T.k. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.16. A. Omnes and M.mo being the reduced mass of the system.3) in the radial Schrodinger equation for the partial wave ip(l. £+**PVV t/. Froissart [223].r).1 in the above. if>oo. S.e. Frautschi [97] and E.5) where K — ik and A n is an expansion in descending powers of K. physical) values of angular momentum as functions of energy E. f H. T. Lovelace and D. . J. i. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. Bethe and T. 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. . $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. bound state eigenenergies and the 5matrix. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.k. (16.e.{l.1. however. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. l+n+l = ^P. See also the other references below.
9) the cut starts at E = k2 = 0. H. Regge [31] and E.k.e.15) and (16. < oo for all 0 < vr/2. < const. Squires [258]. 'In the case of the Coulomb potential (cf.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. as may be seen from Eqs. M. o rV(r) I dpp\V{peie)\ regular at r = 0. 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. Under these conditions the S'matrix is meromorphic (i. (16. W. Longoni and T. W. J. Muller [201] and [202]. Screened Coulomb Potentials as a function of the energy.322 CHAPTER 16. Section 11. Miiller and K.** The energy is later obtained by reversion of the resulting series.z) Then x is a solution of the equation VaX= (16. *See in particular A. one can assume an expansion of the potential V(r) in ascending powers of r. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. for all n.4) to z = —IKv and set TP(l. A.z) = ez2/2zl+1X(l. In the following we follow mainly the last two of these references. starting with the power r _ 1 of the Coulomb potential.k. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. W.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. Schilcher [203]. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. 16. J.ndp. Bottino.19).)/j.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. J. in the present case at fc2 = 0 or k2 — Mi = 0. tf 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. (16. Muller [204]. J. .
j l) + (a + j.10) zm$(a)= ^ j=m Sm(a.9) (16. a + 1) = a — b+ 1.a)${a) + {a. W.j)$(a + j).j +(a + j + l.11a) The coefficients Sm(a. .L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. (16.i 0) = 0.j) (16. (16.a + j)Smi(a. MiillerKirsten [249]. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).z) = Ha). {0) X = Ha. anharmonic and cosine potentials have been derived in L.j) may be computed from a recurrence relation which follows from the coefficients (16.a + j)Smi(a.5)) a = + 323 .awhere (a. By a repeated application of the recurrence relation (16.a — l) — a — l. Eq. where $(o.b.10): Sm(a.7) is seen to be of order 1/K. so that to leading order we have VaX{0) = 0. The function <&(a. K. as in the case of the Coulomb problem. J. a) —b — 2a. and all Sm(a.9) we obtain m l)$(a . (a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.5) is equivalent to a = —n.1).a + + l).16. Sharma and H. which means that the hypergeometric series breaks off after a finite number of terms. (16. j)Smi(a. We also observe that the ansatz (16.b.* Recurrence relations for coefficients of perturbation expansions for Yukawa.0) = 1. (a. (16.j) for \j\ > m are zero.2 Regge Trajectories where d2 Va = z^ and (cf.j) = (a + j l. all other So(a.a+ l)$(a + 1) + (a.11b) ^ The associated boundary conditions are: So(a.
aj 2 \ [a. This follows from the fact that D a $(a) = 0.324 CHAPTER 16. (16. Evaluating the first few terms of the expansion (16.[ a + l.a + l ] 2 [ . the latter can be written . a . Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. 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).<&(a + n) on the right hand side of Eq. [ a . This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0.n.o]i = M 0 .5) the Regge trajectories I = ln(K).e. i R »] = 2K[a. of course.13) one obtains the quantity An(K) and hence with Eq. (16.a]l^a) + ^2(2K)i+1 Y. [a.aj 2 (2K)A _ [a.An{K).13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . (16. (16. and hence Va$(a + n) = ra$(a + n).l.7). [a. oo _ .7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.j). #(a + ra) fjL$(a + n) V. for the coefficients M^> of the expansion Details can be found in the references cited above. Any term /j. Va+n = Va . (16.12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higherorder perturbation terms.' — .a + j]i+l = MiSi(a.12a) where [a.[a>a + J]i+i*(a + 3). + (16. when n = 0. 0 < \j\ < i.1] 2 .a\4. We first cite the final result . 1 . i. .
5). 16.§ This is an important observation which indicates the equivalence of the methods. Burke and C.1) + 3M3M02 + M22n(n + 1) (16. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential. (16. Obviously the expansion — which. .1 by evaluating the first two terms.e.^ 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. i.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . G.1.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . of course. Inserting this expansion into Eq. we obtain the expansion for the Regge poles. Such plots of numerically computed Regge trajectories for specific values of the overall coupling constant and energies varying from minus infinity to plus infinity have been given by various authors. (16.2. including an exploration of the domain of small energies E. Thus the expected appearance is that shown in Fig.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16. . P.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.16.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A. However.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n .l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n .^ p [ 3 M 4 ( n . The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. Tate [6]. Ahmadzadeh. With numerical methods one can achieve more.
The conjugate variable to energy is time.a) = b .12b) we obtain [a. a] 2 = M i Si (a.^ . Solution: We evaluate the first three terms of expansion (16. However. Example 16. so the imaginary part ai of I — an represents its width in angular momentum. where the imaginary part of I is very small. For a bound state a j = 0 and the orbit becomes permanent. Frautschi [97]. one can expect a resonance with the lifetime determined by this imaginary part.2 a = In See S. ±1) = 0. p.1).11b) we obtain Si(a.326 CHAPTER 16. 114. satisfies the relation ajAO ~ h. ai is small and A6 is large. Similarly the conjugate variable to angular momentum is angle.13). (a.—" + In = — . (16. 0). In fact.a)So(a. .0).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. and the angle A#.O) = Other terms vanish since So (a.0) = (a. Hence Si (o. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. and the lifetime At of the resonance satisfies the relation TAt ~ h. For a resonance with a long lifetime. 16. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown.1 Typical Regge trajectory for a strongly attractive Yukawa potential. [a. From Eq. a] 3 = M 2 S 2 (a." Re I 1 0 1 2 Fig. 0) and Prom Eq. just as a decay width T represents the width of the resonance in energy. C. . through which the particle orbits during the course of the resonance. (16. 16.
(n+l)U+^pJ .a + l)S0(a.2o)Si(a. i ) again from Eq. 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 . . 1) (a .2 a ) . One sets z = 1/r so that.13) we obtain 0 (Mo .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I. 14. 0) + a S i ( a .0) + 0 + 0 = ( 6 . (a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.0) + 0 + 0 = ( 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 .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .a)5o(a. a)Si (a.0)+ 0 + 0 = (a1). .n ( n + l Inserting these expressions into expansion (16.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . Solution: Ignoring higher order terms in r.1) + (a .(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . A" An An A" = = = (a.al)S0(a. a)Si (a.0) (a .6 + 1 ) . 0) + (a + 1. We obtain the quantities S i ( a . Sec.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity". .^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . using the Cauchy formula (6.16. we have to evaluate (cf.0) Si(a.b)(a .1 ) + (6 . a)Si (a.11b).2 Regge Trajectories 327 Analogously we evaluate S2(a.6)Si(a. (16.1 ) + (a. 1) Hence 52(a.0) = (a . Thus (with terms which are 0): Si(a.6 + l)a + A.1.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.1) Si(a. (a. 1).
14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A..1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. Screened Coulomb Potentials Mo . Singh [252]. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. however.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . Boukema [34] and [35].* 16. we can use Eq.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . 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 .2. (16.328 Thus here dr K' CHAPTER 16.e.5) together with the expansion (16.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16.15). Example 16. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. Solution: For details of the solution we refer to papers of Boukema.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.l. Here we do not perform this procedure. by expanding the right hand side in rising powers of v and evaluating the individual integrals. ±2HK in agreement with our expressions above. (16.3 The 5Matrix It is clear that if we now work through the usual procedure for the derivation of the S'matrix we pick up a logarithmic phase as in the case of the Coulomb potential. n = 0. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. tSee Chapter 11 and V. i. The turning points in this case are at r = 0 and r = oo.t First. *J. . I.
the behaviour of the scattering phase in the domain of high energies. . for instance. the following *See e. W. Paralleling the case of the Coulomb potential in Chapter 11. Now reversing the expansion (16.14).l)n(n + l)(n + 2) + 2M2M02 .19) we obtain the energy. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. we can now write down the S'matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e r(/ + i^#>)e ' (16 18) ' We observe that the poles of the S'matrix are given by l + l + ^p = n. W.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\. such as.2.18) of the Smatrix now to explore further aspects. J. 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 . J.. This has been done§ and one obtains .19) +2M 3 M 0 (3n 2 + 3 n ...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\.16. i. One can use the explicit expression (16. Miiller [204]. § See H.g. which are precisely the expressions yielding the Regge trajectories of above.0[(KZ + Mx)'6} (16. Miiller [211].17).* 16. H.l (two additional terms are given in the literature).l. n = 0.17) and is therefore real for real values of the potential coefficients Mj. . We observe also that the S'matrix is unitary as a consequence of the result (16.e.
E. MiiilerKirsten and N. VahediFaridi [212] contain as special case the expansions of the first pair of authors. the transformation was introduced in the form given below by G.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 .f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n . 107.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. .3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . Warburton [279].^ 16.2)(n . 282. W. 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) ^ — . i.1) + 2M2M. Zauderer [286].l)n 2 (n + l) 2 (n + 2) 1 + • • • (16.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . J. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A.330 CHAPTER 16. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). 'According to S.l)n(n + l)(n + 2) .11) + 2M22M03(9n2 + 9n . p. Screened Coulomb Potentials expansion. Iafrate and L.e.* We recall from Chapter 11 the definition of the scattering amplitude F(9.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . Mendelsohn [135] and E. in the following expression with an "See in particular G.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . Large coupling expansions derived in H. Sommerfeld [256]. Frautschi [97]. J. C. N. p.5 The Sommerfeld—Watson Transform The basic theoretical tool for the exploration of Regge poles is a representation of the scattering amplitude given by the socalled SommerfeldWatson transform.< 3 M 4 M 0 ( n . B. A. Watson [280] in 1918 and later resurrected by A.
I = n + x.2.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. .iJ 11 —[S(l. \x\ <C l .fe)P . n = 0.k)~l] = ~ k" PiSiW sin 5i(k) and S(l.k) 0ikr (16. (11.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16... (11.2 The integration contour C.k) — e 2i5[(k) ~" p—iirl (16.25) Setting Note the minus sign in the argument of the Legendre function..26.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. For this purpose we consider the following integral taken along the contour C shown in Fig.22) = \F(9. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig. z=o in which (cf.k)\2. ~:' sin nl (16.''' The idea is now to consider each term of the expansion (16. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n). c = 1.k) = 2zfc L " v "'' v .184) to (11. The scattering amplitude possesses the partial wave expansion F(0.k) = ^T(2l + !)/(*.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0. 16. Actually 5j lrr/2 as in Eqs.1.kmcose).23) f(l.187).24) Si(k) being the phase shift. Eq..187)) (16. (cos 9).2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. 16.
3. Thus the direct reaction may. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. Consider the reaction of a particle a{p) with fourmomentum p^ colliding with a particle b(q) having fourmomentum q^ and producing a particle c(p') and a particle d(q'). In order to see the relevance of Regge trajectories in a reaction of particles. For the momenta indicated we set (with metric +.26) P((+COS0) = 2 ( 2 n + l)/(n.—) s = (p + q)2. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. a + c —> b + d. in which case c = a and d = b. u={pq')\ (16.332 CHAPTER 16. we obtain m® = ± I dl^±^f(l. as indicated in Fig. be elastic scattering of a off b.27) a(p) © b(q) Fig. 16. a + d — • c + b. for instance.28) .fc)P„(cos0). t = (qq')2.k) Zi Jc TT{L . 16.—. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. d(q') (16.—. 16.3 Reaction channels and their respective Mandelstam variables. n=0 which is the usual partial wave expansion.n) oo {l)1 Pt{cosO) v ' (16. The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d.3.
30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).t. t = 2q£(i . if we had been considering the s reaction originally. 16.u) (if any of the external particles has nonzero spin.c o s 0 s ) .t) = y^Fs(s.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo. the variables t and u would describe momentum transfers in the crossed channels.The variables s.u are not independent.32) t = 4(q? + ml).16. (16. T See e. if s describes the square of the total energy in the s channel. N. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).. . S. One now makes two hypotheses. A(s. A(s. Symbolically we then have the situation shown in Fig.(<fe)^(cos 6t). the only kinematics we require in the following is given by the relations (e. one can write the amplitude as depending only on two.et) (16.g. Goldberger. u = . In a Regge theory on the other hand. one then has A(s. s = 2qf(lcos0t)(16. Chew [48].u are known as Mandelstam variables. For a Lorentzinvariant normalization of the initial and final states. the dynamics is described by the exchange of quantum numbers. L. u. Treiman [25]. Mandelstam [184] and R.31) i l Taking all particles to have the same mass mo. (16.g.4.g.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. Blankenbecler.9s)\2. N.0s). the momentum k there would correspond to qs) s = ~4(ql + ml). Khuri and S. Thus. F.§ Since the three variables s. Ft = £ ( 2 J + l)F.t. (a) Mandelstam's hypothesis: All three processes (described by the s. § See G.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. there will be several such amplitudes which together describe all three processes). the quantities carrying these quantum numbers are the Regge trajectories.t). which is an analytic function of the variables s.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. B. etc.t) = VtFt(t. except for poles and cuts which characterize the reactions in the three channels.2 ^ ( 1 + cos6 s ). t. M.t. e.t.
16. the Legendre polynomial Pj(cos 9) does not possess a cut. Screened Coulomb Potentials Regge oc(t) Fig.4 Yukawa versus Regge theory. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories.5 for Yukawa potentials V{r) u: lm cose dr a(/j.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. So what can one do? 1 H.e. .r 1+m 2 /2q 2 o ^t "" Re cost: Fig. 16. 16. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. However. the partial wave expansion is not valid in the physical region of the (i.)e~ •fj. For s — oo : cos 9t = 1 + s/2q^ — oo. But for integral values of I. Lehmann [162]. in the domain of schannel physical values of the kinematical of the ichannel).5 The Lehmann ellipse.334 Yukawa CHAPTER 16. But > > wave expansion X)(2Z + l)F.
i.5 The SommerfeldWatson Transform 335 T h e answer is.33) .16. Moreover.t). sin nl [16. B Iml 1 \ i All \ r\ 0 / closed . the contour representation (16. 16. 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. N F(6.^ B 1 A Fig. „ i.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. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.e. to use the SommerfeldWatson transform. contour C / / D ^ ^ .e. all poles of Ql > 0.6 The integration contour C". Thus consider now the same integral but taken along the closed contour C shown in Fig.26) but for a different choice of the contour.1 / 2 .6. and there is only a finite number N + 1 in the domain of 9W = . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l.
of course. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. W. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s.t)&B(t)sa°V. It follows that under these conditions. i. we have (16. Omnes and M. p. MiillerKirsten [29]. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. . this behaviour has been confirmed in high energy hadronic reactions. Regge trajectories have. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. (16. for high energies in • > the schannel). Foissart [223]. 42. K. J. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics.336 CHAPTER 16. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. Limic [177] and E. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Predazzi and T. at high schannel energies. Apart from more refined details.e. for plots and discussion see R. the amplitude can be represented as a sum over ^channel Regge poles.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. i. Eq.32)) and finite. In particular.35) This result is valid for s — oo and t negative (cf. Bose and H. 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. Regge [229]. 16. if ao(t) is the Regge pole with largest real part.e.e. as we discussed briefly in the above. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel.
Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum.6 Concluding Remarks 337 important parameter in string theory. MiillerKirsten [209].** Whereas the former are related to absorptive properties of a reaction. a composite particle being one with structure.g. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. W. 20. J. called "Regge cuts". like a meson or a baryon.1. . Singularities appear also in the form of cuts in the plane of complex angular momentum. The above treatment of screened Coulomb potentials is incomplete. Related aspects are discussed in Sec. and also as socalled "fixed poles" . the latter are related to the distinction between "' elementary and "composite particles". The way to do this is similar to calculations in Chapters 18 and 20.16. For an overview see e. and is composed of quarks. H.
This page is intentionally left blank .
25. however. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. See F.Chapter 17 Periodic Potentials 17. 19 and p.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. 339 . M. The Schrodinger equation with this potential.* These elliptic equations involve Jacobian elliptic functions like sn(a. The reason for these difficulties is. In essence these periodic Jacobian elliptic functions. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. The Mathieu equation can be obtained as limiting cases of spheroidal wave equations and the elliptic Lame and ellipsoidal equations which represent a further level of complication. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. called elliptic modulus. 2K or AK. p. K being the complete elliptic integral of the first kind. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. motivated mainly by the regularities of the crystal structure of matter. since many analogous relations hold. depending on the parameter k.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). the most immediate candidate is a trigonometric form like that of the the cosine function.g. are not much harder to handle than trigonometric functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. Arscott [11]. example 14. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult.0 < k2 < 1. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.) which are functions that interpolate between the n. and are themselves periodic with period e.
J.1. We might mention already here the much less familiar but more general Lame equation.2h2 cos 2z]y = 0. See also the discussion in Sec. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. J.1) where h2 is a parameter and — IT < z < IT. This is a significant difference compared with the Mathieu equation. 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. 8. those of integral order. 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. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. which one looks up in books like that of MilneThomson [196] or Tables when required. Both of these domains are important for a host of other considerations. 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. In fact. For large coupling. Like the trigonometric functions sine. . [205]. Muller [73].2 . such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. W. Dingle and H. the Lame equation possesses 2 n + l polynomial solutions. i. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. (17. cosine and tangent.. The most important formulas for our purposes here are collected in Appendix A. the three Jacobian elliptic functions s n z . which does not possess any polynomial solutions.4. The first of these conditions is already tMathieu equation: R. Periodic Potentials vide periodic potentials like trigonometric functions. each with its own characteristic eigenvalue. B. c n z and d n z are handled with analogous formulas. 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. Miiller. 17. which is conveniently written ^ + [\2K2sn2z]y = 0. The Mathieu equation. and to relate these to the weak coupling bands or regions of stability.e. (17.340 CHAPTER 17. or Sw&ve Schrodinger equation with elliptic potential.1. W. ''The reader who encounters these Jacobian elliptic functions here for the first time.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .. Lame equation: H.7. need not be afraid of them.
1. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. 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. 17. 17.1 The Floquet exponent We consider the given Mathieu equation with argument z and compare this with the same equation but with z replaced by z + ir. the case of h2 small. In classical mechanics stability is treated for instance in connection with planetary motion. these concepts are rarely explained as such there.e. i.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. h2 = 0 belongs to the domain of stability. a bounded function. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. 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). Salem and T. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r).e. Although the mathematics literature on the subject uses the physical concepts of stability and instability. y" + \y = 0(h2).2. Vachaspati [242].17. we compare the For interesting related discussions see M. i. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. since the solution there is of the form cos vAz. i. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. we see that in a plot of h2 versus A.e. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. 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. i. .e.^ 17. Looking at the Mathieu equation for h? —> 0. Adding the negative term — 2h2 cos 2z to A. the semiaxis A > 0. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms.
d(z + TT) cos 2z We see that there are solutions which are proportional. y+{z) . i. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z).e.2/i2 cos 2(2 + vr)]y(z + vr) = 0. y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2). sin V Az or linear combinations. TT. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. — • . (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). Periodic Potentials equations ^ p + [A .3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. ° ^ [1 + 0(h2)}. i. 2/(0)' yfr) . (17.342 CHAPTER 17.e. d2y(z + 7r + [A . y(2vr) 2 a = —7—^ = 1. (17.2h2 cos 2z]y(z) = 0. a = const. we obtain y(vr) y(2?r) a = — — a = —^^r.B and a. y~(z) = 6sin vXz. Setting z = 0.
4).e.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. \y+>y\ = y+{ir)y'{ir) . We can write therefore. 2/'(7r) = a ^ W + ^ . Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. (17. [y+(v) . 7 y\z) = ay'+(z) + (3y'_{z). whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. yV(0) = 0.3) we obtain y(z + ir) = a[ay+(z) + @y(z)].2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. y'(z + n) = a[ay'+(z) + py'_(z)}. y'_(Q) = bV\ with Wronskian W[y+.17. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4.aa][y'_{n) . with constants a and /3.a2b2\ a± ~ 2a6\/A . 2/_(0) = 0.y+(7r)y_(7r) = 0 or abVXa2 . we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. This means.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. Using Eq. i.oWX] . (17. This expression is equal to the coefficient of a2 in the quadratic equation for a. and by'+(?r) = —ay/Xy{n).y+(7r)j/_(7r)} = 0.{ay'_{ir) + bV\y+(ir)}a Now from Eq. we choose these with the following set of boundary conditions: y+(0) = a. y{z) = aty+(z) + 0y(z). the determinant of the matrix must vanish.
1.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. This calculation is demonstrated in Example 17. Prom this condition we determine later (cf. <T__ + <T_ = 2y + (vr) . (5i/ 2 + 7)/i 8 vz + — .5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2).8) "Observe that for integral values of v (in lowest order of h2). (17.22)) the boundary conditions of periodic solutions.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other.344 Setting / = 2by/\y+(Tr). i.. ** An important consequence of Eq. This is the easiest application of t h a t method since only simple trigonometric expressions are involved.e. 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. Eq. <7__cr_ = 1. we have <7± — CHAPTER 17. (17. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 .= — . where the constants are usually taken as unity from the beginning. .e. (17. Periodic Potentials g= 2aby/\.5) a The parameter i/ is known as Floquet exponent. T h e result for A is the expansion A = h4 2 . cr+ = 1/<T_. this implies y+(Tr)/a = ± 1 .0 This series may be reversed t o yield the Floquet exponent. (17. Setting cr+ = e . i. 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 . _ / ± v 7 ^ V . 8.455A2 + 1291A .7. **This point is not immediately clear from mathematics literature. .6) We can derive various terms of this expansion perturbatively with the method of Sec.
11) or smvz or e±ivz.10) A = n2. Meixner and F. p. Example 17. (17.2 how the expansions for small values of h2 may be obtained perturbatively. For these integral cases of v we demonstrate in Example 17. (17. (17. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. It follows that in these cases i? = n2[l + 0(h2)]. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. as will be done later.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17.12) S e e J.3).9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . cos TT\/A 32A(A1) 2 . Solution: We write the eigenvalue equation A = v2. W.35A + 8 . These cases therefore have to be dealt with separately.5) and (17.jl2.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0.„„ x n. Schiifke [193]. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y. VA = ±n. (17.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.35A + 15A2)/l8 /.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.17. n an integer.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. . 8. we must have a2 — 1 and hence e 2iv/A7r = 1} i.1) 15A2 . (17. = J—^ + o(hr).e. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). 124. (17.
With our perturbation formalism — as demonstrated in cases considered in Sees.16) *These smallh expansions are convergent. . v . by J. of course. ("• ^ + 2) (o) 2(2^ . '. i..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. (17. (17.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. (17. when a = 0 or 2v + a = 0. The right hand side of Eq.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. v .3.h2).h2).16) directly into the original equation in the form of Eq. siniy + 2)z + sin(i/ — 2)z. Meixner and F. Proceeding in this manner we obtain the solutions ce. i. since j / 0 ' = yv leaves unaccounted R\.e. .2) which applies when v is nonintegral.3 and 19.2. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . v) = 0. for instance.15) Hence a term fMy„+a on the right hand side of Eq. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .v + 2)yu+2].e.14) = = = cos(y+ 2)z + cos{y— 2)z. (17.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. (17.346 CHAPTER 17.v±2) = l. sev(z. (i/. h2). We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i.11). up to 0(h2) we have A = 0. (17. or A = 0(/i 4 ). as shown.2)j/„_ 2 + {v. W. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.(2) + . se and me as* oo i y = „«>) + yW + i. = . and (v. Next we treat terms yv+a in j / 1 ' in a similar way. (17.e. The recurrence relation can also be obtained by substituting the right hand side of Eq. . me±v(z. 17. see Example 17. i. e ± i ^ + 2 ' 2 + e±i("2)z.2) _ __ _ (y.„ + £ h2i £ P2i(2j)yv+j (17.!/) = 2A We now observe that DvVv = 0.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0.e.14) therefore leads to the following nextto leading order contribution (v. v)yu + {v. h2(u. where except. Schafke [193]. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.
J. See e. „ . * Note.18) Again applying Eq. Solution: We start as in Example 17. Schafke [193]."2). W. (17.e. + y~i) (17.g. the solutions are not yet normalized. The higher order terms naturally require a little more algebra. . . For v = 1 we then have yi = cosz = y~\. 0 = h2(u.14). These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A.17.2 ) v(y2.2: The eigenvalue A for u=l and h2 small Use the perturbation method of Chapter 8 to obtain the perturbation expansion of A for v = 1 and h2 small. (17.7). v) ' ( 2 v + 2) (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)+. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. p. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. i..17) Inserting 1 for the stepcoefficients of Eq...1 with A = v2 — 2h2A but with v = 1. 121. . however. thus determining the quantity A. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. one has to deal with each integral case separately as in the following example. Meixner and F.u) + h4 (". Following the first few arguments of Example 17. fo^ + 2) . (17..R{ ..1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3]. . In the case of integral values of i>.2—r——— (i/ + 2. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. Example 17.v)\ . 2 ( 2 i / . Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• .
Periodic Potentials h4 r ••• . . 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. With normalization (not to 1. and v is to be found in ascending powers of h2. We introduce a normalization constant c and rename the normalized solution y then yc. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . p.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai).24c) below) . the case of A an integer. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. m. W.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. h4 (17. i. J. see below!) one obtains the additional terms given by Meixner and Schafke [193].e. Schafke [193]. We consider this case again in an example. h2). 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 . 123) (there the solution is called cei(z. 120. Eq. h2)) except for the overall normalization which implies that in our (still unnormalized) case above there are no contributions cos z in the higher order contributions. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. The reverse situation is later of importance. (17.* The associated solution is ym+ym. 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].348 From this we obtain 2h2A = h2 CHAPTER 17.n integers. In the above we considered the case of v an integer and calculated the eigenvalue A. Meixner and F.
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. J .4)Av A2v A ~ 32A(A . J. ^S.9b) and considering the approach A — 4 gives > (A4)2 + .• 1 + 5n h 2 s 2932 Setting in cosm.4 ) + . H. (17.. Manvelyan.1)2/A TT2 +h8 (15A2 . (17. Hashimoto [124]. .('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 .1) .2 Cosine Potential: Weak Coupling Solutions Example 17.17.3 1 x 4 + 8) 274233 (.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . Gubser and A. S.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . Substituting these expressions into Eq.1) (A . fe47T2(A4) 4%/A(A . Liang and Yunbo Zhang [189].9b) v — 2 + <5. o 1• • • . W. (11A2 . We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .9b) about A = 4. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq. . we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i .. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. on the left hand side of Eq.l ) 2 (A .3 5 A + 8)TT2(A4) _ 3 / / 64(A .3 1 A + 8) ( A . (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 .Q . MiillerKirsten.
"These conditions. (17.e. M.10) applies.W4W=UtWSta±I(.w^.. for instance.29.21b). Taking in particular Eqs. These are defined by specific boundary conditions. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq.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. pp. bV\ y+(0) 2/40) 07. We verify these for the large/i 2 solutions in Example 17. Fig.350 CHAPTER 17. (17.1 Boundaries of domains of stability. 28 . i.6. . 17. solutions for integral values of v. We now derive these boundary conditions. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.2. 17. in F. = and s+w 2tW±I. as also conditions (17.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. Arscott [11].4) into account. may also be found.21a) and (17.
21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ . we obtain y+(7r) _ i  2y_(f).2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. /7r\j/!_(7r) y L l 7TJ . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. /TTW^) 2j^T±y+V2 From these equations we obtain y+{ir) in terms of functions at TT/2 by eliminating y'+(Tt) from the first and the third equations. which after some rearrangement can be written y+(vr) _ . .4(f) (17.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. . ^\y±M /vr\yL(7r) . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ .17. 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. Then y+ y 351 U . o y+(fy(f)ob\/A _ = 1 + 2I&A/X .
in this order.1. (17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. (17. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. 27r.. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq.3. of v is even or odd.1). These may be subdivided into classes depending on whether the integral value n — 0. There the second term on the right is nonzero.22) * a2n+i(h ). A *• b2n+2(h2). These are defined by the following boundary conditions.21b) must be ± 1 .2 we require this relation for nonintegral values of v.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . IT.— — — • • • .24c) 8 h2 sin 4z \ . (17. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. . Mr7nA (17. . (17.5) that the left hand sides of Eqs. (17.23) We can now see how the domains of stability and instability arise. We see that there are four possibilities and hence four different types of functions.' fh2 V2ceo = l .+ .. 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 . the case of ai with solution cei (of period 27r).24d) v 12 ' **To avoid confusion we emphasize: In Sec.• a2n(h2). A > b2n+i(h2)..2. A 2 y • ce2n=q0i.T + . 19. 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. and whether the function is even or odd.. 17.352 CHAPTER 17.. (17. 17. y > ce2n+i=qo. was treated in detail in Example 17. 2K. This is shown schematically in Fig. (17. y > se2n+i=qo.24b) 8 h2 z —— cos 3z + • • • . Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.. ce± = cos 8 64 h4 5h8 b2 = 4 1 . se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .2): hA 7h8 29h12 + Y l28^04.21a) and (17.
. Miiller [73].2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. Thus in Fig.24e) 17. We can see from the inverse of Fig.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig.3. of course. the walls are not infinitely high and hence tunneling occurs from one well into another.2cos4.3 17.3 we show schematically how the low energy solutions for integral values of the Floquet exponent are related to the asymptotically degenerate harmonic oscillator eigenvalues.hz — 12 . The main objective of the present section is therefore the calculation of the tunneling effect in the form of a splitting of the otherwise asymptotically degenerate harmonic oscillator levels j t Fig. The cosine potential cos 2z depicted in Fig.> (17. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain. In the case of the cosine potential. . 353 . . 17.2 The cosine potential. J.z3 ce2 = cos 2z . 17. 13824 .3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 .17„. 17. W.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.17. 17.1 that the boundaries of regions of stability merge to lines. 17. Dingle and H. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. B.
27) .3. of course. (17. as in the case of the cosine potential. 4 (17.26) into the original Mathieu equation and obtain y" + 2h q + rrxr .26) 17.i r ) ~ 1 — 21 z ± 7r Changing the independent variable now to the equation is approximated by A + 2/t2 dx2 + 4/i 2T y = 0. In the case of finite heights of the potential barriers.2 and 8. Taking the remaining terms of the cosine expansion into account.2/i cos z V = 0. Thus in this case the quantity on the left is not an exact odd integer. but only approximately so.2 The solutions We insert Eq. n 0. we therefore set 9 A A = 2h2 +2hq + —. where A / 8 is a remainder.. 8.354 CHAPTER 17.3) and is. (17.. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. Ion (17.2. and we set this equal to q.1.. simply the onedimensional Schrodinger equation for the harmonic oscillator. We therefore wish to expand it about these points. Periodic Potentials minima at z — ±ir/2. Sees.. there is tunneling from one barrier to the next. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf.
16/i/ (17. and substituting this into Eq.30) D A): COS « = 4z + 1 iq sin 2.3 Schematic picture of Mathieu equation eigenvalues. (17.17.29) is known. Choosing y(z) = A(z)e2hsinz. another solution. (17.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . 2 (17. 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).+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig. the linearly independent one.28) A" ±4hcos zA' + 2h[ q =f sin z \ A.27) we )A = 0. (17.31) We make the important observation here t h a t if one solution of the type (17. can be . z).
(9.4 ] .9) = 0.33) (17.9 + 4) = (9.9) (9. when i = 0 — the terms not involving Aq in (17. (9.Q)Ag + ( ? . (9l)(93).4 ) ^ .36) Then ^4.4)A.94) = (9 + l)(9 + 3). 2 /i D^Ag = 0. .356 CHAPTER 17. ^ ) A + ( ? .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. Aq+4 2Jh ._ 4 .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. (17. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. of course. With some algebra one finds that 16^ + where (9. 2[(92 + l) + A].30) amounting to Ri ] q = . ? . D{q%Aq+4i = 0. 1 + 4 ) ^ + 4 + (^. (17. 2 A A = («.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.32) cos^'+i) [\K . The solution Aq(z) of the first order equation Dq Aq(z) — 0 is c o s ^ 1 ) ( \K+\Z Aq{z) = sml(9+i) ( ivr + \ z ) COS 2 9 ( D I\n+\Z (17. (17.35) But since and so a term iiAq+n on the right hand side of Eq.e.94) \ — Aq4 (17. ? .\z To lowest order the solution A is A^> = Aq.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17. (17.g + 4) . i.^ 6T [(«>« + 4 ) A ? + 4 + (Q.35) not yet taken care of is set equal to zero.
(q. to satisfy Eq. Then for A = A(0)+A(D+A(2) + .4. q. (q. A(z.q + 4)(q + 4.q 4)Aq+4 q + 4 1 q + + + + + <^(.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. except those involving Aq.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ . Thus 0 = (g.17. +4. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. h) = A(z.. h).q + 4) (g.<J+4 Aq+& + 1 2 1 . A.35) and (17.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. q (17. .40) .. .4) (q . (17.37) and so on — left uncompensated — must vanish. h) = A(z. h).37) Here all the terms. (17.g + .38) Aq8 + 1 2 Clearly we can continue in this way. q .39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. q. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.g + 4)(g + 4. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. . q . the sum of the terms in Aq in Eqs. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.4) (q ._ 8 (17. Q . (17..)+ & ! = « ( . q.4. ) ^ 4. (17.39) which is the equation from which the quantity A and hence the eigenvalue A is determined.g4) 4.4 .4) 1 1 (q.8)A. q. q .q 8)Aq+8 (o)' ^±V(q 1 4.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. Thus in particular A(z.
Note different definitions in use.46) 2s(«. where Hen{w) is a Hermite polynomial when n is an integer.45) {B) We recognize Dq as the differential operator of the Hermite equation.d2B w dw2 + w(l w dB_ dw d dw 1.41) so that the early successive contributions decrease in magnitude. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). it is found that D(B) z+ — ]B = 0.44) where D(B) 9 = dw2 $_ !)• (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. We observe that for the large values of h we are considering here.) i ( Q . (6. (17. We return to Eqs. Before we study the eigenvalue equation in more detail.358 CHAPTER 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. The differential equation there is Eq. (17.e.43) 9 2eh . . 2 w2 + H* (17.42) (17.67). this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0.i ) We encountered Hermite polynomials earlier in Chapter 6. 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.28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( qsin Changing now the independent variable to w{z) = iVhcOS I 7T + Z I . and discussed there. i.
qi)Bq_4.e. also the same coefficients in the case of functions C below. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).e.17.50) C = 0.. which is an amazing and unique property of the Mathieu solutions — cf. A{wz (17.34). forms a rapidly decreasing expansion provided that \w (z)\ « Vh. y — Cexp(2hsinz) and Cexp(—2/isinz). Finally we obtain a third pair of solutions. with the equation C" + 4h cos z C + 2h ( q .. The solution with contributions B^°\B^>.e.q + 4 ) £ g + 4 + (q.. (17. (17. (17. . i. i.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. the expansion for the quantity A being identical with that of the previous case.e. i.47) where the coefficients are the same as before.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. q)Bq + (q.7 T H 1 2 Z 4 <1. 16hJ (17. as in Eq. 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.48) This shows that this solution is valid in particular around z = ir/2. 1 COS I .51) .4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized.3 ) \He* ! (9+1) H.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .
Example 17.7 T 1 2 Z 4 <1. 17.4)C g _ 4 .53) This solution is valid in particular around z = —7r/2. 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. q + 4)C 9 + 4 + (q. (17. each of them associated with one and the same expansion for the eigenvalue.360 CHAPTER 17. (17. A closer look at that expansion is our next objective.A Hen . (17. (17. These domains of validity are indicated in Fig. . We note that C(z.4. He'n(w) = wHen(w) Hen+i(w). Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. 2' the remainder 72. q)Cq + (q.54) where the proportionality factor is complex.59d) of Hermite functions Hen(w).(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . (9+3) (2n 2 + 2n + 1) + . we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).4: Hermite function linearization of perturbation Using the recurrence relations (6.q. (17.w2)He'n . Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. 2 J where n=(ql).(w2 + A)Hen.n2(n l)Hen2 .44) can be linearized with the same coefficients as in the case of the trigonometric solutions.h).h) ocB{z. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. ie 1 COS I . show that the perturbation remainder of Eq.q.52) and the coefficients are again the same as before.q.44). We have thus determined three pairs of solutions of the Mathieu equation.
g) 1 1 and so on. q) = 2[(q2 + 1) + A]. ^ 1 Hv" '^ ^' ^ ' . ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.q + 4) (q + 4.17.55) M2 = M. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. (b) The allowed moves are +4.3 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).3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K .l)(q . 2 + ^ M 3 + (17.0 and —4. (q. (?) 1 1 + (9.' 361 in 2*f*1>[i(gl)]! = (q. ) . with (. r j=r. 2)Aq_8} + • (17.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . q)Bq(w) + (q.)Bqi(w).i ) ( w ) 2l(.l)[l(q _!)]. (q.1i . q + 4) = (g + 1)(? + 3).57) .3. . e. We can write the solutions obtained previously. 17.q4) (17.4) = (q .56) (94. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.q + 4)Bq+4(w) + (q.g4)(q4.. l)Ag4 + P2(q. q .. in the following compact form with coefficients Pr(q.q 4. the solution y = exp(2/i sin 2)^(2).q + A){q + A.g. no intermediate move to or from q being allowed. . (q.3).
0) = 1 and all other P0{q.g + 4) 2 1 (g. l ) .g + 8)(g + 8.g + 4)(g + 4. 2) 1 2 ' ' ^ ' ^ 1 (g. i. j) +(g + 4j + 4.34) we deduce that Pr(q.P 2 ( g . (17. . j) = (g + 4j .g + 4) + 1 1 1 and so on.3) ^3(9. p = (<?><? + 4 ) ( g + 4 >g + 8) .j).j). the factor j removing a duplication of factors in the denominator at the junction.l ) . j . g + 4 j ) P r . . j + 1) (17.e. M2 = P 2 ( g .i (g.1) + (g + 4j.g + 4)(g + 4.j)Pr+1(g. Dingle and H. as jP r 2 (g.g + 4)(g + 4.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.j). P2(q 2) Pl(g. Miiller [73].Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j.j)] = 2j]jPr2(g.j).362 Clearly CHAPTER 17. g + 4j)P r _!(g.j). r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. From the coefficients (17.g + 4) ^3(9. Periodic Potentials PlM)=(M+i). (17.g + 12) 1 2 3 (g. Thus we have.g + 4 ) ( g + 4. J.60) M2r+l = = *R. and in general* M2r = J]j[P2(g.j)^2(9. g + 4j)P r _i (g.j ^ 0) = 0.58) with P 0 (g. W. for instance. B.g + 4)(g + 4. and the contribution from a sequence of r moves from q + 4j back to g.4._1) = Mzil.g + 8)(g + 8. 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. x)_ (g.j) = (l)rPr(q.
3. J. 17.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . . of course.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . as a result of tunneling between wells.4. where solutions overlap. [116]. Miiller [73]. h — —h. In fact we demonstrate in Example 17. B.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. [137].( 3 ^ . 17. Up to > > this point.* We observe that the expansion remains unchanged under the replacements q — —q. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. S.2 * +3) + ^ 2 (9g4 . a= (8/i)3(9x) [1(91)]! 1 .17. W. the solutions and eigenvalue expansions again have the same type of symmetries. It will be seen in Chapter 18 that in the case of anharmonic oscillators.92q3 + 70g2 . These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. Goldstein [115].^ . *E. Its precise deviation from qo. as pointed out earlier.62a) f R .61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. q is only known to be approximately an odd integer qo = 2n + 1. Dingle and H. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. remains to be determined next. Ince [136]. L.284g + 57) + (17. We observe that there are regions.5 the following proportionalities there: B[w(z)] = aA(z).^q(33q4 1 220 hA 1 (63g6 + 1260g4 + 2943g2 + 486) 225/i5 rg(527g6 + 15617g4 + 69001g2 + 41607) —3T76 (9387g8 + 388780g6 + 2845898g4 + 4021884g2 + 506979) —WTfs q ( 175045qS + 9702612 ^ 6 + 107798166g4 (17.
1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. The first term of the expansion of A(z) is the function Aq(z) given by Eq.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) .4 Domains of solutions and their domains of overlap.5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.32).364 and C[w(z)] =a~A(z).1 4 ._ ^ l CHAPTER 17. We write and expand this function as follows: .62b) *z domains of overlap Fig.7T H Aq(z) = = C O S ^ .COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 . Solution: We begin with the solution containing the function A{z). 17.1 ) ( 7T + . a . Example 17. and determine the constants a and a.(17.
i ) 1 (g.36).284g + 57) + • The other relation is obtained similarly. (8.!)(.g + 4) (q + 4. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. (6. 17. with e.g.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed. As we saw there.17. we have to impose the appropriate boundary conditions. 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 .22). these are conditions imposed on even and odd solutions at the point z = it/2. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq.e.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.63) In Example 17. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2. \ql)/2(w) 2i(9D[i(. These are the conditions of Eqs. the coefficient of the dominant factor c o s ( .i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a. with (8h)li''1) (3g2 .3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. i._i)]! w*te» 2l(91 (.46).2.92q 3 + 70q 2 . _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa. A*1) given by Eq. (17.53) for the expansion.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .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. (17. we obtain Bq[w(z)} H. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 .61) for the connection and Eq. s . (17. Since the solution with A is obtained from that with A by a change of sign of z.
.e.g. give the following expressions for polynomials with argument zero: ~19)"(^)!.» a \dw _ C[w = 0}c_2h a w=0 = Q (17. i. Magnus and F. With the help of the duplication and inversion formulas of factorials. a ± C[w{z)}c_2hsinz^ a (17. i.22) and to correlate the solutions to those for small values of h2. 80. W.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.e.62a) and (17. (17. The second of relations (17. The functions defined as in Eq.66).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\ ! ' . Periodic Potentials We can extend the solutions (17. and then replacing 2n by i>. Here v is the not necessarily integral index of the Hermite function Hev{w) with Hermite equation ( jir w—+v)Hev(w) \ aw1 aw J = 0.59d).64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.63) to the domain around z = ir/2 by using the proportionalities (17. p. Then y± oc ^ M e ^ i n . He'v(0) = Hev+1{0).62b). Oberhettinger [181]. He2n+1 (0) = 0. Thus y+ = ce and y_ = se. one obtains the first of relations (17.66) ^E. He'u{w) = wHev{w)He„+1{w).66) then follows from the recurrence relation (6.366 CHAPTER 17.
we have cot j ^ ( g ~ 1)} = ~\<(l .67c).67a) and (17.87 4 2 ¥w (17. .! ) ] ! [ . the solutions would be given by q = go = 3.67b) <f ~ 82g^ .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.67c) 4 2 \dw)w=Q C[w = 0] [\(q .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17. (17.106g .39 2~^h? (17.1) 7T .40g3 + 18g2 . and similarly for C and its derivative.» 2 6 /i the condition reduces to the equation cot{ .7. Hence expanding the cotangent about these points.g0)3].17.39 (17. and using the reflection formula (8.67d) Considering now the second of conditions (17.69) If the right hand side of this equation were exactly equal to zero. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.( g .136g + 9) + • • • .68) Inserting here expansions (17. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .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 .67a) g 106g 87 2 14 /i 2 (17.Qo) + 0[(q .82g2 .11. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .
65) yields the same result except for a change in sign.e. Considering now the first of conditions (17. = _a a 6 _4h (17.65) leads to the same result except for a change in sign. .40g3 + 18q2 .. Thus we have 2 (16/i) qo/2 T2 4.120g£ + 467g .136q + 9) + .368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .70) + ^ 2 (9?o .h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.. i.72) This leads again to the result (17.408g0 + 1089) 2 /i +• (17. . The remaining condition (17..70) except that now qo = 1. (17..40gg + 18gg .136g0 + 9) + • Clearly the last of conditions (17.5.(17. Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .528gg + 3307Q .40gg + 18gg .136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg . we use \(q)~\(q0) + (qq0)(^J .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .9.
e. In the literature that we follow here* in all cases several more higher order terms have been given. results from the boundary conditions. (17. The complete. n = 0. i. just as any solvable problem is important for the insights it offers in a concrete and transparent form. *In Sec. The dominant contributions were also found by S.e. (17. (17.3.. and that for the odd Mathieu functions se qo+ i and se9o (lower. Arscott [11]. 9o = 2n + 1. 17.87) = Azpteo). solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. J. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.61).. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.3. + ^ 2 (9<?o + 89o " 78g02 .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.88q0 . B.1. Dingle and H. quite apart from applications. The tunneling effect.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . this becomes 369 = (16/i)^+1e4h l . 120. . 20. though approximate. p. minus sign). example 3.74b) The above results for the cosine potential are in many respects important. . Goldstein [116]. i.17. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . Miiller [73]. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. and not the separation of these as a result of tunneling. W. See also F. M. We see here that *R.A_(g0) ^ . (17.t From Eq.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.3 we call this difference 26E„n. In the literature one finds frequently the statement that the splitting is a nonperturbative effect.2..
M±TT) . and one can compare the methods. . the Schrodinger equation with the cosine potential is a special case of several other more general cases. (17. Naturally one expects the solutions there to be related to those of the periodic case considered here. Periodic Potentials this nonperturbative effect.370 CHAPTER 17. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. As mentioned at the beginning.74b) is also important for various other reasons. results from the imposition of boundary conditions. The splitting can also be obtained with some methods of large order of perturbation theory. (17. The explicit derivation of the level splitting (17.63) — and using only the leading terms — derive the following set of equations with shifted arguments similar to those of the small/i 2 case of Sec.6: Translation of Solutions For the largeh 2 even and odd solutions (17. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). Example 17. y^{z) = Aq(z)e~2hsin *.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential. The exponential factor represents. . ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). in effect. yA{z) = Aq(z)e2hsin '. 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. Thus we do not reproduce every step. 17.2. From Eq. _3/+(z) + ____2/_(z). T y'_(0) y'A±*) y{z).74b). We begin with the solution containing the function A(z).2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . We consider for simplicity only the dominant contributions and replace throughout —1 to some power by exp(i7r) to that power and combine such terms into cosines and sines. Then we obtain first y±(z) = yA(z)±y^(z). evident through the exponential factor exp(—4/i) in Eq. 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.
q).T ( ? =F 1) p + (0). = ±ir). since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0). From the equations for the sine and cosine we can deduce the value of the Wronskian at z = n in the leading approximation for large ft: W[y+.2hAq(n)] . Aq(z) The derivatives can be handled with the help of Eq.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.4ft.31).3 The Ellipsoidal Potential Replacing VA. (17.z ) . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .Ah) ~ 32ft. A' (z = ±ir) = (. (l^ifr^i^O + Cl^ifo^j/aCO cos  .7 r ( q = F l ) } j / . Similarly we find y(z±ir) = = and by putting z = 0: V.4.( ± 7 r ) = c ° s  . we obtain the conditions stated at the beginning.4 17.[A'qM + 2hAq(n)] = iV2(q .%/2(g .VA DV 371 1 V±.4ft) sin j ^n(q . we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .y]z=„ = [y+yL .. y+(0)~2V2. and y+(z = ±TT) = V2(q . which is easiest by comparison with the foregoing treatment of the Mathieu equation. y'_ (0) = \/2(4h . but also to enable a brief familiarization with the Lame equation. Replacing sines and cosines by the functional expressions.4ft) cos i TT(<J + 1 ) 1 .) cos < 7r(q 1) [ = . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) . A ?q( Z = ±TT) = . The Lame equation has not been a widely known equation of mathematical physics so .yV+\z^ = [j/+(0)] 2 (<? . 17.• Then y'+ (0) = 0.c o s  ~(<1 T 1) f y .1 )  .( 2 ) . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). ( . where Aq(0) = Aq(0) = V2.7 r ( g ^ l ) U + ( z ) q : i s i n { . etc.g A ? ( z = ±TT).( z ) From this we deduce for z = 0.17.4ft) ~ 8(q .
§ F .5.0 and .. MullerKirsten and D. 17. 194) remarks to parallels with the Mathieu equation:".. as alluded to at the beginning. so that Floquet's theorem cannot be immediately applied . *A. however. f E . If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). 194) remarks. k) for k = 0. This line of investigation. J.4sn4u]y = 0. L. 19.fi2A.0.98. p. W. M.5 The potential sn2(z. Ince [138]." Fig.75. 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 .Q. Liang.* As Arscott [11] (p. and was continued by Erdelyi. Periodic Potentials far..75) ' T h i s means double well. the nonoccurrence of the Floquet exponent. H. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . but also cubic potentials. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.0. ...1.0. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. (17. inverted double well and cosine potentials (J. The most conspicuous difference compared with the Mathieu equation is. But more recently it has been observed to arise in various contexts.5 < z < 5.K2sn2u . Tchrakian [165]). Arscott [11]. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials. Arscott [11] (p. Erdelyi [85].372 CHAPTER 17..* and the work of both prepared the ground for presentday investigations.. H.
the equation ^  + [A . For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential.1.77) into Eq. . If we put fl = 0. The range of the independent variable u is 0 < u < 2K. in the case n2 — oo.4. The first step is to write the eigenvalue A as A(q.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. (17. and of the derivation of the level splitting'. K being the complete elliptic integral of the first kind. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k. The second step is to insert (17. N = 0. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). J. .78) We follow in this brief recapitulation the description in H.75) reduces to Lame's equation and one writes K.5 for several values of fc. The function sn2u is plotted in Fig.2 = n(n + l)k2. . the usual factor —h2/2mo. (17.17. is the elliptic modulus of the Jacobian elliptic functions snu. 2 .K2sx?u}y = 0.76) 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.e.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. i. In order to distinguish the above equation from that with the Lame potential. . in front of the second derivative has to be kept in mind (mo being the mass). i. In comparisons with the Schrodinger equation. (17. Eq.2k.e. Although the results have been calculated for the ellipsoidal wave equation. (17.K) = qK+^±. 17.77) where q —> go = 2iV + 1. (17.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. Jianzu Zhang and Yunbo Zhang [206]. W.\k\ < 1. MiillerKirsten.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). we consider mainly the Lame equation. cmx and dnu.
determine their proportionality factors) in domains of overlap (their extreme regions of validity). A. ^ (17.82 . one has to derive new sets of solutions there and match these to the former (i. B and C. A are derived. dnu± cnu K Thus one can construct solutions Ec(u).K — > A(u)=A(u + 2K). (17.80) The domain of validity of these solutions is that away from an extremum of the potential. dnu =F cnu . one pair in terms of Hermite functions of a real variable. _ Solving the resulting equations iteratively as before. . one obtains again the same expansion for A.. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. more precisly for dnu = cnu F 1 » .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.374 where CHAPTER 17. (17.K). .Es(ii).e.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed).q. (17. / z J . \/8K (dmz = cn« \ ' F .77). A second solution is written y = A(u) exp < / nsnudu >. but solutions B. the other in terms of those of an imaginary variable. by transforming the equations for A.e. Thus in the third step two more pairs of solutions B. A into equations in terms of the variables z(u) = ^—{k' . C replacing yl. k' = Vl~ k\ \dmi±cnu/ 17.e.q. A(u. Periodic Potentials . i. C which are valid for dnu±cnu <1. /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. i. which are respectively even in u (or snu) or odd.K)=A(u.
i. 17. Miiller [205]. (17. a of _ _ B = aA.e. Oberhettinger and F.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} .3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.85) The first three terms of this expansion were first given by Ince [138].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.6(5g4 . C = a~A.17.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. This expansion is found to be in the more general case of Eq. J. *A. Erdelyi.e. as well as \duJ2K 11 (17.4. F. Es(2f0 = Es(0) = 0. . (17.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. (17." A = qK±(l 1 + r {(l k2)(q2 + l)^{(l + ifc2)3(5g4 + 3 V + 9) + k2)2(q2 + 3)4k2(q2 + 5)} 2WK2 Ak2{l + k2)(5qA + Uq2 + 9) .^ ^ { ( 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£. W. £2 = 0). W. who obtained this expansion for the eigenvalues of Lame's equation (i. 64.38g2 . p. Tricomi [87]. the ellipsoidal equation. Magnus.63)} + ••• .75). G.384n2k4(q2 + 1)} q r {(l + A.
2 /c +256fc2g0(g2 + 5)} (17. go being an odd integer. Finally expanding A(g) ~ A(g0) + ( g .78g^ .88g0 .Ec£ 0_1 and Es2° of periods 4K.Es^ 0+1 .:J {3(1 + k2)2 (9q* . 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.i Here the upper sign refers to Ec*> l or Ec£°.90) + 3.l n 2^+ i J + 0(«i) (17.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .89) one obtains the eigenvalues from which the level splitting can be deduced. J. and the lower to Es*>+1 or Es*5.j 26K2 (1 85) = A(g0) + (q .136g0 + 9) n 3. (17. . from which the difference q .87) (17.91) tt H.go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .4k2(q2 + 2g0 + 5)} + • • • ]. 2K and 4K respectively. Muller [205]. Periodic Potentials These conditions define respectively functions EcX°.376 CHAPTER 17.g 0 ) ( ^ .g0 is obtained by expansion around zeros.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)*. 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). 2K.40g03 + 18g02 . + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ . W. 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 .
4 h ( i e M 9 0 7 2 \. *H. V. but also more easily generalizable.2h2 = A. J. i. the conditions (17. One can see.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods.90) one has in this limit: /l±±\ ~K/U = e .4.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. . h . A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. it = x ± ^ . Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. (17. this equation becomes y" + {A . Replacing u by x ± IT/2. A . in which they dubbed the classical configurations Lame instantons.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. Ince [138]. Thus in the case of the dominant contribution of Eq. K = 2/i.* Apart from the choice of notation. Miiller [206].Ah2 sin2 u}y = 0.2h2 .74a). Hence the conditions fi = 0.* M G . (17. (17.93) (17.e. however./2i and hence A^ A/ A AU I2 .f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«.87) agree with those of Ince. L.2h2 cos 2x}y = 0.86). 17. (17.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation.17. A = 0. K2 = h2 (say). W.J1 O. 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. . then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . + E. Rao [79]. Dunne and K. that calculations with the Schrodinger equation are not only easier.
W. for instance. S. double well.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. **Y. A. A. This is an interesting problem. H. H. 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.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). Rana [287]. Shiraishi [49]. inverted double well and cubic potentials. . W. See Chapter 25. Ouvry [276] and Jianzu Zhang. H.g. all basic potentials.5 Concluding Remarks The potentials we considered above will reappear in later chapters. Quian [123]. Khare and U. Ganguly [103]. J. Periodic Potentials 17. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. M. J.W. Thus.^ The equations considered above also appear in diverse new problems of physics. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine.Y. Gu and S. Tchrakian [165]. N. See e. Sukhatme [148]. A host of related elliptic equations has recently been discovered and studied. D. de Veigy and S. A generalized associated Lame potential has been considered by A.e. § .378 CHAPTER 17. 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. Liang. MiillerKirsten and J. MiillerKirsten and D. Cho. 1 1 See Z.Q . Kan and K. J . i.
Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. The recent work of R. [19].. J.". 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. Achuthan. This seemingly simple problem revealed extremely rich internal structure .1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. M. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M.. J.Chapter 18 Anharmonic Oscillator Potentials 18.* 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. D. Bender and T. MiillerKirsten [163]. 379 . In particular the investigations of Bender and Wu.Q. both of which make the calculation more difficult. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Liang and H. T. f C . W. ' T h u s A. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. Wu [18]. H. Wu. T. Weinstein [281]. Wiedemann [4] and a revised version of parts of this reference by J. MiillerKirsten and A. Priedberg and T. W. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Bender and T.
and thus lead to very different physical situations. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [EV(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . V(z) = \\h*\z* + \\<?\z*.380 CHAPTER 18. These contributions may be given different signs. 18. V(z) = \\h*\z2 I 2i 4 \cr\z .1. with the potential described as an inverted double well potential. (3) Complex eigenvalues with tunneling: In this case. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. 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. *~z (1) (2) (3) Fig.1 The three different types of anharmonic potentials. 18. described as the case of the double well potential.
18.e. We begin with the latter. The result is an expansion in descending powers of h?. . since the potential decreases without limit on either side of the centre.1 Introductory Remarks 381 normalisable wave functions. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. The eigenvalues of this case are given by Eq. It is this expansion which led to a large number of investigations culminating (so to speak) in the work of Bender and Wu who established the asymptotic nature of the expansion. except for a change of sign of c 2 . In this case our aim is to obtain the aforementioned complex eigenvalue. as is sometimes hoped. The level splitting will be rederived from path integrals in Chapter 26. i. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. however the central hump with troughs on either side permits tunneling and hence (if the hump is sufficiently high) a splitting of the asymptotically degenerate eigenvalues in the wells on either side which vanishes in the limit of an infinitely high central hump. The eigenvalues of this case are given by Eq. it will not be possible to obtain the exponentially small contributions with convergent expansions. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). The imaginary part will be rederived from path integrals in Chapter 26. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). 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. that the general applicability of the method becomes evident. Case (3) is seen to be very different from the first two cases.86) below. Since these exponentially small contributions are related to the behaviour of the late terms of the eigenvalue expansions (as we shall see in Chapter 20). The shift of the eigenvalues is best calculated with straightforward perturbation theory. 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. (18. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. (18. Thus we are mainly concerned with the double well potential and its inverted form. however with decay as a consequence of tunneling. 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 17.175) below. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined.
The potential in this case is given by V{z) = v(z).2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. Anharmonic Oscillator Potentials 18.2 The inverted double well potential with (hatched) oscillator potential. 18. This implies t h a t results for THQ = 1 (a frequent convention in field theory considerations) differ from those obtained here by factors of 2 1 / / 2 . (18.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. 18. If suffixes 1/2. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.2.1 and more specifically in Fig. 18.2 18. v(z) = ^h4z2 + ^c2z\ (18. h 4 2 2h1> ~1/2 2c?.1) for h4 and c 2 real and positive.3) "1/2 harmonic oscillator h8/25c2 . (18. . We take here h = 1 and the mass rriQ of the particle = 1/2. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.382 CHAPTER 18. a point which has to be kept in mind in comparisons.1 refer to the two cases.2.
although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. _i\(w) and q = qo = 2n + l. i. 18. (18.1/7  Vg(w)y(w) = o(J^j. The result will be that derived originally by Bender and Wu. the harmonic part of the > potential dominates over the quartic contribution and Eq. \h?\ — oo and c finite.. The problem here is to obtain the solutions in various domains of the variable..4) we can rewrite Eq.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. (18. /i). y{w) oc Di.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.e. 4 . then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.2. (18.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. (18. 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 (18.7b) 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 = 0.5) becomes . and a variable w defined by setting E= b* 1 + A A and w = hz.2. (18.1.18... n = 0.d2 . to match these in domains of overlap.
h) is obtained from the perturbation expansion of the eigenvalue.4(2.2 A (18.384 CHAPTER 18. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.14a) 1/2 VA( ) = A(z)exp (18.11) 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 ) . In order to arrive at solutions we set in Eq. be and A = A(q. as encountered and explained earlier.2 We are concerned with the equation d2y(z dz2 where ^ 1 .+ ^ . y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.9) we obtain 1 h4z2 c2z4 A L2 (18. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. (18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. these solutions are not valid around z = 0.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18. Inserting (18. •A~2 hrz" 2^\ c^z V2 —— + (18.9) z =0 these Thus these c2z4 1/2 + • 2 (18. This observation allows us to define the pair of solutions yA(z) = .14b) .10) into (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 = 0.2. (18. We observe — before touching the square roots in Eq.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18. as we shall see.10) E= qh2 ^ j. 1 d r A!{z)±iA{z)±{ 0.
(18.20b) . We take the square root by setting z2h^1/2 +. 1 2 4 2 Aq(z) + 0 (18. (18. (18.zh2 =()*2"(18.4 .19) We see that one solution follows from the other by replacing z by — z. Since these higher order contributions are of little interest for our present considerations.1 and as a verification again in connection with the solution y# — as the other solutions.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.18.15) where A{z) is the solution of the upper of Eqs. /H 4 Z h +~C Z 21.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.16) For large h2 we can write the Eqs. (18. Clearly Aq(z).13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0. we do not pursue their calculation. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq..34) below) — to be derived in detail in Example 18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. i.e. 1 (18.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h.Aq(z) approximate the solutions of Eqs.13) and A(z) that of the lower of these equations.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18. (18.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z). / c z Aq{z)+0 1/2 . 385 (18.
(18.17) and (18. h2. (18.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. W i t h proper care in selecting signs of square roots we can use the solutions (18..e.• Using Eqs.386 CHAPTER 18. a2 = 2 1 .2 . one observes that with £>.:=*£ +i(5l). (18.. i.20a).! ) ^ . h2..21) Example 18. we write > V±(z) = .18) we obtain A » = H^rL^z) 2 z .20a) and (18. X » / A q + 4 i = Mg+4i . 2 . ft = . „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. A A _ 3 . (18. A = .20b). 04 = 5 ..f . zA'(z) where the expansion coefficients are given by ao = l. . c*i = 1 1 .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).2 ( * ) 2 and from this or separately A"(Z) : 1 (q .l)A(z) 1 = tfA"(z) A .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . 1 . * = — .[yA(q. . to obtain in leading order the eigenvalues E.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). Vq+ii = Vq + 2i. 03 = . h2 —> —h2). Anharmonic Oscillator Potentials i.e. „ . z)). (18. 128 8 16 A = i. procedure.v ' = o (« . i. away from the central minimum.e. z) ± yA(q.
In the next two pairs of solutions the exponential factor of the above solutions of type A is contained in the parabolic cylinder functions (which are effectively exponentials times Hermite functions). Rq '. _1J±w) and D_\.5).23) . when i = 0. (18. [1(9 + 1)]! . ^f( g + i)(±H2^ + 1 > \ — 1. r ( CJw) = ^ . (18.w — > > ±iw) or functions R B w q( ) .18. of course.. w = hz. i.e.. give an equation from which A is determined.—77 [i(9l)].. (18. 2ft4 + e)Av —rAq+i q+8 6 2/i + \+ The sum of terms with Aq in Rq are calculating here „2 must then be set equal to zero. The solutions yq(z) of the equation Vq(w)yq(w) = 0. (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^ .22) are parabolic cylinder functions D i .22) is invariant under the combined substitutions q — —q. +1J±iw) (observe that Eq.2 The Inverted Double Well Potential Thus a term /j.'2H9i) ^faDfr™) and . Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . those in Aq(z). and the coefficient of terms with i = 0.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. We return to Eq. In this way we obtain the next order contribution A^1' to A^°>. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B..(q + 3)(q + l)Aqvri(q '^ ' . = •>— r.
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 + .29) ' A .Aj)(18. pp.27) Now.q + 4j}yq+4j. (18. Anharmonic Oscillator Potentials The solutions Bq satisfy the following recurrence relation (obtained from the basic recurrence relation for parabolic cylinder functions given in the literature^) w2yq= {q + 3)y g+4 + qyq + (q .( < ? .123. (18. Actually these factors can also be extracted from W K B solutions for large values of q. j=i (18. F.3. As an alternative to Bq(w) in Eq. we also have Vq+^yq+tj = 0. Comparison with our notation is easier if this reference is used.28) (A + c w )yq(w) E 2 4 1  2 ^ [q.0) = = = ±(q±3)(q±7). and for j ± 0 : [q.±4) S4(9. since Vqyq = 0. See Example 18.0).q}=A + c2S4(q. (18. Oberhettinger and F.l)2/?4. Magnus.4j)yq+Aj. . G.^ For higher even powers of w we write i w2l V<i = Y. 115 .23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. (q±2)(q±S). (18. W.3)y g _ 4 . S2i(q. Erdelyi.25) where in the case i = 2: SA{q. Tricomi [86]. l(q2 + l).26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.±8) 5 4 (g.24) The extra factors in Eq. and so Vqyq+Aj(w) = Ajyq+Aj(w). (18. Vq+4j = Vq + 4j.q + 4j] = c2SA{q.388 CHAPTER 18. (18.
3 °) For the sum y{w) = y(°)(«. h ) = qh .(!) («.10) we obtain E(q. (18. Thus the next order contribution to yq is y H = ie 2^ —z. (18. .) ty^^w) to be a solution to that order we must also have to that order [q.q]=0. .• q.18. n — 0.e.—v*+*j ( 18 .2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. and c 2  replaced by — c 2 . In Case (3) the parameter q is only approximately an odd integer in view of tunneling.1. This type of invariance is a property of a very large class of eigenvalue problems.—A(q + 1) . .34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. p i(9»9 + 4 J'WK7».[rp ) Y.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.31) Proceeding in this way we obtain the solution y = yW („. (18. (18. so that the entire expansion is invariant under the interchanges q .) + j. Equation (18.32) Evaluating this expansion and inserting the result for A into Eq.35 ) . A = ^(q2 + l)c 2 + o(J^.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.e. 2 . and even powers of q in combination with even powers of 1/h2. h2 —> h2. We can now write the solution y(w) in the form y{w)=yq{w) + Y. ( 18 . i.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. (18. i.
e.q±A) = MM!i±MM+[M±8][g±8. MiillerKirsten and A. o. i. w — ±iw. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1." Since our starting equation (18.i 1 + 4j) in complete analogy to other applications of the method. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. We observed earlier that these are obtained by making the replacements q —> —q. H. ^VO(<L<?) ^i(?. and for j ^ 0 all other P0(q. Achuthan. 2 4tPi{q. Wiedemann [4] . (18.38) [yB(z)]argz=7T = [?/s(2. there is another solution y(—z). q + 4j) = 0. > "p. We thus have the following pair of solutions VB(Z) VB(Z) = B w .g±4] ±4 ±4 [g. their recurrence relations and the solutions of the latter we refer to the literature. q + At] (18. q) = = 1. we may infer that given one solution y(z). q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. J.390 where for instance v J CHAPTER 18.—. ? T 4 ] [ g T 4 . / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. W. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q.11) is invariant under a change of sign of z. g ± 4 ] T4 ±4 and so on. Again we can write down a recurrence relation for the coefficients Pi{Q.37) For further details concerning these coefficients.36) with the boundary conditions p o(q.argz=0 (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.
(18. (18.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. one has to stretch each by appropriate expansion to the limit of its domain of validity. with normalization since the normalization constants are also asymptotic expansions. Integrating and expanding as follows since h? is assumed to be large. .40) "Variables like those we use here for expansion about the minimum of a potential (e. Such contributions arise.q + 4j) as in ys. In order to be able to extract the proportionality factor between two solutions. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.g. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. the solutions of type A being valid away from the minimum.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. 4 . Vc(Z) = [V!B(Z)}q^~q. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. in fact.2. h2) in which odd powers of q are associated with odd powers of h?.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles.39) are with the same coefficients Pi(q.g.z2 . J. Aq(z)) does not appear in any of the higher order terms. 18.3 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.h^ih 391 (18.The solutions yc. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.g. Thus in the transition region some become proportional. Mauss [192].18. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.h^ih.4)). like w of Eqs. see e.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 .
Comparing the solution VA{Z) of Eq.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (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. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > .(«. 4/T 4 Prom (18.42) H W2 (ql)e$W I „ . Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.392 CHAPTER 18.2 « . eh* /12c* e\z*h? (18.z).44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.4 i . with z in ys(z) replaced by — z. We do not require this at present.42) we obtain for the solution ys(z).43) W + 1)]! w^+V A . (18. (18.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.44) In the solution y~g(z). 2 ^i![i(g4il)]!(2«. .45) . y^(z) we see that in the direction of z = 0 (of course. The function Di. we would have to substitute correspondingly the expression (18.43) (since z — — 2 implies > argz = ±7r).l ) ] ! ( .7 r . > n.41) with the solution ys(z) of Eq.0x ! [ £ ( g . Anharmonic Oscillator Potentials Considering the pair of solutions ju(. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. (18.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw..w yB(z) ~ B.
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.49) d yc{z) Again there is no such simple relation between yA(z) 18.41). 18.50) and y+(0) ^ 0.yc{z)pansion (18.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 . 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. Looking at the potential we are considering here — as depicted in Fig. The first of the conditions (18. we see that in their common domain of validity yA(z) where a = =Vc(z)..50) will be seen to imply qo = 2n + 1 = 1. hence we have to .e. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18. At z = 0 the solutions of type A are invalid.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.18.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq. and the second qo = 3.5.7. (18. (18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.21) as even and odd about z = 0.1 ) e 12c 393 Kgl)]^"1) l + O h? (18.y'_(0) ^ 0.. i. the ratio of yA(z)iVB(z) i s n ° t a constant. We proceed similarly with the solutions yc(z).9.46) However..42) into y~c(z) w e obtain Vc(z) = (h*. (18.11.2 — 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. Recalling the solutions y±{z) which we defined with Eq. large probability for the particle to be found thereabouts.2.
Abramowitz and I.1)('») + ^ r + l)]! ^. Example 18.^{(9+1)}.(0) + ^ ( 0 ) l z—>0 2. [C.. one finds that [ 1 IT (q + 1)1! = . [ .(«.s i n { .51) and 0 = V(0) = Jim i»xM . (18.l)(°) Thus with the help of the reflection formula cited above: D B. C."*(«!) ^ ' ~ [ .(iu)].(0) = .(0) = and htoi)(°) ^—'.5 ( 9 + !!)(*"')• i ( 1) .SxW) = \ i»fl(0) ..! ( . IT 4 dio In fact.F ? .56) D 4(.. ( 9 + 3)]! s i n { . . we obtain i V7T24( g .394 CHAPTER 18.54) dio q(w) . Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. y c ( 0 ) .(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}.i ( g + 3)]l (18. Stegun [1]. Expressions for C 9 (0). = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18.55) Solution: Prom the literature.) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.i ) . with the reflection formula (—z)\{z — 1)! = IT/ simrz. — C. We leave the detailed calculations to Example 18.(0) (18.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.2.( g + 3)}.57) [i( 9 _l)] ! 2 i(9i) B.oi(.l ) ] ! [ .2: Evaluation of yB(0). A. M. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i. 1 [7(93)]! n «(°) = ./o^Tpif (q+i) . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.i)H = e.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18. Z (18.' '.fc(0) Thus we obtain the equations (18. + 1)]!' D ' j .58) diu [  ( g .g.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. e.( g + 3)}. y'B(0)..
~ ^ 1 5(9+1) 2i(''. (18.1/2)!. (18.62) ._l) From this we derive C. isin{f(g + 3 .56). _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. (18.49). W .i ( « + i ) [ _ i ( .6 ) } ^ — i t a n < — (q + 3) >. 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) _^_ .sin{^(g+ 1)} = J .+i)(°) [ _ I ( .60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.sin{^( 9 . + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.)[i(g3)]! and ^i(. we obtain 2 i(»+i)[_i(.+1iH]«=o = .(0) = *>i(„+i)(0> 2 .i .18. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.3)}. using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .53) in dominant order and insert the appropriate expressions for a and a from Eqs. Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .46).2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.
7 .2) with potential (18.9. With a Taylor expansion about go the left hand side of (18. z2 > z2. (gg 0 )^±^ y e5?. (18.1).L ^ i . z^ ein/2z.65) Expanding similarly the left hand side of Eq.11.5 B o u n d a r y conditions at infinity (A) Formulation of the boundary conditions We explore first the conditions we have to impose at \z\ — oo.5.2.^ >^_>m e J*. In fact the left hand side of (18. Our next task is to extend the solution all the way to the region beyond the shoulders of the inverted double well potential and to impose the necessary boundary conditions there.i . ^ .. Recall > the original Schrodinger equation (18. 7.^ .63) vanishes for q = go = 1. . 18. (18. ._ e =*.64) for q = g0 = 3.. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.go) ^ (1) It follows that we obtain for the even function with q — go = 1. + 3)} = .63) becomes (? .65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0.63) and (18.64) about go — 3....64) In each of Eqs.9o) ^ cos  .. . For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.. i. the vanishing of the wave functions at infinity). . this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. the energy E is real.. and the left hand side of (18. we obtain cos((g + 3)) = . .396 CHAPTER 18.( g 0 + 3)  + • • • ~ (g . 18.e. (18.9. (18. (18.53).53) becomes 8 m(I(..5..1. This is Case (1) of Fig. equivalently. (18. Thus we have to determine these conditions first.. 1 1 .63) Proceeding similarly with the second of relations (18. we again obtain (18. by the rotations E > einE = E.64) the right hand side is an exponentially small quantity.
Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. i.67) Rotating z by vr/2. we set z w r = \z\eie. (18. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. if TT/3 < 6 < 0. (18.cos 36.e. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. 2 c2>0.e. 1/2. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. in the domain — TT/2 < 36 < > 7r/2.68) . or — < 6 < . replacing sin 30 by " 0 + ^) 1 IT = . 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. y{z)~exv\i[) 7T dV'V.66) z3 = \z\3em.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. the resulting wave functions vanish at infinity and thus are square integrable.e. r / c 2x 1/2^3 = exp{ ± i — —\.e. Now. the case of complex E).2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — .+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. —}.e. i.18. i. i. 6 6 In the case of the inverted double well potential under consideration here (i. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s.
18. We therefore explain our procedure first. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.398 CHAPTER 18. We have to remember that we have various branches of the solutions y(z) in different domains of z. Anharmonic Oscillator Potentials V(z) Fig. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. for z > Too. Eq.21)) were defined in terms of solutions of type A which have a wide domain of validity. Our procedure now is to continue the even and odd solutions (18. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. we match the WKB solutions . We do this extension with the help of WKB solutions.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. This is the boundary condition also used by Bender and Wu [18].68) will lead to our second condition which together with the first obtained from boundary conditions at the origin determines the imaginary part of the eigenvalue E. (18.3 The inverted double well potential with turning points ZQ.68) for c 2 > 0. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. i.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).Z\. Looking at Fig. and there impose the boundary condition (18.21) to + infinity and to demand that they satisfy the condition (18. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. Our even and odd solutions y±(z) (cf.Z\. (18. 18.e.
*R. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18.^ 4 1/2 + icV 7T Jz + (18. 18.70) where z < z\.71) We now come to the algebra of evaluating the integrals in the above solutions.^ T rz\ 1 + yh* . Dingle [70]. ^ 2 4 •K + —(r. t To the right of the turning point at z\ these solutions match on to (r.2. Sec.70). 291.zi)/ \ 1/4 ywKB\ ) z — g/i* . .3 is given by (using Eq.69) The W K B solutions have been discussed in Chapter 14. 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\. Messiah. B. z2h4/A > c2z4/2. p. h2 (18.e. ^2? 1 2qc2 /2? . The distant turning point at z\ as indicated in Fig.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . (18. (18.*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) .e. Dingle [70]. I [195]. 4 2 T ' i. equations (21).4 h + c z 2 1/4 < . i.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. Vol.4.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . f R . z\ ~ . 6.4) for E and ignoring nondominant terms) t o .18.B. 291. p. A J24 1 ? 2 • "^ •z h \—c z ~ — q h .. (22) or A. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. We begin with the exponential factors occurring in Eqs. 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.
^ i ^ ^ c 2 . (18. Eq.2.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18.^ I 2 2 1/2 zV = F 8c2 Jz [i .e.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.400 i.\qh2 + \z2h± . Thus the above factor yields '2c2\l/2 h4 so that (cf.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.73) .211/2 1 \ /j.. Eq.40)). In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c. (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 . We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). CHAPTER 18.2 \ 1/2 . .i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >. Anharmonic Oscillator Potentials E± = exp = exp .2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . (18. zi 2q 1 c 2 [1 . (18.2.40)) E± = exp _/^_2 f ± In A 2c2 .
18.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.74) Comparing these solutions now with the solutions (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.*l) (18.77) Now in the domain 2 — oo we have > \qh2 .1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.(J*i) VA 0 ) = PVwKB (Z C.*l) (3y^(z) (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) . h2 — — h2 is maintained.79) +S_(±)exp .2 The Inverted Double Well Potential 401 Thus at the left end of the domain of validity of the W K B solutions we have (l. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.78) Inserting this into the solutions (18. (18.20a) and (18.76a) P P r 2/1 2 " 9 ( .20b).2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.76b) apart from factors [ l + 0 ( l / / i 2 ) ] .71) and these into (18. > » Returning to the even and odd solutions defined by Eqs.\z2h4 + J Z\ nl/2 \M fZ .21 1/2 .
e..81) Inserting expressions (18. i / 3 ± . Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. i.h2) + {qqo)(^pj = E(q0.. (18. (18.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1. (18.68)..65). ./i2) = E{qQ.85) + ./ ? = 0.6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .h2) + (qqQ)h + (18.2. (18. E(q.5.76a) for (3 and /3.^(q2 + 1) . (18.82b) This is our second condition along with Eq.34). (18..402 CHAPTER 18. (18. Eq..e. h2) = \qh2 .^ j . i. (18. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J .3.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq. we see that we have to demand that 5 + ( ± ) = 0.^ § ( 4 g 2 + 29) + O ( ^ ) . 18.84) Expanding about q = qo we obtain £(g.
o 2 + l)^(4. ft6 ^ = e.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z. S W K B ( 2 ) = PVA(z). The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. where the superscript {r.36) of their Ref.18. Since this is not without interest (e. zo} means "to the right of the turning point ZQ" . 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 . above) (1. h2).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.yA{z).g ( . Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. !*WKB(Z) ~ 2 .~p{q.2 .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .ZQ) . Example 18.g.3. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .e. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. 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. i. h2) of the relations KWKB (Z) =PVA(Z).h6/6c2 E = E(q0.86) will be dealt with in Chapter 20. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18.l ) ] ! > + = ^ 2 .
On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. ./ ^ 2 X 1 / 2 exp(ifi.. the (approximate) solutions* 1/2 exp • ± i ( . Schwind [247]. cf.g In other words. 2 z 2 ) / 2 e \ .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.O.1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation .] { hz combined with the particular constants contained in y ^ ' ^ g .zo)/ \ qh2 2 ZO + z2hA 4 2 . N.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.zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . 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. 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. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .ic224 2 4 qh2 + i^fe 4 . f x Zn 4) h2 2. Anharmonic Oscillator Potentials where zo > z.
DfJ»v we see Comparing now the expressions (18.13) of Bender and Wu [19].18.g.23)) appear quite naturally._ .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.3 18.5 "l _ o f . J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .z ± ) = . We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.(r.zo). i. in fact.41) for yA(Z). as in the case of the Mathieu equation. (18.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. 4" ) (h?z2)^q+1) for q j£ odd integers." / 4 ( ? « / 4 . we shall employ basically the same procedure as above or.e.kyil(. (18. h4 > 0. 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.1)! = ( 2 7 r ) 1 / 2 e .yA{z) with expressions 27WKB i J'WKB ' that the factors p. „ / M 1 / 2 i . The last expression can be made acceptable for q = odd integers by applying Stirling's formula in the denominator of the second of the previous pair of expressions.J ' i J .87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. (18.89) . We can reexpress these relations with the help of Stirling's formula.3. Then .^ 1 . But there are significant differences..88) + [EV(z))y{z)=0 (18. (j . in Eq.
z±).95a) .92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. .90) V(i)(^)=0.V(z±) l{z z±fhA + 0[(z . (18.4 The double well potential. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2. (18.z±) y = 0.91) (18.94) (18. 18.93) w± = h±(z .z Fig. vW(z±) = ±Qch2.i>5. (18.406 CHAPTER 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 . (18.91) becomes T>q±(w±)y(w±) = o(^)y.
3 The Double Well Potential where ^ 407 d2 •« 2 2  d8. n = 0.96).97b) Our basic equation..95a) and (18.3.97a) U(z) = (zz±)2 + 0[(zz±)3}.. 18.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.2 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 (18.93) for E into Eq.18. (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. given one solution.. (18. d2u{z dz2 + v+ 2~r u(z) = 0.95b) By comparison of Eqs. (18. Eq. (18.100) .. is again seen to be invariant under a change of sign of z. Inserting the expression (18.1.87). (18. we conclude that in the dominant approximation q± is an odd integer.96) dz2 + 5 « 4 +  .i 4 f W » = o. i.e.95b) with the equation of parabolic cylinder functions u(z) = D„(z).2. (18. we obtain another by replacing z by —z. Thus again. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)]. we obtain d*y (18. qo = 2n + l.
408 Since CHAPTER 18. Integration of Eq. h2.c W + 1) .e.105) "8 + 4 ^ .102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. (18.e.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —. We leave the calculation to Example 18.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. The result is given by A and E(q. z).h2. 8hw (18.^q(l7q2 6 4/i + 19) + (18.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. h2.h2) . 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 ) . (18. (18. i. (18. Vq = {z2+z2)— + (qz+z). z) = yA{~q.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .4. yA(q. Aq(z) = Aq(—z) = Aq(z). these equations can be rewritten as + (qz+z)A{z) = .101b) To a first approximation for large h2 we can neglect the right hand sides. (18. h2.104) Both solutions yA{z).101b).101a). z) may be obtained from the solution yA{q.103) Looking at Eqs.yA(z) are associated with one and the same expansion for A and hence E.^ A"{z) + 2c V2 *A(z) (18. i. z) = yA{q. we observe that the solution y^iq. h2 h4 2c' with g+ = q. ~h2. but the solution is a different one.106) . n± = V2h2.
102).e.+ 2 ^ ± l + 2^±i + . for instance by componendo z z et dividendo.114a) . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.109) We observe some properties of the function Aq{z) given by Eq.z2 ) N t.108) We wish to rewrite this expression as a sum y coefficientiv4.Aq 1 + 2^± 2 Aq+2 (^M^f (18.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z . We know the first derivative of Aq from Eq. J = 0. (18.qz+) [2z2 .113) 2 .4zz+q+ z\{q2 + 1)]. we reexpress A'q in terms of functions of z multiplied by Aq.106). (18. however.103): . The first such step would be to reexpress the right hand side of Eq.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.qz+)2 .{z2 . i.3 The Double Well Potential 409 Example 18. = Aq+2i+2jAq.2 z\) • Aq$4 £Aq + Aq—4. . First.112) From these we obtain. A Differentiation yields » = ^^w (18.(l + + 2  z y* . z • . as a linear combination of terms Aq+2i. . + Aq+2 + Aq Aq+2 . . 92 ZZJf (18..18. Thus it is natural to explore analogous steps.101a) with A replaced by A . such as periodic potentials.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. (18. (18.105) and (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 .*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. ± 1 . (18. ± 2 . Solution: The structure of the solution (18. (18.110) Aq+2+Aq2 9+2 A. .z\) + 2z(z . .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . . <4q+2 — z — z+ Aq+2iAq+2j (18.
108).2A g + A g _ 4 ] .34). (18. .2Aq + Aq4] 1.e.q + 6)A<. _ 4 . leaves uncompensated on the right hand side of Eq.<? +{q.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.101a) the contribution R .4 ( g .112) CHAPTER 18.116) into Eq.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 . .4]j. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.2\2 (z* .4)A„_ 4 + (q. •••].3 ) A .4Aq+2 + Aq+A Inserting (18. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.aAg+S 12+ 16 (18.(0) V2 ' 2c K + h4+Ao . we obtain 5 K = ( ^ f ) [ ( 9 .q + 2)Aq+2 (18.9=F4) = (g=Fl)(?=F3). Eqs. Since VqAq = 0.118) 2s/2c ~^j[(?. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.117) 4(q + l)2Aq+2 Here the first approximation of A.q2)Aq2 + (q.q)Aq + (q. QA?+4 _ 10 A + 6 . q) = 0.33) and (17. where the lowest coefficients have been determined above as (9.116) [Ag4 . (18. (q. (18. (18.l ) ( g . (q.4 4+* + .410 and from this = Hence with Eq. cf. (17.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq. Anharmonic Oscillator Potentials A 1 . (18.112).114b) . i.l ) 2 A . 1_4^±^+8^±±12^±^ +.119) It is now clear how the calculation of higher order contributions proceeds in our standard way.115) and (18. q + 4)Aq+4 + (q. (18.+6 + • • • ] .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.115) (z*z%) 2 i2 (z 2 . A(°~) = Aq.4A ? _2 + 6Aq . In particular the dominant approximation of A is obtained by setting (q.
g + 2)(g + 2. g .92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.4 ) ( 0 ) ( g .0)=0. 9 + 2) Aq+2Az+ (9.0) = y^(q. Terms jj.2 ) (0) (18. (i) _ / V2c\ [ ( g . (18.124) . where Rq is the sum of terms left uncompensated by AW.94)(g4.e.101a) the contribution Rq .123) ^[yA(q.120) The sum A = A^ + A^1) then represents a solution to that order provided the sum of terms in Aq in Rq and Rq is set equal to zero.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9. i.g) 8z+ (g. 9 + 4) ^+48z+ (18. z)\ = 2^A^ h2 '>z)± VA(Q.0)=0.(0). y'_(q. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A. g + 4)(g + 4.g ( 1 7 g 2 + 19).0) = ^ . in the domains away from the minima. (g. A' (0) = q/z\) y+(q.g2)(g2.e.yA(z) derived above are valid around z — 0.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g . g .105). (18. h2. (18.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.94) 8z+ (9. *)] (18. i. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. g) 8z 4 .g) 4z+ {q q)+ 2r (9.122) The solutions yA(z). h2.28. thus yielding the next approximation of A as given in Eq. for i = 0.h2. of course.g) Az.z)±yA(q.{^ h2.h2. h2 2y^ Age? h4 (18.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.h2.h2. y'+(q.h2.
2A y(w±).125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. 2(9+) .yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). Since w±(—z) = —h±(z + z±).412 CHAPTER 18. 18.^ % [i(?l)]!24(«. yB(z).yc(z) with complex variables and Cq(w) given by Eq.91). We see already from Eqs. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. (18.93) and setting w± = h±(z — z±).126b) These are solutions again around a minimum and with yB(z). we > .99) for z — z±. Moreover. (18.23) with appropriate change of parameters to those of the present case. Bq[w±{ ^ . we have w±(—z±) = —2h±z±.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2).Evaluating the exponential factor contained in VA{Z) of Eq.We draw attention to two additional points.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).95a) and (18.125b) It is clear that correspondingly we have solutions yc(z). the equation is — with differential operator T>q as defined by Eq.1 ) D (18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum.yB(z) is obtained around a minimum of the potential. (18. (18. This means.95b) that the solution there is of parabolic cylinder type. Inserting (18. Anharmonic Oscillator Potentials Our second pair of solutions. (18. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2). + 0(hl2). (18. and another VB(Z) ± 25^ch3wl + c2wi .126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.3. but w±(z±) = 0.
We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. . (18. 24(9D[i(gi)]! hj (18. (18. (18.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.3 The Double Well Potential (cf.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).147)).172) obtained later. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. .172)) we add here comments on this approximation.« + ) 2 (9+1) e (18.97b) ~ exp . Allowing z to approach z+ in the solution IJA(Z).^/w + ± 1 . Later we calculate the coordinates of turning points ZQ. 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. and comparing with Eq. 2 An+z+e 4w+ (18. Thus for h very large these turning points are very close to the minimum at z+. (18.ZI and find that these are given by z+ + 0(l/h) for finite q (cf. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i .• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . Here z± = ±h?/2c are the positions of the minima of the potential. .frW. .97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.146).. .i*l(«^ 4'"+^+ 4"'+v ( « .• • " . (18. ein+z+e in+(z 22+(<?+i) .18. a= ^ . Eqs.127) Recalling that around arg w± ~ 0.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq.z±)dz .129) .164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18. (18.
we naturally expect the wave function there to be similar to that of the harmonic oscillator.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. P.414 and CHAPTER 18. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. The solutions in terms of parabolic cylinder functions have a wide range of validity. even above the turning points. The next aspect to be considered is that of boundary conditions. Various investigations^ therefore struggle to overcome this to a good approximation. Furry [102] . K.z+)]fr+V ' (18.\h\)^+l\\{q l + l)]\ 1+ 0 J_ .(18. but it is clear that none of the above solutions can be used at the top of the central barrier. Since it is more probable to find a particle in the region of a minimum than elsewhere.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. basically.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . The pair of solutions of type B is defined around argz = 0. since. Thus 1 E . We achieve the same goal here by demanding our basic perturbation solutions to be interconvertible on the basis of the parameter symmetries of the original equation. J.3. We have to impose boundary conditions at the minima and at the origin. and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima. S. D. K.7r and the solutions of type C around argz = ±7r/2. Bhattacharya [23]. R. and this means at both minima. S. Rau [22]. de Deus [66] and W. H. The involvement of these WKB solutions leads to problems. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. 18. g . they assume large quantum numbers.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. Bhattacharya and A.
At dtz — ±oo > we require the wave functions to vanish so that they are square integrable.136) . 18. ^(*±)#0 (18.5. and y+(z±) = 0.18.Zii yB{z±) a ± a yciz±) (18.5 Behaviour of fundamental wave functions.134) y'+{z±) ± 0. (18. y'_(z±) = 0.132) Fig. We have y±(z±) = ^VA{z)±yA{z)]z^. 18. an even wave function can also pass through zero at these points.5. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. as indicated there. (18. as indicated in Fig. y(z±)=0. The odd wave function then exhibits a correspondingly opposite behaviour. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.133) y+(z±)^0. 18.135) Hence the conditions (18.132). y{z±) £ 0. However. as indicated in Fig.
138) Using formulae derived in Example 18.2 we can rewrite the second of (18. Eqs.5. 1 1 . 7 . .„*{!<. (1.. .. also (18. _ 8)} s . (18. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.136) the dominant approximations we obtain (cf. +1)} = 4 m. Thus sin {^ +1) } = ^^>^30fe^i. (18.416 CHAPTER 18.54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula. (18.136) as  M .142) (<Z9o)J(l)^°+1)(l)1/2 . (18.«*{=(. .141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i.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 .9.
» ^ B ^ ' . We deduce .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. we must also demand this behaviour here along with a nonvanishing Wronskian.3. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. We do this with the help of WKB solutions. (18.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin. Since the type A solutions are not valid at the minima.3 The Double Well Potential Thus altogether we obtain 417 ^ • . y+(0)^0. 18.z\. We emphasize: We needed the solutions of type A for the definition of even and odd solutions.6 Turning points z$.5 (after determination of the turning points) we rederive this relation using the WKB solutions from above the turning points matched (linearly) to their counterparts below the turning points and then evaluated at the minimum.18. 18. y'+(0) = 0. y'_(0)^0.1 " <813 1 4)  In Example 18. Hence we have to impose at z — 0 the conditions j/_(0) = 0.
We also note that the height of the potential at the turning points is h8 qh V(z) 20. (18. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer.^1 25c2 + V2 2 Thus again we see that for large values of h2 the turning points are very close to the minima of the potential for nonasymptotically large values of q.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.144).147) Since the minima are at z± — h2 /2c with /i 2 very large.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. + £\h%U(z)=Q. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. Anharmonic Oscillator Potentials from Eq.e. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .148) ("f * "The superscript (I.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. ZQ) means "to the left of 20" . In the domain 0 < z < ZQ.e. i. As a consequence. one finds that . h8 25c2 (18.145) \qh\ + \zW Using z+ = h2 /2c. to the left of ZQ where V > E.\qh\ + \h%U{z) (18.418 CHAPTER 18.
Hence we have for ywKB the exponential factor exp ("f * Z+ \*l*e\*(\h% 1 1/2 \qh\ 9/4 + \h\U{z) e\{zz+?h\_ (18152a) .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .149) In order to be able to extend the even and odd solutions to z = 0.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.S2 ) dz rzo 1/2 'Ihl + ^Uiz) I 1/2 / Jz dz (z~z+)  w + 1/2 1 = ±lh+ 1/2 ZQZ+ Z — Z+ 2 i^$) 1/2 ~4 >'^.^ e therefore have to consider the exponential factors occurring in (18.* 4+ .^ + ) 2 + igln2(zz+).18. (18.i (2q ln (18. . )+^*+)2lr} rnD h\z'2)}z Q )\ = . we have to match yA(z). 20 dz \qh\ + \h%U{z) .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ . (18.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.^ j O e .148) and (18.
We see therefore.ZQ) . t h a t the Stirling approximation is amazingly good even for small values of the argument). 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.e. This is the aspect investigated by Furry [102].153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (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. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.125a). of course. However. the results necessarily require adjustment or normalization there in the q—dependence. 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. l/2 ~h\U{z \(zz+?h\ (18.420 CHAPTER 18. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1.125a) and (18. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. using Stirling's formula (and not the inversion relation) — we have —(1.126a). In a corresponding manner — i.154) . Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18.
[\{q — 3)]! are really correct replacements of [\q}\ only for q large.159) 7 i(«i) r(9^3) ^2~ j(9l) . we obtain in leading order (i.e.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. .18. linearly matched) W K B solutions.143) T^^*)) 27T (18.99).123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4. 27T1/2 9/4 1VA{Z). this result is somewhat imprecise. and is shown to reproduce correctly the result (18. (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .153).144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18. .157) is used in Example 18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.103) and (18. (18.155) (2z.K?+1). 7 of t h e matching relations y^NKB\z) i.e. as the results also seem to support.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. \hl.5 for the calculation of the tunneling deviation q — go by using the usual (i. However.104). Returning to the even and odd solutions (18.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. (18.e.157) Since the factorials [^(q — 1)]!.154) with the typeA solutions (18. The relation (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18. Comparing for z —• z+ the W K B solutions (18. (18.158) Applying the boundary conditions (18.
D. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.. . We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^2 —pz I V2 dz ^?^ h+ Z 4? + ^f ^2<«2 (18.. 60 and 361. 213.z2)(b2  z2)..3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18.05.161) we can rewrite the integral as r I2{z) = = dz^{a2 .k'/2.148). formulae 220. p. Byrd and M. (18.149) near z = 0.19.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.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.422 CHAPTER 18. 3G2 12V2c ^See P.164) / 2 (0) z ca [(1 + k2)E(k) .iV2 ! i . p. (18. Friedman [40].163b) The integral hi?) .162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. . G2 K [i 4.163a) (18. F. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. n + uY 2c „ ^6 8V2c2 (18.
[4] misses an n—dependent power of 2 in the result. . n.. ^ ^ . ± .) = 2 \.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. . n = 0.3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .± ' 3G 2 ' T ( ^ (18.1)]' Gfo. which implies small G2.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . .e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. . 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 . /2n + l 2 ± — .r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . 1..3 )l ! .2.( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n . The nontrivial expansions are derived in Example 5.1.1 " e " O T (18 169)  Y^V / 4 tfo.165) in agreement with Ref.. (G\ . [167]. where the result is shown to be (with q ~ q0 = 2 n + l ..18.( ft 7.
(18.3.106).175) ~E0(q0.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.3 concerning the exponentials. (18. .143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec.172) into Eq... Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18.424 CHAPTER 18. h2) the split expressions 29o+1/i2(—W2 h6 (18.174) . (i*™) We obtain the energy E(q. we can therefore impose the boundary conditions at z = 0 by making in Eq.^ W ^ / 4 t o ..157).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. 2 . . 18.»W.1. (18. (18.6 Eigenvalues and level splitting We now insert the replacement (18.. we see that the relation is really an identity (the preexponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton). Inserting the expressions for h+ and z+ in terms of h. and obtain ( * .h2) and hence the splitting of asymptotically degenerate energy levels.173).143) with qo — 2n + l. (18. . (18.n = 0. with the help of Eq.3. (18. 18. .i ) ] ! e ~ s A ' . we obtain for E(q0.h2) + (qq0)^=.
. 1 u2 c 2 (3 9 2 + l) ^ « b .q2 7. Combining Eqs.106).e.1.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.fr2) is given by Eq.178) "In S.^ 2^ h* . the following expression for large q ~ 2n + 1.157).174) and (18.7 =( e N n+1 . i n U n / ! Thus AE(q0.e. for finite h2. i. 2h4' i. The result (18.V2^\ \n+ i . .176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB..143) with (18. K. T \ *n ! 1 ' (18.e. which evidently supplies some correction terms. (18. the level splitting. 1 7 425 2 .h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. (18. the pure WKB result of Banerjee and Bhatnagar [14] (i.n = 0. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 ..2 = 2 . i.2. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .h2)E+(qQ.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.18.3 The Double Well Potential where £0(<7o.. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.h2) = E_(q0. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i.e. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.e.
The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. H.2 / 7. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. the Furry factor corrected WKB result follows automatically. Bhattacharya [23]. however. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.6 \9o/2 e~h&l^\ with (m 0 = l) (18. (66). . h2)\mo=i = . Then the splitting A ^ W B of Eq. as is also explained by Bhattacharya [23]. 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.426 CHAPTER 18. Furry [102]. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». (18. that if this symmetry is taken into account from the very beginning. (18. K. and A there with c 2 / 2 here." Of course.181) "W. as we do here. we identity the parameter k there with ft. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2.3)).178).6) to yield an improvement of WKB results for small values of n. A>0.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).. This.176) together with Eq. ** Comparing the present work with that of S. (18.4/4 here. we would have obtained the result with E. (18.^ 2 + ^h2 If in addition the potential is written in the form A / . h = 1. Eq. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. 2/z4 and 2c2 respectively (see Eq.h2 <> — h2. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. h4 and c2 replaced by 2E. and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18.180) h8 E0(qo. We see therefore.2X2 V(z) = \z2tL) .
Gildener and A. D.h>) * 2V2^^^y/2e^3/3A. (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. ZinnJustin and U.18. Mansour and H.182) implies AlE(l. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.c2 = A/2. however. these are presumably not of much interest in physics (for reasons explained in Chapter 20). H.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.3 The Double Well Potential 427 a form frequently used in field theoretic applications. Patrasciociu [110]. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. D.3. m 2 <• p 2 / 2 . MiillerKirsten [186]. Friedberg and T.181) therefore implies the correspondence rj1 <> A/2.34) and (E. Liang and H. §J. first paper. The comparison with Eq. (18. W. Banerjee and S. P. (18.§ 18.^ *See E. (2. M. ^To help the comparison note that in J. Equation (18. J. Bhatnagar [14]. Eqs.^ Thus for the case of the ground state Eq. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation.11). Jentschura [293].—Q. D. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. so that by comparison with Eq. and g2 is given as g2 = r]2/m3.15). See R. Jentschura [293]. Lee [98]. 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. formula (4.W. J. M.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. ZinnJustin and U.88) h4 = 2/x2. K. . J. We considered above only the symmetric twominima potential.* A similar correspondence is also found in the case of the cosine potential.
Ashbaugh and Harrell [12] and Harrell [130]. 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. in both symmetric and asymmetric form. as he discusses. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle.[131]. P. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. The double well potential. Santi and N. has also been the subject of numerous numerical studies. mostly in connection with instantons.g. 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 his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. . can be found in an article by Coleman [54]. [132] and Loeffel. [19] — that the expansions considered above are asymptotic. N. Martin. Mahapatra. Simon and Wightman [178]. B. A reasonable. 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]. as some relevant references with their view we cite papers of Harrell and Simon [129]. 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. in principle its Schrodinger equation is an equation akin to equations like the Mathieu or modified Mathieu equations which lie outside the range of hypergeometric types. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. we cite papers of the Uppsala group of Froman and Froman [100]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. Perturbation theoretical aspects are "See e. though not exclusively numerical. B. for instance. There is no reason to view the anharmonic oscillator differently. Very illuminating discussions of double wells and periodic potentials.428 CHAPTER 18. As references in this direction. 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]. In fact. like those for anharmonic oscillators. Pradhan [182].
We have 4 ~k' 2u(l . which is assumed to be small. Thus k' and hence k' = v ^ ( l .186) (18. Example 18. 1+ u : Gy/2q. We obtain the expansions from Ref.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. 2 a. fc ~ 1 . T h e r e f o r e it is convenient t o deal w i t h this here.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 .191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' .185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.^ + • (18.183) S o l u t i o n : We have k2=1^.2G^2q + 4G2q .5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.192) 'F 12 16 . Cooper and Strottman [221].18.f + • • •) 2u\ .3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14].184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .u)1'2 = V2u{ 1 . Biswas. q ~ 2n + 1. k! (18. Saxena. Datta. Gutschick.\G\ + • and 1k2 =2Gv/2q4.187) We now reexpress various quantities in terms of u.G2q + . Bender. (18. Srivastava and Varma [24] and Bhatacharya and Rau [23].. 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./4 E{k) = 1 + • l n In (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. (18. X +2 + ' ^ (18. [122] as 13 1.
« ) ^ { 2 4 .— ) .u){4 + 3u(l .u) + 2u} + i u 2 [ 2 + (1 .6w2 + 13u} + 12u 3 (« .u)(u .13)} + ~u3(u .u)} .a = (l«)(jl 2u/ 2 ) .196) Now consider the last line here without the factor 2.430 and CHAPTER 18.3 y .«) + 2} + V [ .195) we obtain E(k) 1 + ln uK(k) .2{«(1 .194) ln '^J + 2 In i/2u )+H l + 2u(l .u 3 ( u . (18.1)(1 . 8 + 6 u 2 .13)}.u)} u{u(l .3 « .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .u) 2 {8 . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .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} + i u ( u .{l u ( l u) / W 2 + 2} . (18.u) + 2}] 1 H 1 o u(u . 8 + 9u 2 .1)(1 .u)} + — u ( u .1)(1 .^ L ) H[(i _ U ){4 + 3u(l .197) .u ( l .13)} + 12u 3 (u .193) 2u(l .13)} + ±u3(u .3) . 1 3 « 2 = .1)(1 .{ M 2 ( 1 .3).« 2 + 3u] In [ .u 13 \ 1 ln(± k> fc'2 +  (18.3u(6u . (18. 8 . ) J 4 (18.4. .3) .3 u ( l .u){24 .u) 2 {24 .u ) (^){« + 5^ 1 )} + 5.3) = = ~ .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .3 .u ( l .195) With (18.6 in the denominator and in the last step pick out the lowest order terms in u (i.3u(6« .3u(6u . up to and including u2): u(u .e. 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3u(6u .3) 1 + In ^ .194) and (18.u){24 .3 u .u){24 .18u 2 + 39M} + 12u 3 (u ..
zo) HWKB normalized _ ( h* y^eM. (18. / to + 3GH~ \ 8 .200) Remembering that u = Gy/2q. (18. (18. i. 2 .u H I .2 /o 2 431 (18.197) we now obtain 4 1 • E(fc) . where w±(z) — h±(z — z±). From (18.2{w(l u) + 2}] = u(3u2 . [(1 . ' 8G 2 q = 4 2 q ln f 26/2 \ o ..3).156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.18.e.3) . .128) and (18.it)} . + iu^Zu2)^~u+ — +0(u3) l .148) (for particle mass 1/2 and in dominant order) is given by H.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs. (18. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .195) with (18..] in (18. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . we have.^HfHGH Example 18.2 (18. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .4u .3 The Double Well Potential Now consider the bracket [.uK(fe) ~ 1 + In I — = Mu2(3u.4u .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .196). Inserting here from the above expansions the contributions up to and including those of order u2. l2 = ^ ( l + uf/2[E(k)uK(k)}.«){4 + 3u(l .198) 9 «. V^LJ 4 2 16 ^ .e.173) t o be evaluated.199) We now return t o the expression on the left of Eq. Eqs.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).2 l_ln(1 JL\ V {2.196) and (18.. i.128) and (18.
g. oo . normalized = We see that the normalization constant is independent of a quantum number.f hi\ VWKB. We start from Eq. (18.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. Then ((r.146) and (18. 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 .** oo / dwD\.158). normalized  V4exp[ ~ f*° dz\/\E~ \EV(z)\V* V z () With / i 4 / 4 = m 2 and mass 1 (instead of 1/2) the result is ' m2 \ 1'4 e x p [ .e. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. (18.e. (26. i.147). (18.203) 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+. W. pp.35)).zo). are given by Eqs. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.hA/2 = h^/4 this implies . Eq.204) . and obtained as Eq.3)]!23<«D [5(91)]' >A. Magnus and F. (18. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e. i./ . ° \Edzy/\EV(z) V(z)V4 VWKB. Example 18.432 CHAPTER 18. Oberhettinger [181]. by using the periodic WKB solutions below the turning points.l)]![i(0 . 1 7(l. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. Solution: The turning points at zo and z\ on either side of the minimum at 24.143).Aw) = s/2i 2("!) V (41) and [(9 . 82.*o)/ ^WWKBOO (18. 93.
i. 1/2 .208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. 6.g.r + ^=co S [r.205) We now apply the boundary conditions (18.2..e.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18. z{) we expect tt (18. \h\U{z) 1/2 ..210) Choosing the point z to be z+. JV = 1. Then at any point z € (zo.207) (18. We could.212) tt See e. Sec.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.2.133) 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 (18.3 The Double Well Potential We also note that d dz 433 y±{z) >2+ 1 F = 7 sln h%U(z) 1/4 f X —2 cos 7 2 2^ + 1 \ 1 / 2 71 .e.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .^r + + ...211) 7 z l . .e. 7T\ + ^ 4 7r\ dz .132) and (18..3.6.iK h\U{z) + / 4 (18.( . this implies sin cos sin cos Thus e.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.. Messiah [195]. (18.g.18.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. from ZQ. approach the minimum also from the right. from z\. i. A. of course. i.J^Wj +(18.
(18._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ .148) and (18. The formula thus agrees with that in the literature. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.3.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 (18.209) and (18.146)).205).210) assume a form as in our perturbation theory.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . where u> is the oscillator frequency of the wells. i. P. K. (18. (18. and cot< (q + 1) '!}» for q = 50 = 1. 1 1 . the level splitting is given by AE(q0. + l ) j } ^ 1_ 27' (18. We now see that Eqs. M..Zi / r. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. The prefactor 2\/2h2/n is. .434 CHAPTER 18.213) The integral on the left can be approximated by 1. 5..2 .214) {«+"i} 0 for q = go = 3. 7r"y in agreement with Eq.215) we can insert for 7 / 7 the expression given by Eqs.149). .90 — =F— for q0 = 1. D.215) we can expand the left hand sides about these points and thus obtain .exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. (18. Lifshitz [156]. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). 7 .215). Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±).e. (18. (18. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18.215) which results from the factor of 2 in front of the sine in the WKB formula (18. and L. W "Cf. K. Bhattacharya [23]. Bhatnagar [14]. Landau and E.27 9 . S.216) Inserting this into Eq. Banerjee and S. . (18.143). ' /^i . they become c o t { ( . In Eq.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. . in fact (with our use of qo).5. 9. 2IO/K.
M. MiillerKirsten [9]. Case [45]. 435 . 'For a review from this perspective with numerous references see H. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. W.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. 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. M. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. W. also in P. H.1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state.Chapter 19 Singular Potentials 19. 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. Feshbach [198]. J. Morse and H. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. Nonetheless it was thought that a certain formal analogy could be seen if the field theory is supplied with Euclidean spacetime concepts at the expense of sacrificing the interpretation of the interaction in terms of particle exchange. n > 2. Giittinger and H.* However. are not as troublesome as one might expect. However. f K . Aly. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. in an early investigation Case^ showed that potentials of the form r~~n.
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. In the concluding section we refer to additional related literature and applications. D. Gubser and A. S. J. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. MiillerKirsten and Jianzu Zhang [226]. Singular Potentials theories in mind. . 19. J. H.436 CHAPTER 19. K. Liang and Yunbo Zhang [189]. Recently. In fact. J.2. Tamaryan. Callan and J. Hashimoto [124]. It is therefore natural that we study this case here in detail. N. § 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). our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. MiillerKirsten. some also permitting further research. for instance. 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.e. W.2) § C .Q. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. S.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. R. D referring to Dirichlet boundary conditions) in 10dimensional string theory. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. Singular potentials arise in various other contexts. that with the hyperbolic cosine replacing the trigonometric cosine). W. (19. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation.2 19. Manvelyan. Maldacena [41]. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. H. G. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. Park.
H.3a) In the case of the attractive potential (as in the string theory context referred to above. Eq. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. (19. This is rarely possible and therefore this case deserves particular attention.mo = 1/2 and h = c = 1. J. 19. It is clear that we draw analogies to our earlier considerations of the periodic potential.19. ig h \j g (19. W.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17.4) first for small values of h2 and then for large values of h2.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. h = ei^y/k^. The following substitutions are advantageous: y = r1/2cf>.28) below).g. (19. (19. J. 8.Q.7 that we employed also in previous chapters. We observe that with the replacements p2 z^iz. r = 7 e z . R. 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. . MiillerKirsten.( I + 0. In the following we study the solutions of Eq. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 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. Liang and Yunbo Zhang [189]. This has advantages compared with higher dimensional cases. using the method of Dingle and Miiller of Sec. Manvelyan.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . (19. h2 = ikg. \[\ in the correspondence is nonintegral.3b) 7 = % j. It may be noticed that since I is an integer.3 below) g2 is negative and hence h? is real for E > 0. and as we have in mind in Sec.
6) terms amounting to 1 „ 2 d2Zu A V (19. The solutions Zv(w) are written Jv(w).1.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions.e.5) so that Eq. dwA w dw { wA ) (lg.Hv (w). VahediFaridi [10]. (19. (19.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. (19. Proceeding along the lines of the perturbation method of Sec. Eq.Nv(w). Aly. 8.2 CHAPTER 19. (19. where these are the Bessel function of the first kind.2. . The zerothorder approximation = Zv[w) (19. where £>„ : = —_ + i _ + J l . (19.I . Singular Potentials Small h2 solutions in t e r m s of B e s s e l functions We now develop a perturbative procedure" for solving the associated Mathieu equation in the domain around h2 = 0. —Zu — (Z„_i + Zu+i). MullerKirsten and N.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. w 2 .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. — Zv+i). (19. J. H. First we make the additional substitution w = 2/icosh z.9) leaves uncompensated on the right hand side of Eq.i l ..4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .438 19.2 A / i 2 . i.11) We follow here largely H. H.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. and H„ (w). W.7. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.
15) (19.13).2)G„_ 2 + (y. v)Gv + (y. I / ± 2 ) = 1. = §. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). v + 2)Gv+2].17) Thus a term fiGu+a on the right hand side of Eq.10) in terms of functions G„ defined by Gv+a = —nZ v+a .2 ) * i ^ 2 + (!/.. where We observe that DUZV = Q. when a or 2v + a = 0. . (19. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . v + 2)*(v + 2.2 ) * ( z / . . z .10) as a linear combination of various Zv._2 + {v. v2)*{y2. (19. wz The expression (19. where the starred coefficients are defined by (19.12) (19. v .14) (19.2)*Zi. (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.i/) = 2A.18) Now </>(°) = Zv left uncompensated Rv . ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.u + 2)*Z„ +2 ]. v .10) is now particularly simple: i?i°) = h2[(v.20) 2)*ZV_2 + {y.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.2.13) therefore lead to the first order contribution jfU = h2[{v. However.16) ( I / .6).i/4)*Z 1 / _ 4 +(u.19. (i/.21) +(i/.u (19. it is more convenient to use these relations in order to rewrite Eq. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.!/+ 2)*4°J 2 ]. v + 2)*Z„ +2 (19. of course. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.v + 2)*{u + 2.u + 4fZu+A}. The terms (19.13) (19. and this means in (19.
26) iy — 2.440 CHAPTER 19. Evaluating t h e first few terms. . .4)(»/2 . (19.v2)*{v2. we obtain 9 h4 = a 2(al) 3 2 (13a . subject to the boundary conditions P2i(2j) = 0 for  j  > i. (19.i/±2)*. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9.v + 2y{v + 2.v) + hA[{v. in Eq.l)(z/ 2 .25)/i 8 32(al)3(a4) . Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. i/ ± 2)*.I/±2)*(I/±2.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . . v + 2)*{v + 2. (I/. i/ ± 2)* (i/ ± 2.v) + {v. Adding these terms and setting the coefficient of Gv equal t o zero.455a + 1291a . (19.2 + 2 9 ) ^ 64(^ 2 . Zv + Y^h2i i=l J2 p2i{2j)Zv+2j.I/±4)*.2)(v + 2j .v + 2)*(i/ + 2. i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. Ri.22) where p2(±2) p4(±4) = = (i/. i/ + 2j). ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . j=i. we obtain 0 = h2{v.j^0 (19. . P2*o(0) = 0. p4(±2) = (i/. p 0 (0) = 1.9) 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.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . v — 2)*(v — 2. etc.2.. we obtain t h e expansion . v) +(i/. .1169)/i 12 . 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v .17)).4 + 58. .24) P2i2(2j + 2){v + 2j + 2. and p 0 ( 2 j ^ 0) = 0. i/)] + • • • .v)] (19. . v + 2j) (19.g. v ( 4 5 a .
32) (19. ±2. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. Thus to O(0) in h we have (jr ' = 4>v — cosh vz. (19.i .cosh 2z)4>v = = 2h2A(/>„ . v)(f>v + {v.7) into Eq.29) sinh vz or e±uz. Thus v is real for small values of  h21.34) . v . Substituting (19. and is therefore real for both cases g2 > 0 and g2 < 0.2)<t>„2 + (v.An(u + n). (19. The solutions 4> of the modified Mathieu equation are now completely determined. 2 . we can rewrite the latter as j ~ . We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.31) It follows that Dv+2n4>v+2n = 0. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. so that J2 (19. (19. £>„ = — .v2(j> = 2/i 2 (A .30) £>„&.^ + e±(v~2)z.33) = = cosh(^ + 2)z + cosh(^ — 2)z. (19.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.h2[(j)u+2 + 4>v_2] h2 [{u. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . 19. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. = 0. (19. .4).v + 2)<t>v+2].cosh 2z)4>.2.19. Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv .
> These solutions correspond in the periodic case respectively to the solutions and me^.h). The use of the symbols (v. CHAPTER 19. 8. M. F. > (f> with </v = sinh vz — Seu(z. In order to avoid confusion arising from the use of different equations. 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.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. Schafke [193].h). 19.J/±2) = . v ± 2) etc. W. • (19. (19. 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.7 — we refer to Meixner and Schafke [193].442 where (I/.38) (f> with (\>v = exp(vz) — Meu(z. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.h).I/±2)*.h) = <(>„ +Y.I/) = 2 A . Meixner and F. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2.h2i *=1 J2 j=i.37) where p M (±2) = (I/.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions.J7t0 p2i(2j)<f>u+2j. . The solutions in terms of cylindrical functions are written *J. Singular Potentials (I/. 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. Arscott [11].2. (19. etc.4 Notation and properties of solutions We now introduce standard notation as in established literature.1 . we prefer to adhere to one equation with different solutions.
h). . this normalized solution possesses contributions exp(i/z) in higher order terms. ' (z. as Mev{z.h). we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. {2h cosh z) • MJ?\z.g.h2) oo = Y.e. W.44) (19.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z).h). 443 (19. as we remarked earlier. h). h). W. (19. 4>v = Hi. i. whereas the — as yet — unnormalized solution (19.42). Mi.40) Cev{z.42) The solutions Meu(z.40) does not. (19. which. see this reference p. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. 'See e. (19. h). p.19.2 The Potential 1/r4 — Case of Small h? similarly: <fi 4> <f> (p with with with with <j>v = J v (2/icosh z) . (19.MP(z. (19. (19. As emphasized in Chapter 8.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19.(2hcosh z). Magnus and F.h).h)±Sey{z. h) are therefore proportional to each other as a comparison of Eqs.41) i=l We see immediately that Meu(z + niri.43) and hence similarly M^\z + inir.e. Me„(z. For the relation of Hankel functions used below. 16.h) = exp(vz) + Y. — oo cv2r{h2)e^+2r>. h) = exp(inwn)MP(z.45) T Below we frequently write the normalized solution as in J. Also from the expansion of Jv{z) in rising powers of z. i.h). 4>v = Hv (2h cosh 2) * M^\z.44) implies. Oberhettinger [181]. 17. Meixner and F. Schafke [193].h) and oo i M^\z. h) = exp{vniri)Me„(z. <f)v = N„(2hcosh z) MJ?\z. h) = av{h)M^ (z.
h) . i. Meixner and Schafke [193]. 2M™ = eiuvMi3) + e~ivlrM^\ .g.8.h) Using further properties like (19.h) on the other hand can be shown (cf.444 CHAPTER 19.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. p. H™(z) = e^HfXz).48) The series expansions of the associated Mathieu functions M„ (z. r=—oo (19.h) can be shown (cf.e.exp(TM"r)MW(.49) Inserting this expansion into the modified Mathieu equation (19. ^ . The functions Me±v{z. the first relation is obtained as follows: 2M<£1 = M^l+M^l. For later essential requirements (i. h). As mentioned earlier. MJ>>(0.46) H^z) = e^H^(z). (19.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. 130) to converge uniformly for all finite complex values of z. h) = M™(z. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.h) . Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. s (19.e.h2): oo Mev(z. oiv{h) = (19. we have the following for a change from v to — v. These points are important in our derivation of the S'matrix below. .e. Meixner and Schafke [193]. i. p.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0.43).h2) = ] T cv2r(h2)e{y+2r)z.50) E. Singular Potentials where clearly MeJQ. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. Mi 3 .
Meixner and F. \) /. ./ i 2 C ^ . W. \ .51) we have 2r (19.^ + 2r + 2 ) 2 ] .19. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. 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 ] . + 2 r ) 2 ] .(/^.T22r[\ A . (39).^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . eg = c~v = 1. 122. 2r (19>51) 2r For ease of reading we give the steps in rewriting this. Schafke [193].^ C 2r+2 (19. p.^ p 2 2 ^ or c 2r+2 2r2 1 (19. . 117.* For 'Actually.54) 2 C2r 4 ^ h. Eq. Thus now c 2r .53) /i2[A(^ + 2 r ) ] . p. h~2[\ . see J.(v + 2r)2} / z . (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.2 [ A .52) in agreement with Meixner and Schafke [193].
. W. MiillerKirsten and N.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(.27) and truncate the continued fraction after the second step._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. Liang and Yunbo Zhang [189].+4)!ll_. + 0(/. Spector [257]. .55 ' Example 19.l ) 19. and also R. Then di 2 1 t" + 2(^1) + • • • .Q .1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. Here we insert for A the expansion (19.446 CHAPTER 19.8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. VahediFaridi [10] (this paper contains several misprints which we correct here).( " + 2)2] ~ h2 h..C + l) 3 (i/ .+0^"> < 19 .^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). W. t For this purpose *We do this in the manner of the original derivation of R.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. 128(i/+ l) 2 (i/ + 2 ) ( ! / . MiillerKirsten.1 for the evaluation of a typical case): hb + 0(hw).. J . (19. M. but follow our earlier reference H. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. H. J. Aly. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ . Manvelyan. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .5 D e r i v a t i o n of t h e S .55). + i).2. (19. H.[v^. J. H. Solution: We put r = 1 in Eq.
2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.19. 22. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). and considering the propagation of this wavefront. In the case of the repulsive singular potential here.57) ^ 1 . Then y reg = = r rV2M®(z. . We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0.3a) and (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 . we have W r = ip + ut — 7r/4' .j7L0 P2i(2j)exp i[v + 2j + 2 2 . p. 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. Oberhettinger [181].5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo.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. I / + 2 2 l\TT +E ^ i\ E j=i. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. Magnus and F. (19.h) 1/2 w) (19. and then the continuation of this to infinity. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. (19.59) which tends to zero with r..56) where by Eqs. W .
Now from Eqs.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i.(2j)<gH • (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. 4%r g >0 4r^r ' rg See W.62) Also with Eq. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z. '2"4 .j¥=o Ki(2j)exp(Tmj) (19.V + § 9 2 > 4hh* . Meixner and Schafke [193].448 CHAPTER 19.h) = exp(Mr^)il4 4 ) (z.A1"—2 z '+"2 A~~ n or rA . 17. 2h and the square of this expression implies 2 2 ~ ' A. '(z. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. We require therefore > > the continuation of Mi. Oberhettinger [181].#./l ~ 2 ~ 2z + " 4 > 0. . Magnus and F. p.60) Using the above asymptotic expressions for the Hankel functions. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.j^Q p2. p./i). The series M^ (z. h) through the entire range of !&{z).h) can be shown (cf. (19. This means that the origin of coordinates acts • > as a sink. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. Singular Potentials so that when t — oo : r — 0. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.3b) ln • 7T Q"*4' r fg ° Vk (19. AV .A14A.61) In fact. (19. 178) to be convergent for  cosh z\ > 1.
i Im z z = In r/r0+i7t/4 . (19. . This interchange in the > solution My interchanges y reg with y+.1 The domains of solutions. 449 This condition.38) or (19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 .49).3a). i.rl)(r2 . is satisfied only for r < r_ < r__ or r_ < r + < r. (r — r+)(r + r+)(r — r_)(r + r_) > 0. which has to be bridged by using another set of solutions.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig.ijt/4 Rez .e. Meixner and Schafke [193].r2_) > 0 with r£ = rg(2 ± >/3). p. Originally we chose the sign of m / 4 as in Eqs. A suitable set is the pair of fundamental solutions Me±v defined by (19. and in order to maintain this symmetry. the domain r_ < r < r+ — as illustrated in Fig. (19.62). 130). r 7r for 0 < r < ro. 19.63) Thus there is a gap between the two regions of validity — i. Since the real part of z changes sign at r = ro. 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.e. These solutions converge uniformly for all finite complex values of z (cf. Thus we choose . (19. we have to assign different signs to the imaginary part of z in the two distant regions.19. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo.3b) and (19.
Singular Potentials Then starting from the region r ~ 0.l/2Me_i/(T.450 CHAPTER 19.^)] dr z=in/4 dr (19.e. 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. p.z=—in/4. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. i. h) (as one can check or look up in Meixner and Schafke [193]. 131). Hence we match the right hand sides of Eqs.66) a = (3 .[ r l/2 M. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. i. Thus.B^0. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.(2)]z=w/4 + 0Me. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. to r^2[AM^ + BM^\ A.v{z)\ . as just explained. (3 = a .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_. at this junction we require also the imaginary part of z to change sign. (19. and we determine these coefficients from this equality and the additional equality obtained from the derivatives. At r = ro the real part of z vanishes and then switches from negative to positive. . h) = Me^u(—z. (19.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.v(z)\ —m/4) + ^[r^Me.
70) by Me'_u (the prime meaning derivative). (19. we use the notation W\f. multiplying the first of Eqs.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}. = aMev(z) + 0Mev(z).65). (19.69) Thus if we replace in the derivative relations of Eqs.68) the derivatives with respect to r by this relation.64) the relations M&Hz) jz[M^\z)] and from Eqs.68) From Eqs. (19. (19. (19.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. the nonderivative parts (from the first term on the right of Eq.71) We now determine the coefficients a. dr BM^(r)]. and the second of Eqs. and we are left with relations expressible only in terms of z. A and B.65) and (19.70) by MeU and subtracting .3a) and (19. (19.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri. (19. (19. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. Thus we obtain from Eqs.g]:=f(z)*&g(z)WZ) dz dz Thus. ^ ] = ^ + ^ ^ .69)) cancel out in view of the nonderivative relations. (3. (19.2 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.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. (19.19.3b) we infer that (z = l n r + const. with the replacements of Eq. (19. (19.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . (19. Then. for instance.
u. we obtain the first of the following two equations. (19.46).e.3] = [1.. (19. Me.75) With these relations and Eq. Meixner and Schafke [193].71) we obtain in a similar way A B = = W\MP\ Me. M^4)] + W[MJ>3). TV W[Mev. av{h) = Me. we obtain immediately W[Z.77) av I 2ism VK \ ctv av ^Thus. p.A] = W[Mi3\z). 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.58) and considering large values of \z\. (19. the leading terms of the respective cylindrical functions.j] given in Meixner and Schafke [193] (p. 7T (19. and Wronskians W[M^\MIJ)] = [i. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.4] = .Me^} (19.] W[Mev.57) and (19. or obtainable by substituting e. (19.M^] J™aV. 7T W[Mev.48) (or cf. (19. = IT <*„. we use the following circuit relation which can be derived like Eq.* « * r r z H .g.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. (19.4] = . W[MJ?\ M^] W[Mev.76) . .M^]= IT e~iv*av. h).73) We now use Eq.^ ^ a „ Q _ .M^} W[Me„.u}W[Me.Mev]' P l ' From Eqs. h)/M^\o.Mevy W[M?\Mev] W[Mev. 7T [1.452 CHAPTER 19. (19.. (19.^ 47 9? MJ>4)] MeV] [3. Singular Potentials the second resulting equation from the first resulting equation. W[M^\Mlf)]W[Meu. i. B = J 2 L _ e .74) Moreover.\ \ ( av 2iwKav ^ v . i.M^] W[Mev. and the second is obtained similarly: W[M?\Mev\ W[Mev.(0..46)) W[Mev. 170/171).e.Mev] = = = av. using Eqs.47) we obtain (av defined by Eq. .
82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#). (19.46) together with the relation Mev(z.(ite™)" 1 { 8 J ' f(k. Z ) e . (19. 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 .83) The Smatrix element is therefore given by ifc*"* .19. in Chapter 16). (19.85a) .79) where the second expression follows from Eq. (19. B=tfD'"". Then f' .80) 2^fc (1981) where \7r/ i t sin VK /(*._ [ / ( f c . 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq./) = (l)1/2R2^2i^)ei^)U^.78) we obtain Ay reg « _ . eii/7r/2 (19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM).f c .l) ' We note already here that with the substitution R = ei7r7. f c r .h) + BM^\z.h) = Me_u(z.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z. as we recall from earlier considerations (e.g.h). (19./ ( . Z ) e ^ ] .(l)le^}Pi(cos 9). (19.
2 The repulsive potential and the various waves. 19. (19. R 2i sin TXV.58)) 1/2 2r e~ 2hir cosh z.58)) in the limit r — 0.2.Smatrix in the form s i n 7T7 Si = CHAPTER 19. (19. cf.56) to (19.454 we can rewrite the . We can rewrite Eq. 19. Fig. we can write this (multiplied > by 2ismvTT.56). (19. (19. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). 9?. R — — = 2i sin ivy.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.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. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. and with the left hand side following from Eqs. we can define respectively as amplitudes of the incident wave. Thus in terms of the (3) variable z.86) . V(r)=g2/r Fig. h) (see Eqs. and recalling that yreg is proportional to a function Mc> {z.
unity minus reflection probability = transmission probability.88) where the factor Me„(0. H. This implies basically the evaluation of the expression R of Eq.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles.h) The function Mo (0. (^)l4lu{h2)Jir(hez)J±u+i+r(hez). h) serves the use of the same coefficients as in the other expansions.e™*/R\2 ' ' . MullerKirsten.e. (p. i.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind.e.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. (19.19. The latter is what happens in the S'wave case of the attractive potential. (19. (19. By inserting in Eq. if R becomes a pure phase factor and eil/7r a real exponential. (19. S.Q. (105) to (108). J. therefore we cite it from Meixner and Schafke [193]. W. A derivation is beyond the scope of our objectives here. (19. in which v is complex. Jv+2r." \R~Ji\2 1  tO _  2 i s i n VK\2 [  . 180).84) of the ^matrix.e.h) is (cf. and then exploit it for the evaluation of the quantity R.89) .h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). / i ) = Y. Eqs. M^l)(0.28)) and R = ey real. Gubser and A. 178). Manvelyan.2COS2VTT = (R. i.m a t r i x Our next task is to evaluate the expression (19. however not here. (19.±)2 + (2sinj/7r) 2 . " S e e S. l=—oo 1 1 lite*"* . unitarity is preserved. J..( i J e ^ ) " 1 ! 2 = R2 + ^ . The expansion is: 4 r ( / i 2 ) M S ( z .88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence.79). i. Eq. 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. Liang and Yunbo Zhang [189]. this expansion is given as 1 oo MJ> Hz.2. Hashimoto [124] and R.6 E v a l u a t i o n of t h e S . i. but in the 10dimensional string theory context.e.In Meixner and Schafke [193] (p.** 19.
one obtains *.91) Inserting here the coefficients given in (19.90) and choose r = 0 (one can choose e. by allocating different values to r. 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.94) .49). (19. e. setting r = 0 in Eq.g. The formula (19. h) in many different ways.28)).456 so that in particular for z = 0 oo CHAPTER 19. (19. ^ (19. Eq. (19. (19.g.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.49) from which we obtain for z = 0: Me„(0. (19. r = 2 as a check)..h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••.e. i.6 2 ) m 6 (^_i)3(z. (19. /n ^ .3 9 ^ .90). u»^.55). Singular Potentials 4r (h )M$(0. /=—oo (19. Next we evaluate M^1}(0. r = 0 and 2.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. As a matter of introduction we recall the definition of the coefficients with Eq./>) = Co> 2 ) £ 7 " = — OO j ± > . Here in the case of the repulsive potential. h2 is complex but v2 is real (cf./i) with the help of Eq. we obtain M&(0. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ .h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).2_4)(l/2_9)iv2y) + <AA.
/I) h u .111) / / i 5 4 3 1+ 4i/ 2 + +' (z..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. one obtains M£!(O.IV 2 2 2(v =F 3i/ . _2_(h\2 .] 2 \v i + {v2 .32i/ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .93) and the coefficient expansions (19.2^ 2 ± 59^ .7) F " (i/±l)2(^^l)(i/24)V2. From expansion (19.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' . 4(z/ T H ^ 3 .1) 2u(4u + 15i/ .l) 5 (19.9) V2 1 fh ±v 1+ 2 + • (19..96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy. + # "+" ( I / .4)(i/ . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .1 ) 2 ( I / + 1 ) ( V 2 .19.55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).96) With this explicit expansion we can evaluate the S'matrix.12i/2 + 64u . .2 _ ^ 4 ( ^ 2 _ 4)2 (19.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .. We observe that this approximation involves v and h4 and hence is real.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.
103) Then SiSt (1+5) i . where n is an integer and a and (3 are real and of 0(h4) We have with Eq. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).R*2 ^ * #2 \ 11 tf2 R* (19.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. (19. This is the case of real .98) (cf.28)). (19.eiun/R2)(eiu*n ~ (19. Then cos 2TT/? = 1 . (19.102) (19.104) We now consider the case of a — 0 implying that g of Eq.458 19.7 CHAPTER 19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case.101) is > zero. + ( i + s w (E 4 .101) where / is complex and g is real.28) that v is real. (19.9 / . (e iun e~iv*^/R*2)' (19. and hence allow for the general case of complex parameters v (the Floquet exponent in the case of the periodic Mathieu equation) although here in the case of small \h2\ we know from Eq.2. Eq. (19. and sin 2TT/3 = ft/. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3.84): Si ST 1 (11/R2)(11/R*2) .^ i 4 2+^ # 1 / •. In this case R — R* and so 1/i?2 ~ Ofji4).99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19.
3.19. . "This is the result effectively contained in R.6C . 1 29/ ^ (1 .3b)) h2 is real for E = k2 > 0. S.105) sin7r/3 (19.l ) 2 2 \) +0{h«) (19.102) we obtain Sis. Liang and Yunbo Zhang [189] (there Eq. f W . Oberhettinger [181]. (19. Hashimoto [124]. W.^ ) ~ 2 2 459 1+ 9 / • & ( ! .97) this can be written A 4TT2 [v\{v . 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 . p. Gubser and A. MullerKirsten. 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.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. Magnus and F. We can also evaluate now the amplitudes (19.107b) = Ah2[l + 0(h4)}. Manvelyan. i.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 .577.107c) Here h2 is actually Vh4. J. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). J. With t h e help of Eq. At = 2i[l + 0(h2)}.61n2 . (19.Q.28). ( I l l ) ) in agreement with a result in S. = 1 . H.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].61n/i} + ~9~ 0(h8) (19. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0.e. (19. (19. Here in t h e case of the attractive potential (as remarked after Eq.1)\} 2 h \2 iv 1 + Au {y . one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .SiSf » 4 sin7rz/ R (19. Hence with Eq.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.106) and evidently violates the unitarity of S.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.
J. 19. W.109) ^We follow here mainly D. 2 2 a = [l + 2 (19.3 19.1.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z .a }tp = 0. 2. Singular Potentials 19. W. completely different solutions. h) for large h and upwards I = 0.108) For the large values of h2 that we wish to consider here.460 CHAPTER 19. K.h) (19. MiillerKirsten and Jianzu Zhang [226]. Park. (19. .3. J. H. we again make use of the replacement (17. N. i.H. H. Miiller and N.4 Fig.3.e. S. VahediFaridi [8] and G. We are again concerned with the associated Mathieu equation given by Eq. 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.$ q(l.h) 10. Aly. Wannier [278]. H. This is therefore a highly interesting case which found its complete solution only recently.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5matrix obtained above by using the asymptotic expansions for large values of h2.3 The function q(l.26) and hence set a1 = 2hl + 2hq + A{q. where the part of interest here is based on H. Tamaryan.
(19. i.h*y_(l. however. which therefore cancels out or can be taken to be 1. and a is its normalization constant.111) we have W[y+. .21b). The replacement (19.e. With the boundary conditions y+(0) = l. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.l + 2y+(l.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case.h*y (19. We begin.yJ\ — 1 and cos TTV .3. to which we have to return in this subsection. (17.113) 'We are considering nonintegral values of v. y± oc A{z)e2h sm * ± A(z)e2n sm z . and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.3.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq.19. with an essential mathematical step.21b). y'_(0) = 1. (19.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. J4(0)=0. 19.y} where VF[y+. which is even about z = 0. The behaviour of q as a function of h for some values of / is shown in Fig. 19. (19. Therefore the second term on the right is nonzero. i.§ a W[y+. (17.5) that this is given by the relation cos iris = ^ M .110) a where y+(7r) is the solution evaluated at z = TV.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. See also Eq. y(0) = 0.63). (17. We also found the boundary condition (17.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. we defined the Floquet exponent v and observed with Eq.e.
114) at z = 7r/2. See R. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. J.118a) "We emphasize again: For integral values of v the normalization constants are different. i. Dingle and H. 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 . (17.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). .111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. and hence with Eqs.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq.46) together with the expansion of the Hermite function from Tables of Special Functions" ) B[w = 0] 7T 4 [lfol)]![l(g+l)]! 2 /i 5 82q2 . Miiller [73].117) a a Referring back to Eq. W. J.32). K [ (z)}2hsinz (19.462 CHAPTER 19. Dingle and H. (19. This paper contains the normalization constants of large/i 2 periodic Mathieu functions.43) we see that z = 7r/2 implies W(TT/2) = 0.112) from evaluation of Eqs. 1 1 Or see R. (19. which is \/2 at z = 0. We now obtain the expressions needed in Eq. (19. B.39 2 14 /i 2 + (19.e. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . Hence we require (obtained from Eq. Miiller [73]. B. (17. in the second expression 2h is obtained in addition from differentiation of the exponential factor).114) a a Our first step is therefore the determination of the normalization constants No. (19. (17.A(z)e2h sin 2 sin z sin z (19. W.
119a) This formula — derived and rediscovered in Park. (19. Tamaryan.! ) ] ! [ .118b) The factors a.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.117) are known from the matching of solutions in Eqs.3).62b) (see also the explicit calculation in Example 17. p.87 2 14 /i 2 [ i ( g . a in Eq. 210. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.2g + 3) 27/i ( 3 g 2 .19.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 .h .! ) ] ! [ .1Q11QM cos irv + 1 ~ •==— e . (19. (19.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19.i ( g +3)]! 1 24/i {2hl>2).112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q . .3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 . MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. (17. (19. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].62a) and (17.
z) = A(q. z) are TP(q. Then straightforward integration yields > inq 4 „. —h. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . h.3. each with a specific asymptotic behaviour.124) Since this function differs from a solution ip by a factor k(q.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. z) can then be written as . h.h)iP(q. V— 2in (19.h) =?==.2. We define these solutions in terms of the function Ke(q.2 we insert the expression (19. —z).3. of course. „ 1 A „ —A cosh z——H . for h — oo).h.h.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. we can obtain the linearly independent one either as ip(—q. dA 1.z) or as tp(q.z) = Aq(z)exp[±2hiswh z] ^~°° exp ±z/te ( ')eT^/4> V cosh z e x = Aq(z) exp[±2hi sinh z] ** ~~°° ^ h e ^ V cosh z e ^q/\ (19.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q.h).120) to (19. —h. z) = ip(—q.z).120) The resulting equation for the function A(q. 17. .464 CHAPTER 19.4.h.h. —z). the expression (19.3 Following the procedure of Sec. h. h. h(q.122) that given one solution ip(q.e. . In the following we require solutions H e ^ ( z ) . z). h.h.109) into the equation (19.3. (19. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19. h.108) and set ip{q.. i = 1.z). it is still a solution but not with the symmetry property ip(q. (19. .? s m h z) Vcosh z (19. z) = ip(q.z)exp[±2hismh z\.z) i/>(q.h. Vcosh z \ 1 .123) We again make the important observation by looking at Eqs. h. With the solutions as they stand.122) Correspondingly the various solutions ip(q. tp(q.109) remaining unchanged.
/h). We observed that these satisfy the relation (19.126) For large values of the argument 2/icosh 2. i. as we saw earlier. [ \ .e.19. In order to be able to obtain the Smatrix.45).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 .^ s i n h 7?)]g./i 2 ) M (19. (0. (19. we h a v e 77 = 0—> 6 = 0.h2). in performing this cycle of replacements the function picks up a factor. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . . Me„(z.e.h2) = OvMPfrh2). 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]. : 7r/2.e. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr.127) sin z T /'1 M TT\ — <>{ 4 . One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . 2j<. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . av{h2) = ^(0J ^ Mi i . (44) of W a n n i e r . (19. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. 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 . * ** = > rojKe( ". the Bessel functions contained in the expansion of the associated Mathieu function M{.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]. achieved with the help of the Floquet solutions Me±v(z. 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 . we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. i. 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.3 The Potential 1/r4 — Case of Large h2 465 Instead. . (z.fi2) can be reexpressed in terms of Hankel functions. i. >• * F $ 0 * e". This is.<5±1) (19 125)  (the expression on the far right in leading order for h2 large).
HeW(z.vr .g.q. z): Re(2\z.h) He (z. 3te»0. (19.q.h) (3) = = = = Ke(q. He (z.h.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. (19./i) Be^(z.466 CHAPTER 19.q.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs. (2) (19.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.h). (19.z).h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.3. 3te»0.130b) r_0 rll2exp\g/r (ig) ' 19. exp[—ikr — in/4\ kr Re{2)(z.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.eW(z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T . Singular Potentials Note that the sine part is nonleading! Retaining only the dominant term of this expansion we have (with z — 2h cosh z) > Me±l/(z. h.g. (19. (19.q.h) Be^(z.4 T h e c o n n e c t i o n formulas We now return to Eq.h).130b).q.130a) Re^(z.q.q. In this way we obtain in the domain . KeW(z.130a).q./i).h) = e(z) exp ihez .
and thus can be matched in this region.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. / i ) + sin7r(7^)He(4)(2. (19.h2) Me„(z.g.i ^ / 2 H e ( l ) (z> ?) (19. Eqs. Changing the sign of z. t With Eq. z) are proportional there. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z.h2) a.li 2 ) Oil. q.h). h. (19.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19.g.134) in the region where the function He^(z.132) into Eqs.19.130b))./i). (19.135) sin irv 'This means. q.(z.130a). h) — sin 7T7 He^(z. (19. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .132) These relations are now valid over the entire range of z.h) sin irv = sin7r(7 + ^)He (3) (2. we obtain in the domain tfcz <C 0 the relations: Me./i 2 ) = Me_j. . h) = Ke(q. we see that the solutions (compare with Eqs. (19. z) and He (3) (z.h) is asymptotically small (cf.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z./i) — sin 7T7 Re^(z. and He(*> exponentially decreasing. q. — sin 777 H e ( 3 ) ( 2 .« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e .g.133) In a similar way one obtains the connection formulas — sin nv HeW(z.q. these are exponentially increasing there. Substituting the Eqs.129)) He (2) (z. h. (19. q. (19. g . h) = sin 7r(7 + v) He (1) (z. h) = Ke(q./i).3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z.q. (19.131) where the second relation was obtained by changing the sign of v in the first.134) Considering now the first of relations (19.(z.q.
( . H.135) is the equation corresponding to Eq. The quantity 7 is now to be determined from Eq. (19.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. The latter is defined by the following larger behaviour of the solution chosen at r = 0. and therefore remarks after his Eq. (19.m a t r i x From Eq. which in our case is the solution He^4^.e'i7ri 1.e. (19. Thus here the S'matrix is defined by (using Eq. cos TXV + sin TXV.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr .Sin TXV. i. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . Wannier did not have the relation (19. (60) in the work of G.468 CHAPTER 19.135).l ) (l„—ikr e From this we can deduce that Si (19.139) ''Equation (19.137) We see that this expression agrees with that of Eq. (19.119a) and (19. of course.119b) as „4/i 1 + COS TXV ~ rrl ( g . q.119b). where 8i is the phase shift. (76): "It is not likely at this stage that an analytic relation will ever be found connecting (what we call) v and 7 to (what we call) (I + 1/2) 2 and h2".5 D e r i v a t i o n of t h e ^ . exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7.(19. is only the dominant term for large h2} 19.133)) — sin TXV He^4' (z.85b) obtained earlier.1 .119a) or (19.133) we can now deduce the 5matrix Si — e2t51.135). (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . .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. Wannier [278]. Squaring Eq. (19. Singular Potentials The factor on the left. sin TXJ.3.
19. i.§ 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.141) COS TTU irvj — i s i n WR s i n h TTUJ. since cos TVU is large. (19. imaginary part uj (so that 1 on the left hand side of Eq. T h i s is r e a l for UR a n integer. for the attractive potential.109). (19. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. .e.139) could actually be neglected) implying  cos m>\ = cosh nvi. h) of the Zth partial wave.138) we obtain. A{l. 19. we can approximate the equation for q ~ h (i.137) together with (19. as a comparison with Eq. (19.4 The absorptivity of the S wave (attractive case). (19.28) shows. the real part of v must be an integer.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.e. 9/2 (91) with the approximation q ~ h obtainable from Eq. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1.e. This is different from the behaviour for small values of \h2\. 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 (19.8h COS From Eq.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. i.140) From this we obtain the absorptivity A(l.
and. . F. D.** 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. N.4 . see also M. I r a n [61] and M. Apart from the papers already referred to." as well as other aspects. B. E. Pope and T. S. Lii and J. small h2) case have been considered by some other authors. A. Esposito [88]. 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.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently.' Rudimentary aspects of a singular potential with a general power n have also been considered previously.470 CHAPTER 19. In essence. Furlan [20].139) 2Tr(16h)q /32\ f e Thus with Eqs. Partial or other aspects of the weak coupling (i. in further contexts beyond those already mentioned. N.142) we can see the behaviour of the S'wave absorptivity as a function of h.e. W Some of these sources can be helpful in further investigations. L. Masson [191]. H. (19. Shabad [248]. The diminishing fluctuations of A(l. Rosendorf [225]. YuanBen [285]. H. Y. Tiktopoulos and S. Handelsman. Lew [128]. A. Jones [78]. Eden [46].P. VazquezPoritz [61]. (19. A. H.107b) and (19. 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. D. Bertocchi. Cvetic.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . C. in fact. Cvetic. S . Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . Paliov and S. J. N. Limic [177]. such as phase shifts. Fubini and G. one may expect corresponding results for other singular power potentials. 11 J. This is sketched schematically in Fig. 19.4 (there are tiny fluctuations in the rapid approach to 1). and G. "G. Treiman [271]. 19. Challifour and R. **R.k) in the approach to unity are too small to become evident here. **A. Pao and J. Lii. Dombey and R. Apparently this is one of the very rare cases which permits such complete treatment.
W. Wu [18]. Bender and T.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. M. Dingle and H.* In fact. 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. 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. *R.g. J. B. 471 . the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. Muller [74]. B. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. e. [19]. Thus in the following we shall not only obtain the large order behaviour of the coefficients of the eigenvalue expansion but also that of the coefficients of the wave functions as well as the connection between these. We shall also see explicitly.* However. f R . This is indeed a remarkable connection. We concentrate in this chapter on the large order behaviour of asymptotic *C. the asymptotic solutions in different domains all have the same coefficients.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. T. Dingle [71]. also convergent expansions are known and a lot of literature exists on the equation.
Presumably this has never been done so far.2 and below. 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. Since the method is also applicable to convergent expansions.2 The function ur = r\/pr vs. Fig. 20. r for p = 4.1 The function ur = pr jr\ vs.2.472 CHAPTER 20. 10 12 14 16 18 20 Fig. This approach leads .6.8. 19.2.1 and cases in Sec. 8. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.7.6. 8. 20. as evident from Example 17. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. 5. r for p = 4.
. 20. The behaviour of such terms is illustrated in Figs.1 Introductory Remarks 473 automatically to asymptotic series. The traditional method of using an asymptotic expansion is to truncate the series at the least term. 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.1 and 20. The behaviour of the rth term (r large) of the convergent expansion is generally of the form of a power divided by a factorial (as. For an authoritative discussion of these aspects we refer to Dingle [70].3) n=0 Thus for any given value of g . (20. and the integral E(g2) is called the Borel sum.20. It can be seen from these. the function E(g2) is given only approximately by Yln=o En92n.J2 2n h E (20.2. equivalently for g = l/h. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). as we discussed earlier in Chapter 8.e.e. even if each En is known. and the expansion is said to be Borel summable. for instance.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. i. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. The formal perturbation expansion of a quantity E(g ) .2) or. A considerable amount of work has recently been devoted to the question whether it is possible to reconstruct a function exactly from its asymptotic expansion. i. T V lim h2N 2 n EQ?) . this term represents the size of the error. in the power expansion of the exponential function). In some cases the answer is affirmative.
5) then with the integral representation of the gamma function F(z) = (z — 1)!. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . (20. (20. n=0 (20. (and E0 = 0). In any case.474 CHAPTER 20.e. El . .1)? Asymptotic expansions originate from integrating a series (i.6)). The function F = 1/(1 + z) is an analytic function which can be continued beyond this circle to the entire positive zaxis as required by Eq. .4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. (20. n=0 (20. 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 . Thus if CO F(z) = aiS(z) + J2 anzn. In this ideal case the integral of Eq. In fact. . Q. .4). many asymptotic series do not even alternate in sign (as in the case of the series in Eq. . Large Order Behaviour of Perturbation Expansions Eq.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. 1 we obtain oo F(z)£(*)" = — .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. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). .n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. In general this is not so easy. (20.
Popov and V.. This is the case if the real potential has a hump of some kind. $s E can be calculated by the WKB procedure. § where 7(ff) = To exp  . (20. and E represents an eigenenergy. . or can we determine Ar independently? Fig.9) The question is therefore: How can we determine 9 E(h). ? When ?s E ^ 0.3 The potential V(r) gr». 20./ \p\dr\. (20. (20.. 20. e. S..N = l. Weinberg [228]. if the potential is V(r) = +grN..11) V. The potential (20..1 Introductory Remarks 475 except for possible subtractions.2.10) r as shown in Fig.. if the potential or a boundary condition is complex such that the problem is nonselfadjoint.e..10) is considered in detail in this work. N = l. M.20. we have a quantum mechanical problem where probability can leak away. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).. i..g. 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. for < ^ 0.3 and g < 0.2.
2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. of course. . Thus in the following we apply these methods to our typical examples. Example 20. V. (14. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients. S.1.9) for Ar.^ The logarithmic factor in Er is a novel feature of this case. (11. the cosine potential being t h e most completely investigated will also be seen in the following. Popov and V. (20. 20. Solution: The solution can be found in a paper of Eletsky. as we expect on the basis of Eq. L. (20. Here n is the principal quantum number of the Coulomb problem (see Eq.113)). In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential.11) is. An interesting application is provided by Example 20. 8. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available.7 with its focus on the structure of coefficients of perturbation expansions was seen to permit even the formulation of the recurrence relation of its coefficients. as was (and still is) not the case for t h e equation 1 V . M.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.134)).76) (see also the comment on the comparison with the Kepler problem before Eq.73) in the W K B or semiclassical approximation. Popov and Weinberg. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. Equation (20. However. Eletsky. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. Large Order Behaviour of Perturbation Expansions where the frequency of particle bounces against the potential barrier is 70 ~ ^classical/27r and w c i assica i = 2ir/Tkepier oc 1/n 3 . the perturbation method of Sec.476 CHAPTER 20. (11. the BreitWigner formula (10.11). Weinberg [83].
(17.j) a n d M2i can be deduced. M2i = 2j2 J[Pi(q.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.58) and (17. j .34)) that the coefficients Pi(q.l ) ( g + 4 j .13) In the special case j = i the boundary conditions stated after Eq. (17. % = Aq. 3)? . i. (17. (20. or equivalently of the quantity A.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.13) represents effectively a partial difference equation.l(q.e.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.j) of functions ipq+4j.J) = (g + 4 j .M2i+1 for large i. 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. .55)) X = E(q.Bq. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.14) Thus these coefficients can be obtained once the coefficients Pi(q. B. J. The recurrence relation (20.3 ) P i _ 1 ( g .58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.60). Eqs. W. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. (17. obey the recurrence relation jPi(q. Dingle and H. Miiller [74].l ) +2[(q + Aj)2 + l + A]Pl.26) and (17.12) (20. Eqs.20. are given by i . h).j) are known. (20. (17. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.j + 1).h) = 2h2 + 2hq+j.Cq.2 A = Mx + ~M2 + 7^WM3 + •• • . (cf. Eqs.
17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. see also below) .18. Hence exact Borel summation is not possible.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i .478 CHAPTER 20.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20. .16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20. Using this relation and the duplication formula lw('\W±^.*. (20.16a) (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.•91 {[l(? + i)]![(? + 3)]!}2 (20. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .l (20. also written lCj.91 l)]![i(g3)]!} 2 { [ .
4. but with different boundary conditions.2 solutions obtained earlier. i.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/?. .3. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions.20. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary. . .3 Cosine Potential: Complex Eigenvalues 479 20. the difference q — q$. we emphasize a few points *M. this difference was found to be real.* However. n = 0 . we follow here a formulation in the context of our earlier considerations of the Mathieu equation. The potential we consider here then has the shape shown in Fig. l .4 The cosine potential shifted by TT/2. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. Previously. At the risk of repeating what we explained earlier.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. in the calculation of the level splitting as a result of tunneling. A means to achieve this was found by Stone and Reeve. 20. . We retain the previous boundary condition at one minimum. J. Stone and J. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. 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. \ 1 \ \ \ \ \ 0. f H . W. remains unchanged. 20. 2 . Reeve [262].e.3 20. . MiillerKirsten [200]. when this is g0 = 2n + l .5 \ \ / \A Fig. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2.
X + 2h2 = 2hq+—. which means replacing z by z±ir/2. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there.20a) As stated. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.480 CHAPTER 20. Thus one chooses not the solution with real argument at the minimum.4 (around z — 0) to the one on the right (around z ~ ir). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . we shift the argument by n/2. We can distort the potential there or alternatively impose a different behaviour there on the wave function. as we found these for this potential.4.2h2 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). Effectively we consider the tunneling from the left well in Fig. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . 20. i. Now we change this situation at the minimum on the right. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. 8 the equation can be approximated by d2ip(w) (20. but t h a t with complex argument there. a boundary condition which at the face looks somewhat abstract. and allows its imaginary part to tend to infinity. Setting w = 2h1/2z. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them.e.21) <u> + ( i « . (20.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. Previously this situation was the same at the neighbouring minimum.4. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. (1 w < 2a22) cylinder T h e solutions of this equation are the real and complex parabolic functions D ±(q~i)(±w)> In the following we find it convenient to select solutions of this type in the domains around the minima of the potential in Fig. 20.T K ^ a D_i_{q+1){±iw).Ah2z2 • • • }^{z) = 0. Here we choose t h e second way. 20. but achieves exactly what we want. .
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. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. .e.3 Cosine Potential: Complex Eigenvalues 481 As explained above. that at z = 7r. 20. . A (which we can loosely dub solutions of WKBtype).24) Considering t m O i n Eq.20.e.23) . however. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. The calculation for the general case is nontrivial and so will first only be written down and will then be verified by specialization to the case of the ground state. 17. Since the large/i2 solutions valid away from the minima were the solutions of types A. We consider here explicitly only the case of the ground state around z = 0. 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. 3 . we naturally constructed the even and odd solutions with these (cf. 2. A derivation is given subsequently. whereas here we start off with a solution even or odd about a minimum. we see that ips ~ cos \j2qhz for z ~ 0. . the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. by a different method. whereas here we construct these from oscillatortype solutions. such that this deviation becomes imaginary.4. The proper harmonic oscillator solution there would be that for infinitely high barriers to the left and to the right with a dependence on an integer n which enumerates the quantum states in the well from the even or symmetric ground state with n = 0 through alternately then odd and even states upwards with n = 1. (20. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. (20. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. Eq. i.> 0 for z > ±ioo. . Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. Thus. .63).22).
w) = e T Dv{w) 2TT n») AvViDv^iw). Magnus and F.24) are linearly independent around z = 0 for q ^ qo. (20. and then evaluation at u = 0. this can be handled in a few lines. From Tables of Special Functions'. This is the feature which gives rise to an exponentially increasing contribution in the neighbourhood of z — 0 and hence provides the dominant contribution in going away from the origin. (20.482 CHAPTER 20. . we know from the solutions of types B and C in Sec.27) requires the differentiation of this expression with respect to the index v. §See e. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. Thus the only nonvanishing contribution comes from differentiation of Y[—v). 92.27) differ by an angle IT. W.28) Extracting Du{—w) we have A.h(2h1/2 J £2=0 (20. p.g.3. p. However.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. 17.29) Equation (20. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. W.30) 'See e.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. Oberhettinger [181]. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. tp(v).we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{. Magnus and F.(. (20.•w) + 2ir T{u) <"+!)% Dvi(iw).26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. 2. Oberhettinger [181].g. The arguments of the functions appearing in Eq. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. (20. We see this as follows.25) qo) E2 0: (20.
we obtain 4>s D0(2h^z) D0(2h1/2. We now go to the WKBtype solutions.27) exponentially decreasing contributions. those of type A. This result is » not easily generalized to <jo > 1. Oberhettinger [181]. which are valid around z ~ 7r/2.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. For u — 0 one obtains the result (20. W.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. 5 7 7 .+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .//n and ij>{y) = C . 92. pp. 3 . Magnus and F.20.29). . (20. Reeve [262]. p.33) = IV2~TT.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.g.. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT.e. Oberhettinger [181].4.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. Magnus and F. and "Prom W. i/^0 483 T(v) It follows that d dv [T(u)\ (20. .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. . r(i/) We now return to Eq. Differentiation of this equation and subsequently setting v — 0 yields. Thus calculations for qo ^ 1 are not given by M. . we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. i. (20. dropping from the derivative part of Eq. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20.e. (20. "See e. Stone and J. including the derivative of DE2ufl(2h1/2z)).31).
cos z/2 ~ (20. (20.39) Now. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left. we obtain the required solutions from the earlier ones.36) and consider this at the other end of its domain of validity. near z = n the boundary condition (20. 2 c2e2hehz UJ2 . that in approaching z = ir. Eqs. We observe that in the matching domain (i.40) From Magnus and Oberhettinger [181] (p.29) and (17.sm I We now want to match this solution to that of Eq. Recalling that we have shifted the potential by 7r/2.36) cos  . x . by there replacing z by z + ir/2.34) and (20. Then V'WKB becomes (with cos z ~ — 1 + z' /2. i. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. Thus in the present case and with q ~ qo = 1 we obtain „2ft.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. y COS c\ J (20.38) in its limiting domain towards 2 = 0.37) Comparing Eqs. To this end we set z' = z — ix.e. (20. We now return to the WKBlike solution (20. (20.40).484 CHAPTER 20. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).e. (20.— 721 . (20. i..41) 2TT . (17.34).35) We construct with constants c\. 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. where we can then match these to the solution ips.e.cos z — 2/icos z ipA^ Tcos  and i n A —.36) V'WKB cie~2hehz' 9h  . 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . smf (20.
1 2 4 1 2 8h ^^2 2 (20. (20. .39) and (20. (20.2/.^.46) Repeating these calculations for the general case (i. ^'mliw <2a47) For qo = 1 we recover the result (20.44) = DQ{2h / z) ± (2h7r) / e. with c\ = exp(—2h) from above. i.38) can be written TAWKB (20.45) Comparing this result with tps of Eq.34) we obtain E2 = i%\ = ±ih22Qe~8h.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'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 .20. (20.1/V)_le^_ Z With this result Eq.43) we obtain. (20.e. J Hence Eq.e. The general result stated by Stone and Reeve [262] is therefore presumably guessed.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real. We derive the general result in the next subsection using our own method. (20. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . (20.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(.46).43) Comparing now Eqs.
5 The cosine potential from — TT/2 to n/2.486 20. MiillerKirsten and A. Considering the original equation. i. We have to impose two sets of boundary conditions.48) These conditions provide the required quantum number qo = 2n + 1.3. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .()#0. At the minimum to the right of the barrier shown in Fig. On the other side of the barrier at z = —TT/2 we consider different boundary conditions.Ah2 ( z + . the result is formula (334) there. Wiedemann [4].5. Achuthan. 20. i. J. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. . this equation may be approximated there as d2ip ~dz^ + E + 2h2 .2 CHAPTER 20. Fig. W. at z = TT/2.e. i.46) we consider a half period of the cosine potential as illustrated in Fig. 20.5. (20.e. Eq. •_(:)*<>. V(z) .z. (20. Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e 7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P. H.*.e.20a).) + 7T V> = o. •+(=)* V(f) 0.E curre it \ j. 20. and expanding cos 2z around the point z = — TT/2. (20.
' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.e.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. (20.z)=tpA(q. (20.z). (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. i.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. B and C.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.50a) V'AfeM) = where (the second expression is needed for later comparison) Aq(z) cos§(9i)(iz + lyr) + W) [ t a n (±z + W)]«/ 2 [sin(i* + ±TT) C O S ( ^ + ±vr)] V' ' sinfr+Vaz (20. Proceeding as in our previous cases and substituting the expression (20. three pairs of solutions of types named A.21) into the Mathieu equation we obtain. (20.h. as shown earlier.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish. For particles or probability to leak through the potential barrier from the trough at z = —ir/2 in the direction of negative values of z where we choose the potential to be zero.20. Recapitulating these from Chapter 17. For the continuation we use again the WKB formalism.51b) 2i(«1)[i(9l)]'.h. we demand that ^(*)L<. (20.r/2 ~ e i(2hq) V 2 .21)).
53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20.52b) The first solution is valid asymptotically around z = —vr/2.e. 5. the relation q . Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function.48) and proceeding as in Chapter 17. Here the upper sign refers to the even states and the lower to the odd states.=F2 2\1/2{28h2y/4e~Ah l + O (20. (20. Finally the third pair of solutions is Cq[w(z)}+0 Cq[w(z)}+0 Cq[w(z)} where Cq[w(z)} = 22(9+1) •((73) 1 h1/2 1 hW> (20. \vw). we obtain our first transcendental equation.488 CHAPTER 20. we obtain 1>±W = \ a a (20.57) where go = 1) 3.qo . We saw in Chapter 17 that some of these solutions can be matched in regions of common validity. ^ W H . (20.52a) = Cq[w(z)}. (20.56) Applying the boundary conditions (20. and the second around z = 7r/2.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}.^ i . We return to this condition after derivation of the . Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . i.55) Extending these solutions to the domain around the minimum at z = 7r/2. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2.
D.e.58) Ignoring the correction A / 8 . (20.2hq] / dz) We know that the solutions I/JA{Z). The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. p. I^A(Z) overlap parts of the domains of validity of these solutions. Thereafter we extend the latter across the turning point down to periodic WKB solutions. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. Jz+ ' T In principle the integral could be evaluated exactly — see P. .Ah2 cos2 z ~ 0.5 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.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum.49) and we can apply our boundary condition.67). We begin therefore with the matching of V'WKB ( Z ) to IPA{Z) and integ rait* V'WKB (Z) to I=f V'AW Hence it is necessary to consider the elliptic (20. Some nontrivial manipulations are again required to uncover the generation of important factorials. These then provide complex exponentials like (20. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.01.[ [Ah cos z . F. Our procedure now is to match these to exponential WKB solutions. Byrd and M. e x p ( .20.61) [Ah2 cos2 z2hq]1/2dz. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . formula 280. as we shall see with the result of Eq. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. we see that the turning points at z± are given by 2hq . Friedman [40]. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (20.Ah2 cos2 z ip = 0. and we obtain the proportionality by appropriate expansion.
sin2 z = l .62) as follows.64) . Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. 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 /•«"(.63) We handle the logarithmic terms in Eq.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.cos2 ( z+^) = (z+ . (20. we write cos2 z = l . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).490 CHAPTER 20. 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. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.» _ i y v .59) 2h COS Z+  ~ 0.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.) +0 z+ IT Inserting this into the integral I. (20. We have for the following expression contained in (20.
62).50b). .60) by (4h2 cos2 z)1!4). we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .+ qj V2 4 A.20. (20. . but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  .3 Cosine Potential: Complex Eigenvalues Inserting the relations (20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . 1/2 (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. on using in the second step Eqs. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20. (20.50a) and (20. (4^2)i/4 [ i( g _3)] ! V/ l (20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.64) into Eq.66) The WKB solution therefore becomes (approximating the denominator of (20.63) and (20.68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated. We ignore this here.
KW4^8 4 .50)).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. i. Thus with IPA(Z) and TPA(Z) taken from Eqs.2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .T 7 77 r )V2 /4(2 . Large Order Behaviour of Perturbation Expansions again using the Stirling formula. S1H (2/ KZ ) / (z f 2 (2^).1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .+) + B^) HAh^/\ir.2 [ ? ) Imposing the boundary condition (20. Eqs.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i).i ( 2 / i g ) 1 / 2 ( z + .l^l !T ^^* T2 7 r ( ^^ /. we obtain l(8/i)g/ 4 e.67) and (20. io(8/*)g/V^i6ft2)V4[±(g3)]! .?/ 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.™) .. (20.U ^K^.49). 1 2 1 ±2 .e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.z)}].49). where _ 1 + [±) *) + £ (20.69) l(8M9/4e2^ \2 [ i ( g . (14. we obtain T _ ( ± ) = 0.55) 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+. (14. We now return to the even and odd wave functions defined by Eq. (20. i 27r(8^/ 4 2/t /4 ' r(±) .492 CHAPTER 20.
74) This result agrees with that of Stone and Reeve [262]. Ex = 2h2 + 2hq0 + A 1 (<? .20.3. Eqs.h) + i l + O '4{[^ol)]!}2 E(q0. In both cases we wrote (cf. the result (20.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. We now have the two equations which have to be satisfied. h) of Eq. for q an integer both sides are imaginary. Eqs. (20. who derived it for qo = 1 and guessed the form for general values of qo. Combining these equations by replacing (28/i2)<?/4 in Eq. i.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. (20.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .h) + (qq0)( BE — 90 E{qQ.57) and (20. (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.3 ) I = }(»« !7rl/22I(.72) We observe that with the square root as on the right hand side both sides are complex. we obtain E(q. Equation (20. in fact.h) = = E(q0.e.h)+2h(qq0) (16h)i°+le'Sh E(qQ.21) expanded about q = qo.72) is our second transcendental equation.72).h)+i%Ego. (20.qo) El Ah' (20.75) . (20. 20.57) by the left hand side of Eq.25)) E = A = Ei + E2.73) Inserting this expression into A = E(q.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .47). (20.26). (20. (17.
Wiedemann [4] and Example 20. f E . i.9) the coefficients Ai of its perturbation expansion. its tunneling properties and the large order behaviour. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj.I g  ± l ( 2 „. cf.80) *P.3. H. J.W [£(901)]! < (20. . 25Eqo=8h[ 2\1/2 .494 CHAPTER 20. We had A= or _2ft2 + 2ft. A.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. Setting ng AEqo:=2iSfEqo. Bogomol'nyi and V. Fateyev [26]. (17. (20. Achuthan. The same formula can be derived in the case of the double well potential. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo.2 below.74b). the discontinuity across its cut from zero to infinity. W. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20.e.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. 7 8 ) i=l V . B.+  . Eq. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions.) {I6h)™/2e4h 1 + . MiillerKirsten and A. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. we can recalculate with the help of Eq. (20. A = . 20. by others.
(20.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . Using again n = (go .13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq.1). (20.j + 1). there is nothing special about this case. B. (20. We also observe that these terms do not alternate in sign. *R. (20.l ) ( < ? + 4 j .3 ) P i _ 1 ( c ? .. Dingle [74].83) One now sets ftH = ( 2 % P * ( 9 . MullerKirsten [200].81) We observe that this result agrees with the result of Eq. For convenience we write Eq.19) for the large values of i under consideration here. and the method can be applied in many other cases. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. The integral is recognized to be of the type of the integral representation of the gamma function.13) out again: jPiiQj) = (g + 4 j . j .j) satisfy the recurrence relation (20. But as Dingle remarks.74). J).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. so that an exact Borel summation is not possible.14). a = J + \<1> (2084) . f H . 20. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. the integration is seen to give (20. (20.20.l ) +2[(q + 4j)2 + l + A}Pi_1(q. J.j) + {q + 4j + l)(q + 4j + 3)Pi1(q.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq. W.
z)WaA/2(z) Setting V.e. we regain Eq.^(Z).90) See M.h) 2+ 8h a p(a.l. we obtain (2a . Pi(a) i. formula 13. nx pi_i(a + l). (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.l/2l Wa.e.89) Replacing here K by a and \x by 1/2. (20. a.h).a R .g.e. Whittaker functions WKjfl. d2WKh dz2 + 1 4 K z zM2n WKtll = 0. the recurrence relation (20.(a) oc . Now. (20.86) one can convince oneself that Eq. (20.31. 4j(4j + 2g + 4) + / .K + ^ ] L + K^)WK.85) Considering now the generating function of the perturbation coefficients. 1  1 / 2 (4 (20.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.1) + 2 (a — 1) a .h) = ^2pi(a) with p.87) Taking terms of 0(h~l+1) of this equation.4 ) ^ . i.85). 0 4j(4j + 2g) pi_i(a . (20.27h pi(a) a = 4j(4j + 2 g .h) +p(a + l. g < 4 j .4. (20. . (q + 4j) 2 ~ 4j(4j + 2g). A.83) can be written j . p(a.85) implies p(a . e. i.91) + Wa+lil/2(z) = (1 . Abramowitz and I. Stegun [1]. z2 (20.1) + 2 P i _i(a) + Pi_i(a + 1). (20. which are solutions of the differential equation known as Whittaker's equation. Taking dominant terms.ll{z)= L .496 CHAPTER 20.1/2(z) (a1)! (20.
e. we have to remove a factor (recalling a = j + (7/2 of Eq.g(h) are functions of h which in view of Eq.j — —j. . Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).94) Hence to insure that p(a. > (20. i. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. p.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.88) is known. (20. Magnus and F . Oberhettinger [181]. such that p(a.93) (a1)! ' where f(h). h) is expressed in descending powers of z = 8/i.\)V .. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . a (20.4 Cosine Potential: Another Method of Calculation we obtain Vai.20.84)) {8h)q/2( Ah for j < 0.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. Dingle discovered the solutions^ p(a.i/2(z) + Va+lA/2(z) = 497 2+ Va.(« . (20.i/2(z).(« 1) 2 ]. (20. In the present case this means the replacements q — —q. (20. > > ^See W.87). another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged.h) = \~\Pi(&) with pi oc h l.92) Comparing Eq. h —> —h.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. ^Observe that if one solution of Whittaker's equation (20.(« .86) have to be chosen such that p(a.h) ~/(fe) ' (a1)! .92) with Eq. h) is expressed in descending powers of h. 89. v . g(h)(20.
(20.498 CHAPTER 20. i). j — —j. J.83) and is given by 1 1 . . Muller [74].). n = 0. (20. (20. _ i s. g — —q. Comparing > > > this result now with Eq. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.95b) Then for j > 0 we obtain from Eq....95a) Inserting (20.+ i f or 3 < 0 (20. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2.96) and a similar expression for j < 0 with /i — —/i. 2.l ) ] ! 11 R. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .1/2(8^) ( .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) here we have a series in descending powers of h. h\ oc (8h)q/2e~4h ( _ .j _ g / 2  1 / 2 ( . This expression can be obtained from the recurrence relation (20. Dingle and H. B.1. Correspondingly we set for j > 0 / 1 A pla = j + q. W. we obtain the result of Dingle: (20.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1.8 / t ) for j > 0. ^ .
e. (20. of Eq.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.157(4).3)]!2 2 '+^.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . Dingle and H.l ] ! {[. l{Q.j) (i + lg)[2i + i ( g .97c) We assume that in a leading approximation we can cancel the factors (i + \q).97d) Then the coefficient M<n of the eigenvalue expansion.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20. (20.20. B. S. A special case of the formula can be found in I.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. note the misprint there: (r — s) ought to be (r — s)\. W.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.l ] ! (20.14).e.0) = 1 as desired (although not really enforcible for large values of i). .+i?_l]!}2 (20.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq. Gradshteyn and I.100) **R. Ryzhik [122]. E* J'=I i+M 2 [2i + 9 . p.{j + \<l). 5.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. is M2l = 2j2j[Pi(<l. (20. Miiller [74].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. formula (30). Thus ° 9 (i + \q) 2«(i+Igl)! p . i. formula 0. J. M.
(18. i\ M i large 1 .86).l l ! with the large i behaviour in agreement with the result (20. With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator.105 ) the latter because we see from the eigenvalue expansion (18. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.e.1. .Am i .e.86) and set (cf. [^ + i ( g . n = 0. (18.4)) £ = ^ = ^+ w and y= ^ > ( 20 . [19].e.l.5.34) that £ is an expansion in ascending powers of h 6 /c 2 .17).. i. the result of Bender and Wu [18]... replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.102) Thus the result is M m 1 2«Fi + g .i ) ] ! [ ^ +  ( g .81) for q — qo = 2 n + l . i..3 ) ] ! _ 1 Hence M or. Eq.e.. i. since \{q — l) + \{q — 3) = \q.5 20. We refer back to Eq.e.101) This formula allows us to cancel all factorials [i.]\.o i • . i. (18.1 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.. Eq.' 1 i • (20.2. 20. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. i.500 CHAPTER 20.
the discontinuity across its cut. 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. and then the relation (20.l ) r . .e.109) or.110) This is the result in agreement with the large order formula (4.(20. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . in terms of K where go = 2ftT + 1.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.77) can be verified.108) Proceeding in this way. and hence inserting here the imaginary part from Eq.4) of Bender and Wu [19].e. (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.86). i.107) we obtain (i)r r 71" JO yirZ£{y')dy'. i. 20.2 The double well It is clear that one can now proceed and apply a nonselfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue.106) With £= E r=0 Mr (20.5. in fact as one with Borel summability in view of the factor ( . The result establishes the series unequivocally as an asymptotic series. (20.20. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20. We leave this to the following Example. (18. This can then be compared with the level splitting we obtained for this potential.
Damburg. Solution: For the solution we refer to the literature. J. Achuthan. J . K. J. J. 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). Silverstone [62]. Bender and T. V.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. Nakai.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. J. J. . Eqs. + 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. R. (357) to (371). " P . J. Silverstone [52].* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. *Some of the literature in this field following the work of C. Paldus and H. *R. Wiedemann [4]. Grecchi. R. of International Workshop on Perturbation Theory at Large Order (Florida. 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. Wiedemann [4]. as in later chapters. Damburg. A relation similar t o (20. Propin. S. Liang and H. J. G. MiillerKirsten [164]. 1982) [139].2: Late term behaviour of double well eigenvalue expansion Derive for the case of the double well potential — by application of a nonselfadjoint boundary condition (as in the case of the cosine potential) — the imaginary part of its eigenvalue and hence the large order behaviour of the asymptotic expansion of the double well eigenvalue. Gram. Cizek. H. J. MiillerKirsten and A. S. Paldus. J.^ 20. Grecchi. E. The relation (20. Cizek.Q . J. K. M. M. 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. Large Order Behaviour of Perturbation Expansions Example 20. it can be used to obtain the other.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. J. V. t p . Harrell. S. E. H. [19] can be traced back from articles in the Proc. Propin and H. Wu [18]. in t h a t when the quantity on one side is known. R. M. MiillerKirsten and A. T. Harrell. W.502 CHAPTER 20. W.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . Its usefulness has also been demonstrated there. Graffi. Harris. Achuthan. W. . t The relation (20. J.
Chapter 21 The P a t h Integral Formalism 21. Chapters 2 and 5. This formalism is particularly useful for evaluating scattering or transition amplitudes. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. Feynman [93]. ^ Our interest here focusses on quantum mechanics. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. The path integral is. unknown). however. P. A standard reference is the book of R. Feynman and A. Felsager [91]. 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. Kleinert [151].* 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.e. in general. The formalism is not so useful for boundstate problems (in this case the initial wave function is. Feynman [94]. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. However. *R. P. Probably the most readable introduction. Hibbs [95]. For the application of path integrals to the hydrogen atom (i. can be found in the book of B. Schulman [245]. A further wellknown reference is the book of L. Coulomb problem) we refer to existing literature. since in such cases the probability of this system choosing a path far away may be considered to be small. which explains also difficulties. R. S. P. 'See in particular H. 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. A brief introduction is also contained in R. We illustrate the method here by rederiving the Rutherford scattering formula with this method. However. 503 .
ti){x. we wish to know* ip(xf.x).ti)ip(x. U.2) where K is a Green's function.tf\ip) . */ = U + (n + l)e. the initial time. =J dx(xf.504 CHAPTER 21.e—»0 Sn. x. The Feynman Path Integral Formalism 21. ti).x. (21.tf\x.tf) = / dxK(xf. (21.t). (21. Obviously K(xf.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo.ti).ti\tp).tf.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). (21. The Lagrangian L of a particle of mass mo moving in a potential V(x) in onetime and onespace dimensions is given by L = TV= m0x2V(x). be ip{x. i.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. *In Dirac's notation we write this equation later (xf.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf. ti) = 5(xf .1) Let the wave function i/> at time t = ti.5) Sn = X / fc=l mo (zfe .e. we can compute the action S1. 2mo dx2 (21.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.
21.t). the Fresnel integral (13.1 A path in (x.1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig.6a) . 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.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e.21. Here and in the following we need one or the other of the following integrals (cf.tf). We now consider intervals in position coordinates and in particular the following integral over intermediate position coordinates Xk which corresponds to the sum of exp(iS/h) over the uncountable number of possible paths from Xi — XQ to Xf = xn+i. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi.28)): f dr\ exp ia a) ' (21. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation. 21. The usefulness of the method therefore depends on the fact that some paths are more probable than others. of which one is illustrated in Fig. as will be shown to be possible below.ti) to its final position at (xf.
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.m0 .6c) In the evaluation of these integrals we replace a by a + ifi.8) Note t h a t in spite of the n integrations. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.Xi.6b) —c ia o drjr] e x p—V € (21. J—oo dxneiS/h — • 0. dx exp z—{a.5)): G(xf.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT. (21. dxneiSn/n.37)) dxk exp d(xk .. n > 0.mQf Xk L ~2e^ ^ . ie f me\ 2a\a~) 1/2 (21.xk+x oo —oo oo .t ~Xk+l> + xk+i 2 ) +{xk . (21. 2 + 2a.506 CHAPTER 21.xk+i) exp . one has j ^ dr1eifa+i^ /£ = .Zfci) 2 } I +{xk+i x*. for example.. We then integrate and finally let /x go to zero. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.2 + (x + x f c + 1 .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.. tf > U.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.x f c _ i ) 5 \m0 J (21.0 ^7 (x f c + i . One therefore removes the troublesome factor by defining (Sn being the sum in Eq.tf.z f c _i) 2 } } .(x fc+ i .ti) '•= f g m ) (N(€\]n+i dxL.xk_i) dx exp i — { 2 a .s f c _i) + (xk+i .
ti) for tf > U.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman.21. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. (21. i.2 The Path Integral and Green's Functions 507 Eq.26)) that — ignoring the potential — the multiple integral (21. 21.7) for n — 1) to give a time interval. in fact. It is neither a variable of integration in T){x} —> Ylndxn. i. and will therefore not be attempted here.t In the following we demonstrate that the function constructed in this way is.tf. (21. However. however prefixed with a new factor.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure.5). which all vary from —oo to oo. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. 11 For discussions see e.24). we arrive at a path integral without the measure factor N(e) above. (21.g. p. 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.5).e.Xi. B. . B.ti) For tf > ti we now write* G(xf. (21.2). (21. 183. G(xf.e. These integrations are indicated by the horizontal arrows in Fig. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq.U) = / JXi=x(ti) = K(xf. See the square root in front of the exponential in Eq. it may be noted that with a different philosophy concerning the summation over paths.Xi. nor the limit of integration of some variable Xi.Xi.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. Different constructions of these families of paths usually lead to different measures. Felsager [91]. 183.g. as in our discretization in Eq. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. We shall see later (with Eq. and so on. (21. (21. summation over all possible paths. § See e.tf. the Green's function K of Eq. p.1. Felsager [91].tf.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).
One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.1 and allowing n to go to infinity. I We employ this method later in Example 23. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. Consider a time t = tf + e for position Xf = xn+2. 187 .188.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf. We have G(xn+2.tf + e. . It is for these reasons that — unless required — the measure is frequently ignored. we have G(xn+2. so that the piecewise linear paths fill the entire space of paths.ti) = 1 j^p: f°° cte n +iexp } • fm0.74)). Then we have on the right the factor G{x+rj). Felsager [91]. pp. one with the full action S.tf + e. (23. (21. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. however.ti).508 CHAPTER 21.exp 2e n ieV(xn+2) G(xn+2 + V.3 in the evaluation of I a specific path integral (see Eqs. we obtain the same as before except that there is no factor N(e).72) to (23.ti. Each of these linear pieces is completely characterized by its endpoints. the discussion in B. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig.Then we can construct a recurrence relation as follows. Then the measure drops out.xi.U) imo 2 G(xn+i W)Idr. 21. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths.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. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.Xi. Expanding this in a Taylor series around x. we have See. the other with that of the action of the free particle SQ.Xi.tf.
ti) +\if(^G{x.tf. U) + .tf. we obtain: G(x. tf + e.Xi.Xi.xi. tf.13) We integrate with the help of the integrals (21.ti) .14) = G(x.xi.— p G ( j . tf.tf. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.tf. tf.tf)= dxG(xf.xi.ieV(x) + ri< —G(x. tf.x.tf.Xi.Xi.6b) and (21.15) We see that G(x.tf.ti) = j^rr dnexp I ^rj1 .tf + e.16) .ti)ip(x.t Thus + 0(e2 (21.ti) = 1 d2 + V(x) G(x.(21.U) +••• \ 509 G(x. Xi. It follows that ip(xf.tf''Xi.xi.6a).xi. (21.Xi. Retaining only terms up to those linear in e.G(x. d and this means d i—G(x.xi.tf.tf + e.tf.xi. .ti) = G(x.ti) +V(x)G(x.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.tf.xi.6c).Xi.tf.tf + e.tf.Xi.21.ti) for tf > ti (21.U + 0(e ).2 The Path Integral and Green's Functions so that G(x.ie 1 d2 2nV0dx^G{X. 2mo dx2 (21.Xi.ti)\ . U) = e^G + 0(e 2 ) dtf 1 d2G(x.U).Xi. U) . ti) satisfies the timedependent Schrodinger equation.ti) 2 —l€ — ^2 + V(x)G(x.ti) N(e) or G(x.
x.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. as may be verified by inserting this into Eq.Xi. denoted by KQ.ti) = 5(xj — x).tf.U).*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. 21. K0 = 9(tf . (21.tf. In a problem where V is small it is convenient to calculate K with a perturbation method. (21.1)).15) is a wave function (cf.tf.x.tf.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.16).x.U) = G(xf. (21.8)) Go(xf. The result. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.3. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. We have therefore established that the Feynman integral (21. or K(xf.U) = / T){x} eiS\ [fdt \mx2. and the Green's function KQ\ S0= Here (cf. now called So. (21. Eq. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. is the Fourier transform of the nonrelativistic propagator.U)G0.510 CHAPTER 21.20) .ti) = = 9{tfti)—G + G5(tfti) . The function G which satisfies the timedependent Schrodinger equation (21.x.U.10) is indeed to be identified with the timedependent Green's function. (21. Hence our original function K is the same as G. as this is also called.U)6(tf .17) Differentiating K with respect to tf we obtain —K(xf. 21. Eq.
TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.22) • \ n/2 . 2t Go(xf.25) T h e quantity Ko(x.23) We now insert our expression for In into Eq.tf.23) for Go and obtain Go(xf. We obtained the same expression previously — see Eq. exp 2 (21. we have Go(xf..t) K0(x. (21. (21.tf.ti)9(t).3 The Green's Function for Potential and hence Go{xf.1 we show t h a t oo /. (7.ti) = = G0(x. . oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a. In Example 21.Xi.tf.Xi.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function./ .tf.tf.Xi.x 0 )" n + 1 (in agreement with Eq. But t = tf — ti = (n + l)e.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a. (21. exp where Xj = xo and Xf = xn+\ and there k = 1).Xi. Then Go(xf.37) — by explicit solution of the differential equation which the Green's function satisfies.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.8). n + i .21) fe=l and t = (n + l)e = tf — U and h = 1.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.Xi.Xi.z / c .ti)= iX (o. f c .Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.tf.21..24) ( s / . (21.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.ti) Ko(xf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.i ) 2 .ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.
1: Evaluation of a path integral Verify Eq.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i. In our proof of induction we assume the result is correct for n. i.t0) (21. the latter being a Green's function with interaction like that for the harmonic oscillator.ti. or else by induction.26) with n integrations and n + l factors KQ.e.x „ + i ) 2 = ( x n + 2 Xi + Xi .5) is carried along.f_^y e. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 . ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l. (21.. (21.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.xi)2 = 21 xi ^2o(„. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 . Then oo 2 / (21. Example 21.xo)2 + (x2 ._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.tn)••• K0(xi.512 CHAPTER 21.6a).22).x0.54). e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .10) in a product form.)!} J — OO .x ^ 2 + y2 . For n = 1 (see below) h This result can be verified like Eq.25) we can rewrite the Feynman path integral or kernel (21.22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. (7. (21.x „ + i ) 2 = (x„ + 2 . The Feynman Path Integral Formalism We can now see the group property of the Green's function or propagator by observing that with the result (21. Eq. (21. ^ (xi .2y(xn+2 . If the potential V(x) of Eq.tn+l•xn.7) by direct integration with ^ .)+(1"+2.. Solution: The result can be obtained by cumbersome integration. as / JXi f r D{x{t)}eiS°lh . we have instead of 5*0 and KQ expressions S and K.
2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ.21.i) H —3/ . (21. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A . We pass over to momentum space with the help of the following two integrals.a .i) .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m. n + 2 . (21..x2rriQ 1/2 toij_.2y(xn+2 . (21.2.3. This quantity is also described as the nonrelativistic free particle propagator. it . 21.29) Hence with KQ from Eq.27) we have (with a = t/2m0) . The first.2 y ( a . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. Using Eq. (21. The second integral is an integral representation of the step function.n+2a. e > 0 .Q . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i .25): Ko{xJ) = = (S) e{t)(^X'2 y J . ) H — (xn+2Xi) n+2 n+l n+2 y2 .6a).Xi) + ( x n + 2 Xi)2.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.27) '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.e.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj).i j ) H n +— z 1 . A(n + 2) 21. a dp exp [ipx + iap2] = exp — i(21. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 . 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. i. and + 2 2 l n + 2 ( x n + 2 . (n+l)A n n j(a.
28) this becomes K0(x. an expansion in rising powers of the coupling constant (contained in the potential).2 Contours of integration. 21. ( 3 b) ° The expression E+ V — it 2m. the second order two factors of V and so on. Thus in first . This is the type of expansion most people describe as perturbation theory.Q (21.51)) is fairly selfevident. i. The generalization to three space dimensions (required below in Eq.e.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£.*' where 2mo — r. Replacing 9(t) by its integral representation (21.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. The first order correction contains one factor of the potential V.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. 21.514 CHAPTER 21. The Feynman Path Integral Formalism Im t \t>0 Rex Fig. (21.
21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/. .t)\.)} (21.. \2meJ n+l J.t) JSo dtV(x. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.t)] n+l .. dxn exp i ^ —. dxr. } e °° f l f •/ oo /"oo F ( a .U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2. (21.(xfc . (21.t) the first order correction is (G\ ~ G — Go) Gi(xf.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.20)) and with elS = exp i S i o / hi dtV(x. .ti.i.U).t/. (21.tf.i. li f f 515 order (cf.21.xfc_ OO J— OO J—OO L J.32) Since — we recall this for convenience G(xf.5) a n d (21.i V e / D { a .Xi.33) xG0(xi. dxi ( m0 V V27rie/ m0 \^He \ exp fe=. Eqs.Xi.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq..4 Including V to First Order: Perturbation Expansion h=l. x (n+l)/2 \ lim .^._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 .tf.^){iV(a. ..^ J_00 V(xk.xi. .
t2) xKo(x2.x2.e. t') + • • • . / 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'.37) Jti The last two contributions do not describe the time evolution from ti to tf.t)V(x. = i dt2 / dx2Ki(xf.U) xV(xi. .516 CHAPTER 21. (21. They must therefore be deleted. This is cancelled by n! different time orderings. t) Ju Thus the term containing n times V contains a factor 1/n!.t2)V(x2. The factors of "V" originate from the series expansion of the exponential exp ftf .36) {if •j.e.ti) t2)K0(x2. i.t. t2. in writing down the expansion exp i dtV(x. (21.I dtV(x.ti).xi. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.t) =li dtV(x.tf.ti)K0(x.x.ti).xi.t2)V(x2.x2.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. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.t2.tf. Jti Consider the term of second order.t)K0(x.35) We also observe here that the problem of timeordering arises only in perturbation theory.i / dtV(x. Xi.t) (21.xi. i. dx2Ko(xf.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). OO /"OO / oo dt / J—oo dxKo(xf. (21.
tf. These may be formulated in configuration space as below.i ] P I dt J dxKn(xf.t2 K o (x 2'l2 .t)= f dpdEeipxiEtV(p.x. x"^ X x"2. . (21.38) . or in momentum space as mostly in field theory.ti).t)K0(x.t)Ko(x.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. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.21. Here.t) ) K K K0 'V X K o Fig.Xi.t)V(x. 21. called Feynman rules. U) (21. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj.tf. x r l i 'V(x.Xi.3 Feynman rules and representation of the Green's function.tf. Such diagrams in the context of field theory are called Feynman diagrams.t) is V(p.Xi. E) which is given by (in onetime plus onespace dimensions) V(x.ti) i dt = K0(xf.xi.3 along with the diagramatic representation of Eq. 21. (21.t.Xi.39) dxK(xf.x.tf. tf.39).E). we are concerned with quantum mechanics.t.ti) (21. The Fourier transform of V(x.ti).Xi.ti) = K0(xf.40) The rules for the present case are given in Fig.t)V(x. of course. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. For various reasons — such as illustration. The diagrams represent mathematical quantities and hence are designed on the basis of rules.tf.
Xi.ti).5 Rederivation of the Rutherford Formula As an application of the path integral method we now consider scattering of a particle off a Coulomb potential. (21. The plane wave function normalized to one over a box of volume V = L3 is. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. (3) The transition probability per second is = \A\2 . i.t) = .k = 2wn/L. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. ^(x. mo V since ip is normalized to one particle in volume V. T = duration of time for which potential is switched on.= l. . The Feynman Path Integral Formalism 21. There the wave functions ipi.ipf have to be properly normalized. 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).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. with p = Kk. to one particle in all of space. (21. i. (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.tj)K{xf. dn = (L/27r) 3 dk.518 CHAPTER 21.ti)il)i(xi.e. and dn = dnxdnydnz. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics. here with E — k 2 /2mo. We recall for convenience from **Note that from the wave function we obtain.41) We have to pass over to three space dimensions. * k 1 • .e. for the allowed values of k: kx = 2imx/L.7 = e i k ' x _ i : l i with f i/}*i/jdx.tf.
i. V(x.4). and with these the following perturbation ansatz: ih^* with an(t) = HV. 47T 1 3 7 (21. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.44) = u(Ek.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.= t7Ei4 1} wi 2 k It is frequently assumed that the only time dependence of H'(t) is that it is switsched on at t = 0 and switched off at time t > 0. Proceeding in this manner one obtains Fermi's "golden rule". Then H'ki can be taken out of the integrand in Eq. in which T is a large but finite time. (21. P(*) = j r (2145) (cf. Eq.44) and f_oo can be replaced by fQ.— . (10.t) = e~^lT\ r where r = Ixl.117) and Example 10. «= — . The transition probability per unit time is .21.. (21.47) . # = ^an(i)VnM) n = a^(t) + \a^(t) + .k. u = ^\H'kfp(k). Then for stationary states M^t) and itm^ = f HUt'V^t'dt'..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.e. H = H0 + \H'.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.
x.30a) K0(xf.tf)Ki(x.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.. also called Born approximation.ti)'ipi(xi.x. we have to replace the onedimensional quantities dxf etc.i x (i)V(x.t) = ^ J*ie*<*'*><<*'*>. (21. by threedimensional dxf and so on. t. (x*.^ / .tf)ip*(xf. (21. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.t. we obtain A1 = clS(ti tf) / dxf(i)V(xf.49) i.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .50) Here x — xj — Xi and t = tf — ti.520 CHAPTER 21. £.Xi. ti) = co5(t — U)5(x — Xj) etc. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x. x.tf. Ai = dxf / dx. (21.39) OO /"00 / dt / OO J —OO dxK0(xf.t)V(x. We can relate this amplitude to the amplitude ark of Eq.yLi.tf)ipi(xf.>/>.f.xuti) = ^ J°° d p e i p x _ i ^ * .ti). T a large but finite time. (21. Thus with the ansatz Ko(x.ti). dt dxip*f(xf..44) (here in the scattering problem with initial and final energies equal). t.tf). This now has the form of the amplitude (21. The threedimensional generalization is (taking tf » T and ti <C —T.f.tf)K0(xf. (21.51) Inserting this into Eq.* / .tf.t)K0(x. t)K0(x.tf. Xj. In our previous onedimensional considerations we had with Eq.44) by contracting the propagators to instantaneous point interactions.tf. i. (21. The lowest order approximation of A.48) where from Eq.e. so that 0(t) can be dropped) K0(xf.52) .t) (21.tfx. (21.). (21.
56) (21. Then we require the following integral. this becomes A(p.~ .fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .k.54) Now inserting the expression for the potential V(x.).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. 2 „ x v~ m o (21.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. (21. Then integrating over q and q' we obtain . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i 2mo (t — U) + ik • x 2mo .p ) .59) . £).p 2.T) = JdtJdx(^\V(x. k2 ipx + i—t + ikx — i 1 2mo 2mo (21.58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.iaT2/8: L With 0 0 • .57) we have A(p.k. . (21. P2 .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 .T) = x exp L z/ q/2 k2 521 f dx.t)exp A(p.T) = ^JdXJdt%*t2/T2exp i ( k . which is evaluated in Example 21. (21.i ( k 2 .r) = = y .k.k.
< ^ £ \M VTTT2. i. 66) o . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . This expression assumes that the integration over time t is normalized to 1.e. Using the result (21. T ) ~ .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.T2f 8 V/ 2 .59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . k .„.64) Inserting the result now into our expression for the total cross section. otot contains the factor \A(p. y 2 [ ( k .^ . For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . ( 21 . mo m (21 .60) Now. dE = PjV = ^ E ± . In our case this has not been done. (21. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).a ( — ) ^ . (21.65) But E=*.63) We establish this result for T — 1/e in Example 21.522 CHAPTER 21. e 8{k) = lim 7=*—^ (21. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. k2 .P ) 2 + M2]2 4 VvrT2.63) we have: Probability per unit time = 77777.\ 2?7lo / 1 (21.k. <rtot. e (21. write down the probability per unit time which is ( 8 \1'2 _a2 16^ .59). .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.T)2/T.3. Using Eq.62) We want to consider the limit T — oo.„ 2." + M ] p) ' "•"• ^2 . { I i _ F T l } .51 — 2 /w2 y ~ —. we obtain . / / P 2 _ .
69) This is the wellknown Rutherford formula. 2mo / dft. This result may also be obtained from lyukawa of Eq. (21.21.cos 9) = 4k 2 sin 2  .p ) 2 = 2 k 2 ( l .M £ M * = / m0kdn = m0^2m0E (k . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21. Then (x = r) foo „2jj. —Mr 4w = .j . (21. setting M2 = 0. and inserting the expression thus obtained into the final result of Example 10. giving the potential a coupling constant.59) by replacing there q 2 by ( k — p ) 2 .^ .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 — .e.e .* 9 ' d r e .2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .^ .67) and (21. we obtain f mlV2 dQ v / 2 m ^ 3 2 7 T 3 a 2 atot r) 3 fcV 2 4fc 4 4sin 4 (0/2) ' JW) da _ d& = i.67) T h e delta function here implies energy conservation.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) . 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 . p 2 = k 2 .65) with the cutoff M > 0. so t h a t (21.j .68) Inserting (21.70) .4.= e " f c a ^ o 2v^fa /4a = lim £^o —. i. /Yukawa = ) Solution: We choose the 2axis along the vector q.^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. (21.e. Example 21..68) into Eq. esfn (21.
(21.6 P a t h Integrals in Dirac's Notation It is instructive to devote a little more time on the Feynman formalism by rewriting it in terms of Dirac's bra and ket notation.oo J / )eikx. however..70) and (21. co oo dx co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). Since dxe" 1 1 1 = « / .524 The function required is CHAPTER 21. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx.75) . a simpler method suffices. We have x> = I J ^ l p X p l x ) . Differentiation of (21. We write the state vector of a particle of threemomentum p in the momentum space representation: p). (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 ) . In configuration space the state of the particle is written x). ' i a = lim e —0 pk 2 /e2 e^TT 21. Va we obtain g(k) = Je~k V a = 0.x) (21.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p .* . In the present case. a.71) with respect to k yields the same as °° 2 .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.q). Direct integration /4a .72) co From (21.^0 a^C 2^/TTO.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 consider again nonrelativistic quantum mechanics.
4) is to satisfy the Schrodinger equation. i. The equation 1 = J ^p)(p. (21. Thus (qPp) = p(qp) = p(27r)3<5(q .t).t) = +iH\x.. x. t) = (x.. i.t).x.160. d/dt\x. (21.^ l x ) .21.t'x. Then ( X  ^ .p).77) (21.If '/'(x.i) = e iH4 x.83) (21. at initial time ti to its final position xy at time tf. i. Similarly we have a position operator X with Xx) = xx).0) = e""x). The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H. so that \if>t=o)s = H>)H.We define x.81) Comments on signs in Eq.t).i'x. i.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . Hence for the infinitesimal transition (x'.79) Feynman's principle is expressed in terms of the propagator or transition function (x'.£) = ( x V ^ ' .80) (21.We now introduce the Dirac bra and ket formalism. The time development is given by the Hamilton operator H. t\ipo)s. We consider first an infinitesimal transition. See also L.e.t) = —iHip(x. as had to be shown.t)} constitute a "moving reference frame" in the sense that ip{x. so thattt d — x.i) = K(x'. 159 . x' — x infinitesimal.t> =iH\x. i.t\.82): The general Schrodinger equation has the state vector to the right. . (21.e.82) (21.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). which describes the propagation of the particle from its initial position x.e.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.76) These relations imply the completeness relations 1 = J dxx><x.t) = e* H t x). H. Ryder [241].e. (21. Then the states {x.t). (21. t' — t. pp. M^ip(x.t'.e. (with ft = 1) d/dt\ip) = —iH\tp). ) H = (x</>o)s. we must have d/dt{x.t\ = iH{x. (px) = e~ipx.
Pq) = J 7^3Pp)(pq) = J dppp>*(p .V 2 < 5 ( x .P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.89) Using this and previous relations we obtain for the Hamiltonian (21. (21. (2187) (21.y<&(p'ix')(x'ivix)<Xp) = f dxe.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).i p '.x y(x)e i p .85) the amplitude or expectation value or position space representation (x'#x) = .91) .90) (P'IVIP) = y<&.526 CHAPTER 21. The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P '  e ""'"'' )  P ) < P  X > e.x).e.88) Thus P has the operator representation P= /dxx)V(x.* .x = V(p .x') + V{x)5(x' ..p'). 2mo Since (21. (21. "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (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). i.
i.i'x. But e'ipx + 0[{t . (x'.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.p) + i{t . 527 (21. tf subdivided into infinitesimal transitions (xj+i. EQ — Hi .t) (cf.21.92) We now return to the expression (21.p') + V(p . for the amplitude of an infinitesimal transition.tj+i\x.t') \^(27rf5(p 2mo +V(pp' where we used (21. (21.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.92).p') • • • j W < 5 ( p ' . (21.84).e.95) when applied to the bra vector (x. We now consider the transition of the particle from Xj.t'f / w? eip''x'^(p ~p/) = / l r ip''x' / "x"e~JP'x"v{x")e" = = Hence + y(x') 2771Q d /dx"<5(x'x")F(x"KP'x' y(x')e i p ' x '.p'). i. We have <x'. t).j.tj). ti to x j .6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .fx.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 .82)). Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj. the signs are reversed when these operators are applied to the ket vector \x.
ti.. . x n which the particle passes at times tQ..2. zmo T h e n t h e integration over all p ^ can be performed by using Eq. d x i . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/. . (21.x i ) H h pn • ( x n .i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .X n _i)] h) (21. d x i ( x n .xk (21.528 CHAPTER 21. We now insert for each of the infinitesimal intermediate transitions the amplitude (21.a — —e/2mo and divide b o t h > sides by 2TT. dx. x i ) ( t i . ..tixo... d x „ _ i .e.i/xj.xn)(tn in_i)]. x i . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21.xic) by fc — Xfej H(pk. x exp [ .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . i.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf. t n  x n _ i . T h e integration over the n—1 intermediate positions.tn respectively.tj) = / ••• / d x „ _ i d x n _ 2 .x 0 ) + p 2 • (x 2 .+ V(x f c ). then corresponds to the sum over all paths.tf\Xi..27) in which we make the replacements x —> eifc.t 0 ) + H(p2.i { f l " ( p i .99) .94) and thus obtain a phase space expression.98) tk — *fcl # ( p f c .t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . .. d x n .ti) rx/=x(i/) d x i dx2 .97) H hH(pn. i. .dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .. = 1 X X<jPfc We now replace H(pfc.tfei) fe=i .e. x f c ) = ^ .tf\Xi. . (Xf. . t n _ i ) t„2><xi. .
] ex H ^I (. (21.H(x. (21. Truman [254].6 Path Integrals in Dirac's Notation 529 Forming products of expressions of this type we see t h a t this relation can be generalized to the threedimensional case. G. .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.ti) = J D{x(t)}D{p(t)}expi [' dt\pi. > Thus the phase space representation of the pathintegral* (21.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.21.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21.e. Tokarev and A. i. to / m0 \ 3 / 2 las. Smolyanov. i.tf\xi. (21.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . (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.98).8) as N(e) so t h a t the integral does not vanish for e — 0. G.e.101) Then Eq.p)]\. / {xf.103) is another form of the pathintegral for which the integration with the help of Eq.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.. A. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.
t. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x.530 CHAPTER 21. (21.tf\xi. £ J ) with no operator in between. We have now obtained the path integral representation of the matrix element ( X / . (21.104) where for the particularly simple Lagrangian L under consideration here m0±l .t'}. operator (21.ti) where nl D{x(£)}exp '. Garrod [106].8). 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'.e.tf\xi. x. . Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. pk) = L(xk. i.iJ(x fc .x) (21. p . Thus finally we have Feynman's principle: (Xf.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. £ /  X J . (21. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. xfc).V(xk) = m0±l . this agrees with the corresponding discrepancy in the onedimensional case of Eq.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).x) .H(xfe.105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. pfc) = Pfc • xfc . t to x. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. nl (xf.8). I hi dtL(x. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.
5 l^ t 'lp / > e .*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(. Thus <xV'xx'. x fdxG  ^ P <p"x3)<x3e^V) x (21. ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.107).* * . (px).t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".t"\x5){x5\x'. . .21.* * f dp = j (2^ p) ^ d .v ' x ' and thus {x". With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'.t"\x\x'.t'}.6 Path Integrals in Dirac's Notation For the verification we use Eq.0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. so that we obtain a cnumber position coordinate which we can shifted across other cnumbers. (21.76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). With the wave functions (21. .i p '• x .109) .77) and thus have 531 *!*') = *M = * 7 ^ I P > « .108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e./ ^ I < « . (21.
0 dp x„ / (27T): 1 . (21.£/xxj. This is the expression corresponding to that of Eq. Thus with x^ = xn. (21.£ n xx n _i.tj) at equal time intervals e.94) for the case with no operator in between. tn2) • • • (xi. We now consider the transition of the expectation value of operator x from Xj.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .85) (xVxx'.t') dp .532 CHAPTER 21.f ) i f e i p .110) Recalling Eq.t'){x"\H\x') = X + •••} (21.t') (xV'xx'. ] .£'> = x"[«5(x" .it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .i(t" .i t " ( x "  #  x 5 ) ( x V ) +. . (21.x') . we obtain (here and in the following the limit n —> oo at the end being understood) (x/. .x ' ) (2^ i^+vM + D «p(x"x') (21.ti = to as before.£j to Xf.113) Inserting for each of the infinitesimal intermediate transitions the amplitude . t0).£j) = / ••• / dx n _idx n _2 . x / / e . we obtain with (21. (21. ( x " . Inserting here the integral representation (21.• dx1 (x n .t"\x\x'.*/<x"x5)(x5tfx'> + . = x"6{x" .74) of the delta function.tj+i\x\xj.i{t" . tixx 0 .i ( t " .90) we can rewrite this expression as (x".t n _i) x ( x n _ i .x') .iit!' 1') 2mo + vtf) + <5(x"x').112) where x" is a cnumber.tf subdivided into n transitions (xj+i.tf — tn and x$ = XQ. We now proceed from there along parallel lines. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK. £ n _ixx n _ 2 .111) 1 ..
dyi l xi x 2 .« • > We reconsider therefore the transition amplitude (x".... Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.{^( P i.t"\x'.2 .ti) 533 rxf=x(tf) = / • • • / dXidx.10) or (21. .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.t"\x'.. dxnn _i Xi X2 .t') = and con 5x"Vx„{x".7 Canonical Quantization from P a t h Integrals We saw above that in the path integral method the evolution of a system is expressed by a basic functional integral like (21.t"\x'.. x n _ i x n reduces to the single factor x^.xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. (21.114).x 1 )(t 1 Z & ) +H(p2.y (27F (27F •" (2^(2^)3 exp[.2 .112).115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .3 " n d x f c . (21. We demonstrate first that the operator representation of P given by Eq.iyxxj.and hence may be put in front of the integrals to yield (xf. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .x n _ i ) ] . (21.89) is consistent with the conjugate expression dL (2L116 * . tj) = xf (x/. .tf\x.x2)(t2 ~ h) + • • • + H(p ••• + p „ ... the string of factors xi X2 .114) x exp [i{pi • (xi .7 Canonical Quantization from Path Integrals (21.( x „ .x 0 ) + p 2 • (x 2 .t') sider the variation S(x".{x".\Xi.t"\x'. Xn _i X„ xi dx.105).t') .t') = (x" + 5x".21. x (21. . we obtain* (xf. tj \ xt. U). tj  xj1 x».t"\x'.x . For simplicity we consider the case of one spatial dimension. n—>oo *•»• 21..117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq.t') (21.
Then. " (t). we separate the metric factors from T){x} in Eq.i/(x)/fi ) j ^ ) dtL(x.122a) = 17) 5x"Vx/. dx dt \ dx _ (21.(x".120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').118) — = (21 5x"(x".tt).119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. Thus in this case of classical mechanics dL_ d fdV.10) and set D{x} = D°{x}/JV(i/ .t').l&cdi. . " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.x). and we demand the validity of the equation of motion.t') =  6— / 1 V°{x}t .t i_(dL_ x't' h\dx J t„ (21. <9L' dt —dx + — 5 x ox ox f . Hence w « = (t). (21. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. x) Jt Jt dL.t"\x'.t'): 1 fx" 6(x". For the purpose of enabling some formulations below. The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.t"\x'. (21. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. jt/ 37 ( £T. dL r /•*" . t".534 CHAPTER 21. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. with N = N(t" .118) JI(x)/h . (21.
.125c) Corresponding to our relation (21.t'). The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.21.Xf = x(tf).124) where with A = tf — ti.x).89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.t')(x'.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. (21.115) in three space dimensions. (21. . Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). 6 (21.125b) and the measure property (3) N~\A + A') = A^. / Jxi V°{x} / Jx' V°{x}.tf\xi. ^2\x'.7 Canonical Quantization from Path Integrals or JL(x»y\x.126) R. I(x) = / JA dtL(x. Wightman [264]. Streater and A. S.1 (A)iV.125a) (2) the completeness relation holds. (21. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.1>) = ^{x".122b) By comparison with Eq.t'\ = l. § The evolution of the system is expressed by the Feynman functional integral (xf. x' (21.t"\p(t")\x'.1 (A / ). 535 (21. F.
t"\x'. (21. Eq. (21. (21.131) = ih(x".t'). (21.t'). Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.132) The generalization to higher spatial dimensions proceeds along parallel lines.. tn lie correspondingly in consecutive order between U and tf.t'){x\t'\x2\xi.ti).536 CHAPTER 21.tf\x(ti)\xi.t"\x'. Thus (xf. We consider the case n = 2. (2) and (3) above.t>). (21.127) We can now write down an expression for a timeordered product of operators. (21.t\x'.t"\x'. .129) = ^ y With Eq.tf\xi. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.128) where tf is later than ti and t\. (21. Then the relation follows with the help of the properties (1).127)) (x".133) = ih{x".t') (21.t') + x"^J^{x".. We can rewrite this expression as the commutator relation (ITU*" x"^](x". Thus if T denotes such ordering.tf\xi\x'.t') \ i ox" i ox"J or in commutator form [P.. (21.ti) = x' 'Y^{xf.t"\x'.ti) = —!— p V0{x}xx • • • xneu^h.tf\T(xix2)\xi.t"\x'.t')=x"(x".ti) = Xi(xf. we have the relation (xf.122b) we can deduce the nontrivial canonical commutation relation arising here.114) (xf.t').ti) f ' V0{x}xlX2eu^lh.130) ~i5x7jX"{x".t"\X'.t') = ih(x. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.x](x.. We have (cf.t\x'.t"\x"\x'.tf\T(Xl • • • xn)\xi.
although the first few terms indicate a convergent behaviour. the modulus of the ratio of the (r + l)th term to the rth term is larger than one.e. for sufficiently large r. as we have seen in Chapter 8. N. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. An essential singularity is really an ambiguity rather than an infinity.e.g. i. Such an expansion parameter is usually known to be or assumed to be small in some sense. p. In physical applications such cases are rare. z = a + i/3 with a — oo or f3 —> oo or both.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. If the expansion is mathematically well defined. i. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. Whittaker and G. 102.e. Such expansions are strictly speaking. it has some finite. Besides.g. Expansions of this type arise. The vast majority of perturbation expansions discussed e. Watson [283]. such as the fine structure constant of quantum electrodynamics. divergent. nonzero radius of convergence in the domain of the coupling parameter. E. It was therefore natural to search for alternative methods of expansion. integrated over) beyond its radius of convergence. T. see e. Consider e. but asymptotic. in the sense of rigorous convergence tests.e. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. in the context of quantum mechanics are not convergent. i.Chapter 22 Classical Field Configurations 22.g.g. the series converges. 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. W = e x p ( l / z ) . > 537 . so that successive contributions to the first approximation do not invalidate this approximation.
specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". In the following we consider various typical examples.. 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)..e. The fundamental extension of the method of canonical quantization (i.h. and has led to insights which previously seemed unimaginable. A further challenging task was the development of methods of quantization of theories which incorporate classical. i.h3. In searching for other means of expansion.) l/h2. the latter's topological properties if any. such as soliton theory. These methods are often described as methods of collective coordinates. One therefore faces the problem of quantizing a system which is subject to constraints. The classical... A method which achieves this is in particular Feynman's path integral procedure.538 CHAPTER 22. On the contrary. cnumber first approximations. before we reach the stage at which quantization can be considered we have to deal with numerous other aspects such as the stability of the classical approximation. The study of expansions of this type has turned out to be extremely fruitful.e.h2.e. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). higher order corrections are of orders ±. There.g. although this terminology is not precise. nature of the dominant approximation does not imply that this describes classical motion.h°. it is an expansion in rising powers of a semiclassical expansion parameter which plays the role of Planck's constant h in the quantum mechanical WKB approximation. the 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. cnumber. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. and in general this change is accompanied by certain constraints. one considered in particular an expansion which is such that the first approximation is purely classical in a certain sense and such that this ignores quantum effects. such > a series begins with a contribution proportional to (e. This type of expansion is decsribed as "semiclassicar. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. i. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. However. Many of these aspects are interesting by themselves . and the corresponding expansion is called the "loop expansion".
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). to keep this case in mind.g. modified by going to imaginary time).2 The Constant Classical Field 539 and in any case deserve detailed study. . 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. 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. it is important for a better understanding of the considerations which follow. d^] = d^d^ . We begin therefore with a brief recapitulation of the simple Higgs model which exemplifies this case and exhibits the phenomenon known as "spontaneous symmetry violation" or the Goldstone phenomenon. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). or even a solution to a modified field equation (e. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. vortices and kinks. V{</>) = m§0V + A0(<^)2. We write the complex scalar field <f)(x) in four spacetime dimensions The spacetime Lagrangian density of the specific theory we consider here is given by C[4>. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". Thus references to field theory will be exceptional and hopefully do not irritate the reader.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. 22. may like to consult the article of D.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. Tong [272]. However. (22. This means theories with field densities which are therefore infinitedimensional. a static (time independent but spacedependent) solution of the classical equations of motion.22.V(4>).1) ' T h e reader who wants a highly advanced overview of instantons. monopoles.
§ (22.3) See e. 22.g. £)) is bounded from below we must have Ao > 0. i.2) We ask: Are there classical (i. V ) . 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). then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. cnumber) solutions <j)(x) = 4>c = const. V[<t>] m 2>o o Fig. 22. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x.540 CHAPTER 22. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. as illustrated in Fig. of the equation of motion? For these we must have all derivatives of <j) zero.1 Different potentials for different signs of m§. 433.S£/<90(x. B.e.1. p. Felsager [91]. Classical Field Configurations where d^ — (—d/d(ct). m l + \04>*4> )$ = <).£). .e. The EulerLagrange equation is d„ dC d(d. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. and hence here [d^ + m20}(l) = Xo(f (22.
</>2)plane. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. We can convince ourselves that <fic is the field associated with a state of minimized energy.2 The Constant Classical Field For Ao > 0. 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. H = —C = V. 22. TTIQ > 0. 22. But for Ao > 0.22. with 0 = 0. i.e. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>. aq> For (b = const. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. The Hamiltonian density H is defined by the Legendre transform H[<f>.e.5) . The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected.2 The spontaneously chosen phase. w A iP 3ml 5* "XT" (22.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. not the solution x(t)).ir]=7r<fiC. mg < 0. and so ft[<M=m6:(^* 1 + Ao(0*0)2. the equation has this invariance. TT = —J. Fig. we have 541 0. (22. the only such solution is the case we wish to consider.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.
(22. along the trough of V (which we can call a longitudinal direction). 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). &H/d(fi = 0) if [m20 + ^>]</> = 0. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .7b) In Eq.Then m\ m\ = ml + A0A2 = m2.7a) vanish on account of Eq. V?) (22.7b) the constant on the left and the coefficient of A on the right of Eq. and partly by climbing up the parabolic wall on either side. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. we set cf>(x) = <f)c + ri(x). (22.7a) (22.3).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.542 CHAPTER 22.e. i.3ml = 2ml = ml + TA 0 A 2 = 0. 22.2.g. Suppose we choose one such configuration. (22.e. the one for (3 = 0.3) obtained above.2) and separating real and imaginary parts. (22. and we wish to investigate the behaviour of the field cf> in its neighbourhood. i. as indicated in Fig. in a transverse direction. .6) Inserting this into Eq. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. Classical Field Configurations the density Ti is minimized (i.tp2.e. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . e. This is the condition (22. >° for m 0 < °> . i.
the existence of individual perturbation contributions). then a massless boson exists. (22. The elimination of the . There the wave function with an associated vanishing eigenvalue is called a "zero mode". i. (3=0 Fig. tangential to the classical path. 22. whereas ipi(x) is y ' " " " " " ^ . Thus the Green's function is similar to that in electrodynamics.e.t').22. 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. ^ = h eip / \ ^2 l . though not identical phenomenon.3 The spontaneously chosen phase seen from above. We can see the problem here by looking at Eq. t. 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 £.7b). In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar.e.r')S(t . Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. t') = d(r . r'. 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.2 The Constant Classical Field 543 We observe: The field ipi(x) has acquired a real and positive mass whereas the field ^2 0*0 is massless! We observe that the field ^2(2) is that component of ip which is directed along the trough of the potential. This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction. of course. irrespective of the question whether the perturbation series as such converges or not. Goldstone's theorem applies to fully relativistic field theories. has to be removed in order to allow a well defined perturbation procedure (i. which. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r.
zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. see e. we shall need only at a later stage. 7711 = — 1 ( 1 6 Minkowski manifold). MullerKirsten [291]. Zimmerschied and H. i. W.£) in one spatial dimension. We make one more assumption concerning V which. Lohe [179]. N. *S. spinless field $(x. Khare [17]. J. Behera and A.d^} = ~d^d^V[^) = ^2^(^j V[$]. A.t In particular the sextic potential can be considered along parallel lines. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. Rebbi and G. 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. t Potentials of this type have been discussed by M.g. equates to zero the coefficient of this wouldbedivergence.* We consider the theory (later: quantum theory) of a real. however. and we choose the metric ?7oo = + 1 .3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. To insure that the energy (see below) is positive definite we require V[$] > 0. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. The Lagrangian density is taken to be C[^. 22. instead we shall consider static and timevarying solutions in important models of one spatial dimension. (22. Later we wish to develop a perturbation series about a classical configuration of $.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. C. . Soliani [237]. F. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail.544 CHAPTER 22.8) where V[<&] is a selfinteraction or potential of the field (to be specified later).* Potentials of higher polynomial order in $ than $ 4 can also be considered.
e.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.13a) We write such fields <j>(x).«{x) = V'(<p). From Eq. (22.13b) We are not interested in just any solution of this equation. such that J > — — — = 0.9) with quartic potential is defined by ^V\g$\ = ^V\g*. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0.22. t) is an observable (which in general — in field theory as compared with quantum mechanics — does not have to be hermitian although we assume this in the present example).g] = \ v m For instance. (22. The first such condition to be imposed is that the energy of the solution of interest must be finite. ^ ^ * + V'[*]=0.e.t) has an explicit timedependence it is a Heisenbergpicture operator. the reason for this condition being that we visualize the classical solution as representative of a lump of energy.§ Unstable classical configurations are also important and will be considered in later chapters. the socalled ^theory = ^V[g$. i. The other condition which we impose (and study in detail) is that of stability.l}.l]. — m$ 1 — cos Im 1 I ^ $ (22. (22.12) we see that they satisfy the Newtonianlike equation <j. In a quantum theory the field $ is an operator defined on a space of states. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. Since $(x. (22. i. The field $(x. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. .11) (22. (22. at least in the present case.12) where the prime denotes differentiation with respect to $ . 1].10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.
. (22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. the classical solution will therefore have to be such that this integral is finite. (22. 22.e. HI tanh m(x>k) 9 Fig.T^ = 0. 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. We observe here already that the integrand of the integral in Eq.546 CHAPTER 22. (22.15) The zerozero component is equivalent to the Hamiltonian density H. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.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>.7r) where IT is the momentum conjugate to $ defined by dC ? r = . <9$ i. n = & . dC • Too = r?oo£ + =r$ = C + vr$ = H.17) Since we have to integrate over all space.e.c = <£z + .14) where the overdot implies differentiation with respect to time t. (22.T = <I>. d.2\dx 2 ' 1 fd& + V[$].4 The wellknown soliton of $4theory.
e. In fact.22.777. 22.3 Soliton Theories in One Spatial Dimension 547 As we shall see below. Again XQ is an integration constant. O±m(xxo) (22.4. m ( x . 22.x n ). In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .13b) for the specific potentials (22. in the case of the first.21) This sineGordon configuration is depicted in Fig. i. (22. < = — tan [e >  ° ] 9 2 rrm/g — Fig. the $*theory. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". (22. 22. Rubinstein [240]. it is not difficult to find classical cnumber solutions 4>c to Eq. . 4m .11). > . — XQ). (22. See J. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq.1 .18) is depicted in Fig.20) (22.18) Here XQ is an integration constant.10) and (22.5.5 The soliton of sineGordon theory. "The name sineGordon is derived from the analogous KleinGordon equation or theory.
10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . . ± 1[tanh. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . In the case of the sineGordon theory defined by the potential (22. Classical Field Configurations The solutions (22. (22.* ( m' m 2 r)] ^2" . Inserting the potential (22.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ).13b) which we can write d \W? = !"(•».9 2 4>{x) = i.1 "»JtKxo) 9 1 . Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. With we obtain g(/>(a.20) are obtained from Eq.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I. and hence 4/71 9<t>{x. <^>(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 . Here the constant is zero for V(<p) and <j>' zero at x = ± o o .18) and (22. 4m In tan tan 5 l/e±m(*xo)tan^£^)U.o) =m7r. or \(<P'? = V(4>)+ const.e.548 CHAPTER 22.
(22. the discussion of Bogomol'nyi equations) in a simpler way. the curves rising monotonically from negative to positive values are known as kinks.22. 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. SE((f>) 5cb(y) 0. Suppose now that we demand stability in the sense of minimized energy. However. instead of doing this now. (22. The relation (22.17). The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". (22. Jackiw [141]. (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) . .4 Stability of Classical Configurations The concept of stability can be complex.22) and that its second variational derivative be positive semidefinite. however. A state of stability is therefore associated with minimized action and/or energy. Then we have to demand that the functional derivative of the energy E(<p) be zero. this does not require the action to be minimized. the others as antikinks. i. The expressions (22. for instance. in the onedimensional Ising model. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. we obtain these expressions later (cf.19).23) See for instance R. 22.e.21) for the energies can also be obtained from Eq.
25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. a very general phenomenon which can be established as follows.13b) with classical solution 4>c. in fact ip0(x) const. = 0. (22. i. In fact.e. (22.25). if 4>c{x) is a solution. dx (22. Eq. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. i. Since it arises from the violation of translational invariance by 4>c. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space.e. namely (p'c.550 CHAPTER 22. and the EulerLagrange equation retains this property. of course. The original Lagrangian was. with eigenvalue w2. (22.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. in fact. This is. The Newtonlike equation (22. in particular the invariance under spatial translations.13b).27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. (22. the integrated contribution vanishes. Thus the vanishing of the first variational derivative yields again Eq. we see that the latter has one eigenfunction.e.26) Comparing this equation with Eq. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. The eigenfunction 4>'c{x) is for this reason called a "zero mode". Classical Field Configurations Since y ^ ±00. if we apply the generator of spatial translations d/dx to the classical equation. it is also called a "translational mode".A. the function 4>c(x + a). Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). dx2 + V"{4>) (22. its eigenvalues w\ > 0. i. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. (22. written down in Lorentzinvariant form.a / 0 is not the same.13b) retains the invariance although the solution (f>c does not.28) . 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.
]. / oo Thus for E(4>c) to be finite. (22. (22. We see therefore that the occurrence of these zero modes is a very general phenomenon. this integral has to be finite. lim [V(<f>c) + const. We can choose the potential V such that the constant is zero. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero. in fact. 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. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. (22.31) Inserting this into Eq. (22. i.e.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. Thus TUC is a zero mode.22. and we have seen that Eqs. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). We return to our considerations of stability. In the onedimensional case under discussion we can rewrite Eq.18) and (22. i.] = 0 or x—>±oo lim V{(/)c) = 0 for const. 0 = S'[UC] = S'[gUc}.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const.e. (22. it is an expression like (22.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22. we are interested in classical cnumber field configurations of finite energy.20) are such configurations in two particular models.32) = V'(</>). (22. As stated above. Then the vanishing of the first functional variation implies that 5S = 0. (22. = 0.33) x—>±oo .17) we obtain oo dx[V(4>c) + const.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.27).
Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. In the sineGordon theory > • defined by Eq.552 CHAPTER 22.c o and another for x — oo. in view of the antisymmetry of eM^. . (22. the potentials have degenerate minima (i. . For a configuration to have finite energy. This number is defined like a charge in field theory by the spatial integral of the time component of a current. . (22. Thus the classical energy or vacuum configuration is not unique. (22. the classical field (bc must approach one minimum for x — .1).e. Classical Field Configurations V Fig.6.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. V{4>) = 0 at different values of </>). 22. ± l . The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. 22. n = 0 .35) and it is seen that in this case q can be ± 1 or 0. . .34) jfc" = e^dv(l since d^k^ = 0. ± 2 .6 The minima of V in $4theory at x — ±oo. oo J—oo y The topological quantum number q is defined by 0. 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. » In the <fr4theory defined by Eq. In the case of the 3>4theory we can define a conserved current k^ by (22.
x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . .4 Stability of Classical Configurations 553 Since we call q a topological quantum number.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. n = 0 . in general. Again we see that E(\<f>c) is finite only for A = ± 1 . . The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . In the case of the sineGordon theory the minima of the potential (22. A property is. Tfl \ (22. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 .7T (22. termed topological. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. continuous values. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. We can see this as follows. First. one may ask how topology comes into the picture. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. . ± l ± 2 .7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). and we call this "topological stability. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense.cos(27rnA)].A2)2 + 0 except for A2 = 1. zero is not a meaningful topological number. . 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.11) are given by ( so that for x — ±oo: > — J.37) for the onedimensional case here.38) 4 V(\<t>c) >m ~2[l .22.
the factor A = exp[ix(#)] varies over a complete circle).33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22.9 the foregoing arguments in a higher dimensional context. the energy is minimized.13b) or ^ ' ( * ) ] 2 = V(</>). 22. if oo / dxy/2V{<l>)<f>'{x).554 CHAPTER 22.41) dxy/2V{$)4>'{x). i. Thus with one space dimension higher there. pp. W. (c) the time component k° is a spatial divergence. (d) the topological quantum number iV arises in a completely classical context. 00 (22.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). Eqs.17)) f 00 > 0. (22. 425 . . From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.^ = N.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). (22. (22.40) (22. (For a broader view we consider briefly in Sec. Eq.e. H.31) to (22.42) Thus the energy has a lower bound given by the right hand side of this inequality.43) **See e.5 Bogomol'nyi Equations and Bounds We have seen previously (cf.428. Classical Field Configurations involves the difference An = rioo — n. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. its integral is nonzero only on account of nontrivial boundary conditions. MiillerKirsten [215]. 22. The inequality is saturated. 00 (22.g. We observe (a) the topological current k^ of Eq. J. (22.
Eq. 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. for instance.44) This first order equation which saturates the inequality (22. (22.n > 1. (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. In the sineGordon theory we can consider.39). y/V(ft) > 0) oo =2nnm/g r<p=2imm/ g / dxy/2V{<f>)</>'(x) oo /•rj)=2Trm/g = / J<j>=0 d</>y/2V(<f>) = = n Jd>=0 rt=Mmlg n / J<p=o d<f>y/2V(<l>) 2 / ^\ — # W 2 1cos— . We discuss this in the following Example. (22.5 Bogomol'nyi Equations and Bounds We observe that (cf.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.^ Clearly it solves the second order equation or (22. m 9 V V / and with x = g4>/m this becomes P IA\ CTnin(0) = = /^ J /^ V ™3 f2W A I A • 2X n—7T I ax\/2ll — cos x) = n—7T / ax* Asm — Z 3Jo 2ir r 9 Jo V 2 2m 2 cos — n — n^[2(l)+2] 2 ~ m3 8m 3 = n—rrin agreement with Eq. Here (V(c/)) > 0. 555 (22. Bogomol'nyi [27]. B.22.40)) this occurs when 4>'(x) T V2V(4>) = 0. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g.19). n E .21). .
with mass m and connected with short strings to neighbouring bobs.e. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. Fig.46) **See e. rotate) only in the (y.t) measured e. from the yaxis as shown in Fig. At equidistant points xn = na along this string we attach strings with pendulum bobs. Solution: Consider a string along the xaxis as described. xn+\ — xn = a) 00 a —> dx. Assume that the pendulums can move (i.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis. 22.t) 4>{xn.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.t)\2 +mgr{\ .2 V dt + k[(f>(xn+i. all being identical. Construct an expression for the energy of such a classical system and its Lagrangian. z)plane and only such that their elastic strings remain taut.g. Classical Field Configurations Example 22.cos</>(x„.t) <j>(xn.7. t)). 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. 22.7 The pendulum analogy.e. We can now write down an expression for the energy of such a system of (say) n pendulums. 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. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. Initially the pendulums are suspended freely in the gravitational field.556 CHAPTER 22.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. . so that oo > u. H. Assuming that the pendulums can move (i. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) . ka —> K. } n= — oo roc —> / L J . Chapter XI. Goldstein [114].ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .
t) of Eq. This number.49) exp(iwkt)ipk(x).49) would imply an exponential growth in the future t > 0 or in the past t < 0. wk would be imaginary and so the factor exp(—iwkt) in rj(x. (22. and so would invalidate the procedure which assumes that rj(x. (22.13b) we have for small fluctuations rj(x. of the continuous curve in Fig.13a) and (22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. Eq. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt).6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. which represents an expansion in terms of normal modes.t) = <l>c(x) + r}(x. (22. It is instructive to obtain this equation by yet another method.47) into (22. Hence we set ${x.50) This equation is identical with Eq.47) where n(x.t) = ° (22'48) (22. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). 22. The equation is called stability equation or equation of small fluctuations.22. We can understand what stability means in the present case.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. the topological quantum number.t): {w With the ansatz r)(x. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.t). is therefore also called "winding numbed.e. ' Equating coefficients of exp(—iw k t).6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. (22. t) is the fluctuation (field). (22.7 linking the pendulum bobs).25) obtained earlier.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. (22. If some w\ were negative.t) is a small . 22. Inserting (22.
52) tan 6 := e±m^xo). In fact. we shall see later that the fluctuations n(x. we obtain V"(<t>c) . For x — ±oo the quan> tity acting as "potential". the zero mode. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction. Inserting (j)c of (22.A \ Inserting <f)c of Eq.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22.50) has the general form of a timeindependent onedimensional Schrodinger equation. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. i. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. . The equation of small fluctuations (22. In fact.e. This means that the stability is not universal but only local in a certain sense.558 CHAPTER 22. Thus for stability w\ must not be negative. (22. Classical Field Configurations fluctuation. i.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ .e.20). 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£^)*.1. V"((/)c(x)). cos46» = 2cos2(2(9) . and need 9 = tan" 1 j e±m^Xo) 1. If that is necessary. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22.2 m 2 + 6m 2 tanh 2 m(x .18). V"(<b) = 2m2 + 6g2(f>2.
We can therefore construct solutions of definite parity.53) = m(x — (22.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. even and odd solutions. Those of the other case are complex and periodic at infinity. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.* For such a potential the spectrum can be b o t h discrete and continuous. Consider the equation fd2 dz2 .55) The differential operator is even in z.2 (22. .22. ^+(0) d 0.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 . i.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.e. ip+. We can investigate the spectrum as follows.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. which satisfy the boundary conditions V>(o) = o. — XQ)] cosh[±m(x — XQ)]: 1 . A  *(* + !) cosh 2 z tp(z) = 0. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.ijj^ respectively. (22. (22.
Since for £ — oo: * the function ^>+ is normalizable provided Z > n.560 CHAPTER 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.e.Z ) 2 .4 n ( n . i. and in the odd case n+\)U+\l)+\(\ i.60) terminate after a finite number of terms.e.I) . when in the even case n ( n .e. A ^ An = .Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22.Z ) + ^(A + Z2) = 0. i. (22.e. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± .59) Setting (22. 2n < Z. A * An = . (22.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.Z)2.l2 = . + l2) = 0.61) Discrete eigenvalues are obtained when the series (22.57) Then the equation for x{z) 1S (22.( 2 n + l .58) = (1 + sinh2 z)~l/2x(z).( 2 n . i. .
\=Kl2 From (22.62) where even (odd) iV are associated with even (odd) eigenfunctions. We can now derive the spectra of the stability equation for the potentials (22. d w = tanhz. for N = 0.e..2Z = ..50) we obtain with z = m(x — xo) tp(z) = 0. (a) ^theory: the equation Inserting (22. From (22.e. 3m 2 .10) and (22. i.54) by setting X = k2 Then 1 .1: w2 = 0. (22. (22. °2 i. (22. N = 0.6 The Small Fluctuation Equation 561 and the function ip. 2 . We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .e.62) we see that the equation possesses discrete eigenvalues w —2 .2. (22. (22.u2)—^V 2 dio and ip(z) = elkzV{to).65a) _dz2 cosh 2 z l=2. w2 = Am2. the continuum is w > Am2.22.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.( 2 .51) into (22. m N<2. m i.63) we see that the continuum starts at A= 2 2^ = 0.1.j < I .N<1.. We can see the continuous spectrum of Eq..63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.is normalizable provided 2(n + .11).Ny.
It is now particularly interesting to look at their wave functions. m d . Classical Field Configurations (b) SineGordon theory: Inserting (22. N<1.8 The zero mode as typical ground state. We know that 4>c is a monotonic function. d . 22.27)) . Thus either spectrum contains the expected eigenvalue zero. . of course. . The zero mode is depicted in Fig. i. (22. .65b) dz2 cosh^J.50) we obtain with z = m(x — XQ) the equation (22. Thus 4>'c(x) has no node.=1A=^_. so that its derivative is nowhere zero (except at ±oo). Eq. and — as expected — this eigenfunction is even. In the $ 4 theory we saw that (cf. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. This property can. also be deduced from d(j)c/dx.53) into (22.. %(x) Fig.66) We see that this is a nonvanishing function for any finite value of x.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2.e.. the zero modes.562 CHAPTER 22. for N = 0: to2 = 0 and the continuum starts at w2 = m2.8 and we see it has the shape of a typical ground state wave function. . Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. 22. .
. Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za.3 .22.70) . .t a n h x . (22.e. d . without quantum corrections). (22.6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to .e. Finally we consider the normalization of the zero modes.oc (22. 1 o rrr tanh x — .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).z 0 )]oo = 8 .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.= E(<f>c).o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. . — XQ) OC Hence the conclusion is similar to that in the previous case. 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 . (22.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. . i.
Goddard and D.e. P. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). Kastrup [145]. J As a first example we consider the expression (22. H. Olive [113]. This question has important consequences. I.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . Affleck [5] who uses in his Eq. we consider E(<P) which we also write as (22.T(4>).564 CHAPTER 22. however. f G .g. ddx {V4>f + v{4>) (22. ^For basic aspects of scale transformations see e. it is important to understand their physical implications.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax .7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. (22.g.72) with T{$) = I' ddx]{V(j))2 > 0. i. H. (2.g. *The considerations of this section extend in part beyond quantum mechanics and into field theory. Derrick [68].* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. .2d. P(</>) = f ddxV{(f)) > 0.41) for the energy of the classical configuration but now with respect to a ddimensional position space. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. *See e.> 4>a{x) = 4>(ax). They may help. Classical Field Configurations 22. I.4) <j>(x) —• a<f>(ax). A. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. One can have different types of scale transformations — see e. for if such configurations exist in a realistic theory with three spatial dimensions. to underline the fundamentally different nature of topological and nontopological finite energy configurations.
79) In order to emphasize that (f> — (p^ in any direction.xd). (22.75) Since both T and P are > 0 and d is integral.e. since d2E(<pa) da2 2(2d)T(<f>). (22. x = (x1. (22.80) .7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1..78) For the energy of a static configuration.x2. to be finite the field <P ~^ 4>oo £ Mo for x — oo. a=l (22. Next we consider more complicated cases. In fact.76) we see that E(<f>) can be minimized (i.. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a.e.73) we see that this implies (2 .71).V(<P) (22. we see that the stationarization of E with respect to a is possible only for d = 1 as in the examples we have been considering so far. we write • = lim J r . we set 0 dE(<pa) da J a = i (22. the second derivative is positive) only for d = 1.74) From (22. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension.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}.d)T(<f>) = dP{4>). i. Consider first the Lagrangian density £[&M = \{d^)(d^) .\ eMo(22.22. i.. (22. d = 1 minimizes the energy. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}.
and. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. of • course. one set can be mapped into the other. i. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . x = 00.e.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. If we go to two spatial dimensions (d = 2). We see. —00. are the same. i. that we have a set of disconnected classical vacua (</>oo><^oo. A correspondingly topologically nontrivial field theory would therefore have to possess a!7(l) symmetry as in the complex Higgs model (recall that in our considerations of $ 4 . i.38).81) and (since d = 1) S = ± 1 . . = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . S1. — 1). as is the case for the constant solutions. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00.1. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. Thus in this case S^1 is a disconnected set consisting of two points. (ptx = <f>oo. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. Then each point at infinity is characterized by a different value of the polar angle 9.566 CHAPTER 22. These observations are summarized in Table 22. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. Table 22. therefore.e.e.and sineGordon theories we assumed $ to be real. —1. is a connected region. —00 or equivalently S = + 1 . in fact. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively).
if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. for instance.g.e.e. i.• M0. map into. Thus.83) . given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. .e. phase transformations 6Q ^> 9Q + 59.e. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. we would have only two points at infinity. $(r) : Sl . We see. the field must map position space into field space at infinity. On the other hand.e. * n reach any other <^(oo) by simple phase shifts. if <p is real. VMx)) with (0(oo) = linvxx) (p(r. 9) would depend only on r = x and hence the problem would be effectively onedimensional. (22. —00 or S = 1 .2. i. This will be "nontrivial" in the above sense only if 0oo is independent of 9. we have w e ca {Mx)}> and 1 3 1. that if the problem is to be topologically nontrivial. Similarly. Thus. to have topologically distinct classical configurations. This can be achieved e. the one point S = + 1 .82) Now Mo := \ 4>(oc) ^24>2 = a2\.22. therefore. if cj> were real and position space twodimensional. cj)(r. i. i. assuming invariance of C under the transformations of the group SO(3) in internal space. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial.e. 00. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. i. i.3 (^real) v{4>) = / (22. If position space were onedimensional.1 .7 Existence of Finite Energy Solutions Suppose. by allowing <fi to have three components in some "internal" (isospin) space.
3. it is effectively a surface (and so a continuous set) S2. Georgi and S. Nielsen and P. through each point at infinity). circle. . invariance under replacements $ — —$ in $ 4 theory.e. % . so that details of these models are irrelevant.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) £(</>. If. Olesen [220].V{4>a]. The symmetry of the potential (and Lagrangian) — i. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories.e. 11 H. the winding number is n.. 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.. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. we cover <S^_1 n times.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. $ — $ + 27 > • T in sineGordon theory.a M J . In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. sphere. A) = T(4>. $ — $exp(ia) in a U(l) symmetric theory. the appropriate rotation. . A) + T{A) + P{<j>). a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. going once around S%~1 (i. i.2. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit.„ where G r = 9 M £ . the subsequent discussion requires only the general form of the Lagrangians. Glashow [108]. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). i. Example 22. B.e. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line.</>] = ~G^Gai. L. + i ( D " f t W ) a . a = 1.e. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A".568 CHAPTER 22. $ — > > $ exp{i Xlo=i Pa.
84) in this way one finds that the quantities involved transform T ( ^ . >• A„ x ( x ) = AAM(Ax). A.e.22.d)T(4>.86) (22. Thus we set dE(<fix.7 Existence of Finite Energy Solutions where T{<j>.84) and investigate the existence of finite energy classical field configurations.d T ( A ) + AdP(</>). negative.A).. and so B ( ^ A .dP(4>) = 0.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. = 0.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). > Since T>IJ. (i)^ 2d T Considering the contributions of (22. i. we have {^/»)A^ (22. A). (22. (22. T(A) = f ddxV(4>) > 0. A) + (4 .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'>.T{A). i.i T(AX) = X4~dT(A).d T ( ^ .87a) C?M„(x) — \2G^{\x). (2 . some contribution must be . (22. A A ) = A2dT(^. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. We now consider scale transformations with A £ R. l j ddxGijGai] > 0. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.89) P{4>) are positive semidefinite and not all zero. <0 >• <£A(X) = <KAx).d)T(A) . ^ A ) = A 2 .<^[x) transforms like a vector. Solution: Each contribution in (22.e. (22. (22.84) is positive semidefinite and so must be finite by itself.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd .Ax) dX Since T((f>. J 4 ) + A 4 . Mx') = E with a scale transformation x — x' = Ax.
89). L. T^{<t>x) = \4dT(</>). J. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x).e. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. 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. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. Landau [111]. This case is.89). (22. (22.570 CHAPTER 22. One can thus define a scalar field or wave function ty(x. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. MiillerKirsten [268]. of course. 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.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. H.89) the contribution (4 — d)T(4>). Ginzburg and L. 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. In the following we consider a related theory. a = a(T) = a T —T c . R. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives.92) The expression ^ ( x ) ! is known as the "order parameter. . (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. 22. W. (22. is temperature dependent and changes sign at the critical temperature Tc.t) such that  ^  2 describes the density of Cooper pairs. D. A) or P{4>) for d = 3. (e) For d > 5: In this case it is not possible to stabilize Eq. ft V . a > 0. Thus such a contribution could counterbalance T{4>. (22.g. H. We then have a situation of the potential like that described in the context **T.** e. Skyrme [253]. Tchrakian and H. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a socalled pure gauge theory with socalled instanton solution (which we shall encounter in a simpler context later). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i.90) and we would have to add to Eq. exemplified by soliton and sine—Gordon theories.
but for T < Tc it has a double well shape. Sec.e.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. The parameter 7 serves to put the minima of the double well at V = 0. ^(00) = a (22. i.e. (22. Clearly the length (see Eq.^ A n (this is called "minimal coupling of the electromagnetic field^ with vector potential A to the Cooper pair field ip in a nonrelativistic treatment) where g — 2e~ is the charge of the Cooper pair. 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. i.94) to check the dimension) 1 = I—a.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ .95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. setting S£.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf.2): For T > Tc the overall potential has a single well (minimum). Then the static energy density becomes (cf.22. assuming there is no superconducting material in this domain). is a measure of the distance from the surface of the superconductor to where . 22. from x < 0) means — as is familiar from electrodynamics — replacing V by V . 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) = 0.93) J' (22. Example 22.2) £(^. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material.(tp) = 0. (22. Variation as before.
i. V x M = j s . W. i.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B .e. H. as indicated in Fig. J. exhibits a completely different behaviour — in fact. MiillerKirsten [215].g. Classical Field Configurations superconductor (x>0) Fig. yJ—a/[3. 2mi Setting < / > \P\e vp B = ViA. (22. The density of the Cooper or super current is given by the typical expression of a current.9. the order parameter \ip\ attains (approximately) its asymptotic value.96) ipH. In the Maxwell equation See e. We can see this as follows. 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 ).9 Behaviour of ip and B in the superconductor.98) where M is the magnetization resulting from atomic currents. 22. The magnetic field on the other hand. .M. (22. i. p.e.572 no superconductor CHAPTER 22. (22.e. 22. h [^DV'V'DV].
div grad" and V • B = 0.L B = O. m xi A B . In the onedimensional case we have B{x) = B(0)ex/XL. (22. 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 . Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. Also dD/dt = 0 in a static situation like the one here.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. V x B = [i0V x M = /lois.M i. 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. These vortices therefore carry magnetic flux which is necessarily quantized . those of the socalled type I and those of type II.5 to 10~6 cm). 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. Approximating p by its equilibrium value. K i ™*. and V x (V x B) = / J 0 V x j s . (22. This is zero in the present case.22. this becomes A B = no V x j 5 . Using the relation "curl curl = grad div . i.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B.99) Inserting (22.e.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. Hence 0 = V x H = V x ( — B .e.
.9) ^e^ '>\<poo\. (22.37). If the energy E is to be finite — a basic requirement for the field configurations of interest to us — this integral must also be finite. (22. de Vega and F.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.e. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or.36).101) (p(r. We can think of (e. Abrikosov [2]. where exp(ix(0)) is an arbitrary phase factor. we must have that lim V ( H ) = 0. De Gennes [67]. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. See also P. J.g) twodimensional space as being bounded by a circle at r = oo. i. Finiteness of E requires \<t>\ = \<fioo\ on this circle. *A. i. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. A. 'See the remark in H. A. Schaposnik [69] after their equation (3. in other words. In the NielsenOlesen model defined in Example 22. in fact a periodic function of the polar angle with period 27r..574 CHAPTER 22.e.9). the space consists of two diametrically opposite points on the circle. 22. G. (22. In the two models of Example 22.102) In the onedimensional soliton case of Eqs. 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). Classical Field Configurations (see below). of two points on a line.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. The existence of these vortices has been confirmed experimentally. Thus the asymptotic field </>(r.
.10.e. W. Fig. (22.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. The two theories we considered above in Example 22. defines the orientation of this line element. The 1sphere is the unit circle in the space of all complex numbers. 22. topological group with the usual multiplication as group operation. Suppose <j>. 2n] remains continuous if the function assumes the same value at the two endpoints.1. In mathematical language S1 is therefore a compact. Sl := {zeC\\z\ = 1}. Finkelstein and C. Then a continuous function of (j> on [0. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. commutative).X ( 0 = O)].118)). 22. X is called the domain of / and y its range. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. i.102) is called the exponential map of S1 onto S1. Naturally an integer cannot change continuously. abelian (i.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. 0 < 0 < 2TT. The map A : S1 > S1 defined on S1 C R2 by (22.e.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".22.* We can think of this as a fixed line in space which is provided at each point with an infinitesimal direction element that is free to rotate in the plane perpendicular to the line (like the set of pendulums considered in Example 22.10 States of winding number 0. Misner [96]. Thus smooth deformations *See D. (22. We may define this as « = ^[X(0 = 2 T T ) .1).
i. We observe that since time evolution is continuous..e. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.e.105) the corrections falling off faster than 1/r. The different kinds of connectivity of these topological spaces (e.) remain the same under homeomorphic transformations and are therefore called topological invariants. the winding number must be a constant of the motion.. In other . since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism.g. Thus two homeomorphic spaces have equivalent topological structures. i.. i.i'ldx(0) 9A Thus for (22.104) so that d2x hJ Since A has direction e#. finite.576 CHAPTER 22. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite.105) is described as "pure gauge" since it is determined entirely by the phase x(#).104) to be finite we must demand that g . it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). (22.g.e. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). elements in the same class have the same topological characteristic. if it has one twist as above or a hole. The gauge field required in the asymptotic limit (22. we required that lim r—>oo w o. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. one hole. r^oo 1 dx(9) n r o0 (22. The winding numbers n are such topological invariants. no hole. For that reason n is called a "topological invariant?. we can write 1 d h J ee rde w At h r^oo . connectivity.
e.105) insure that B ^ O . going once around S^. However. the higher order contributions in (22. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. the vector potential A. i.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. into S i . It follows that the field Fij. However. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). is also zero at r = oo. cannot be deformed continuously into one with two loops (n = 2). for B / 0 . 9){^ and hence the functions (22. we have B = V x A = curlgradx = 0.106) J g j r d d g Thus the magnetic flux is quantized. A field configuration with one loop (or twist). we can consider transformations which describe continuous transformations of a configuration with fixed n. the group space of (the phase transformations of) U(l) continuously. (22. we can go n times around S^.108) map the spatial circle S^.e. An example is the mapping H(t.110) .9):=t(f>£\9) + (lt)(f)W(9) with te[0. with n = 1. the number of flux quanta 2irh/g being the winding number n. i. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . cannot be pure gauge everywhere. (22.109) A W(0) = e^ W.<b rd9—4^ = n. the gauge field.107) The fields <f>(r.22.l]. i. Ffiu. it is determined by a gradient (l/r)d/dd.e. Of course. (22.9 Introduction to Homotopy Classes words. 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". if A = V x . Therefore.
Dittrich and M. S%. i.9) = <fiW(9) and H(l. H. A brief introduction can be found in the very readable book of W. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). (fin (9) in the above example.e.9) is called a homotopy (emphasis on the second syllable).e. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. Classical Field Configurations which is continuous in the parameter t and is such that H(0. it is clear that this is also a homotopy.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. f T h e standard texts on homotopy theory are N. Steenrod [261] and S. n' ^ n. We can consider (22. the unit interval I = [0. Thomas [270]. Reuter [77]. — S\ > (22. i.e.111) as a map from the space described by 9. i. Hu [134].e. we have to show that the map satisfies the three properties of (a) identity. H.578 CHAPTER 22. like <pn (0). 8) = cfinl\9). (b) symmetry and (c) transitivity. We verify these in the following example. Since the passage of time is continuous. The map H(t. time cannot make a field jump from one homotopy class (see below) to another. e. But this type of combination is excluded sind H(t. Example 22. i. (22. n. Two curves which can be related to each other in this way. the phase factors of (f)(9).112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).9) = ^°\0).111) In principle an expression like (22. to the internal group space of £7(1). Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).113) . and the space of t.g. 9) is an equivalence relation. A superposition of classical configurations with different winding numbers does not minimize the energy.T. i. In fact.lQSl^S^ with H(0. (b) symmetry and (c) transitivity.0) = $\6). (22.8) itself must be a configuration with definite winding number.1].110) could be written down for configurations with different winding numbers. Solution: As stated.3: Homotopy an equivalence relation Verify that the homotopy map H(t. t The homotopy relates a representative function or curve (fin ' (0) to another function or curve 4>n' (9) by giving a precise meaning to the idea that the one can be deformed continuously into the other. H(1.e. these two maps (fin from S^. The relation (22.
a class being a set of equivalent maps.ffi(O. i.* 22. (b) Next we check that if 0<°) ~ <p^.e. 0). is isomorphic to) the set of integers Z (positive. U(l)). 0) = ^ (0) and 6(1. 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'.e. . Hence <f>(0^ ~ <p 2 ).9) H(0. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. \ M)\ H!(2t. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. i<t<l. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i.0) = </>(0) is continuous there. 0) = tf2(l. <f>W . as they are called. is such that H(0. 0) = 4>W (0).e) tf2(2il. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). below).0) for for 0<t<. (c) We can demonstrate the property of transitivity as follows.22. i? is a homotopy connecting </>(°) with <^>(2'. Returning to the maps \(x) : Si ^ S1^ (i. being the continuous map connecting </>(°) with <j>(1'>.e.9) = . 0). Moreover. we see that of the set of these maps some are equivalent and others are not.9) = <f>(0\6) and H(l. 0) = <£{2). 9) connects <pn (0) with itself continuously. which therefore demonstrates that the homotopy relationship is transitive. 9) = <f>W (0). so that H(t.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. Homotopy classes are sometimes also called Chern—Pontryagin classes. 9) = H(l. 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. We now define the function H ( f 0 ff .10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. then </>W ~ <t>{0)• We define an H(t. negative or zero). since and tf (1. Then H(t. <^(2). these groups are then called homotopy groups (cf. ' . 9) := H(l . . On the other hand H(0.t. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'.0). 9) = H(0.
p.e. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9. T < 9 < 2TT.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. 22.11 The trivial n = 0 map allowing shrinkage to a point. n=0: t=1: t=1/2: i.580 CHAPTER 22. S 2 the surface of a sphere in a threedimensional Euclidean space. Appropriate maps.g.11 the circle drawn with the continuous line is the circle S%. Sketch of idea demonstrating that nilS1] = Z. Fig. 22.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. How can we see this? We first sketch the idea. 52. In the diagrams^ of Fig. . Rajaraman [235]. 22. to make the result plausible. In order to really appreciate how topology comes into the picture here one should understand (e. T (22. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. and then introduce some of the related mathematics. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. which is to be mapped into another circle Si indicated by dashed lines. but TTI[S2] = 0: S 1 is a circle.) why 7i"i[S2] = 0.115) Varying t € [0.
The maps for these cases are illustrated in Fig.117) in ±.e.s. to the case of winding number zero. (22. of (22. and hence the winding number. One can prove that the expression (22.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. 22. Coleman [55]. In the cases n = 1.e. It is seen t h a t in these cases the continuous curve can be distorted into t h a t of case (a) only by cutting the dashed circle — thus these cases do not allow a continuous shrinkage to a point.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. For winding number n we have the mapping (i. 2 the maps are wound around the dashed circle. We make another observation at this point. 22.117).12 The n = 1 and n = 2 maps not allowing shrinkage to a point. i. the transformations^ (the result may be looked at as 8n of expression (22. d (22.116) In fact from this we obtain (see below) 2  n and this means r.s. is invariant under infinitesimal (continuous) deformations. .22. 0 < X(P) < 2vr.h. (22.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n.12.h.e.117) 2^ d6X(x)\1(x). i.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S.
8. 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. Classical Field Configurations Under this transformation n changes by 5n Now. there is a theorem which says thatH TTg(Sn) = 0 for q < n.e. Sec. 1 1 See N. In fact. . only one homotopy class. i. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). the cyclic group with two elements (since 50(3) is doubly connected).e. the identity class (the group space S2 is simply connected. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. 15. whereas 7ri(50(3)) = Z2 (i. an integer modulo 2) i. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point.582 CHAPTER 22. Steenrod [261]. Thus there is only the trivial map into a point.e. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided.
Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. finite Euclidean time) classical configurations (which are nontopological). 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.Chapter 23 P a t h Integrals and Instantons 23.e. in such a way that the factors appearing in the result are exhibited in a transparent way. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. In the following we employ the path integral in order to derive the same result for the ground state splitting (only that in this chapter!). Since we have not yet introduced periodic (i. and in particular the exponentially small level splittings of the double well potential. different from literature. 23. 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. We consider such a theory now with action S((p) depending on the real scalar field 583 . In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies.
note that here the mass of the equivalent classical particle is taken to be mo = 1. i. 23. (23. Gildener and A. The Hamiltonian of the simple harmonic oscillator (cf. Chapter 6) 1 2 .584 <)>{t). 2 2 4 ™ = ~. Patrascioiu [110] and Chapter 18.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. and ^2(0) : m* .1)) and hence E0 = mh/y/2.2. (23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. 23. Path Integrals and Instantons S(<f>) / dt \i>2 .1. "For comparison with literature.* /•tf CHAPTER 23. Fig.v{4>) <?VV m4 __ 1 .2) Fig. in particular E.1 The double well'potential.e. ! " ^ j = ^ 2 . n = 0.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf.
This tunneling affects the eigenvalues. We can develop a perturbation theory around either of the minima. The states of the system here must be even or odd under x — ±x. 23. Fig. e. must have definite parity as in a physical > situation.2 — and we can deduce from this the boundary conditions. this alone does not yield the level splitting.In mechanics of a particle of mass 1 with ~ position coordinate x(t). as we did in Chapter 18 — but as we saw. i.2 The wave functions of the lowest states.e. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering)._(0) = 0. </. and in the above 4> corresponds to x. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig.23. we have the Lagrangian L(x. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator .x) = x2 V(x).2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. 23.g. but there is tunneling between the two minima if the central hump is not infinitely high. W(0) = 0.
7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo. (23. Equation (23.7.0o.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.e.8) V(4>) + const. The transition amplitude for this.6) d*l dr 2 . dr We observe that Eq. In the semiclassical path integral procedure that we want to use here. (23.2 m 2 ( l m z (23. that given by Eq.9) .586 CHAPTER 23. (23. i.  .e. Euclidean time T = it. dr m+™ (23.3) for tf — oo.e. Then (0o. r . A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.£'(0) = 0.3). (23. i.5) and iS = dLE d dr[d(d(f)/dT)} i. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.<h. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.7) so that (23. SE.e. But quantum mechanically it can tunnel through the hump from one well to the other. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. For small h we can attempt a stationary phase method.U) = J'D{<f>}exp[iS(<l>)/h] (23. classically forbidden domains then become "classically" accessible. dLE = 0.4) J <bi =—(An rTf (ITLE = I JT.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).2. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55.tf\ .2 V'(<l>) = . is finite if we go to imaginary. i.
timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. 23.10) where TO is an integration constant.23. i. .e. we see that the equation is the same as the classical equation for a static soliton (i. Fig. 23.3.2 Instantons and AntiInstantons 587 In fact. 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. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). i. — (23. is depicted in Fig. in the present case (T — To) = m tanh m(r — TO). and hence kinetic energy gained is equal to potential energy lost. Its trajectory. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. this means the total energy of the particle in Euclidean motion is zero. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). the classical path.3 The instanton path.18).e. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".e. Thus we know the solution from there which is the kink solution (22. from fa — —rn/g.
14) But this is not yet enough.13). Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .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 22.11) For this quantity to be finite for TJ = —oo. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.' . The configuration (23. Using Eq. but with the same Euclidean action (23. we obtain lim f+ OO.7"/! ).8) we obtain T f SE dr[2V((j)c) + const. (23. The classical equation (23. (23. thus indicating the nonperturbative nature of the expression.r 0 ) ' (23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.13). 23.^ .4.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.TJ T fro. This antiinstanton configuration rises monotonically from right to left as shown in Fig.5). Tf = oo.Tj) oc exp 4 m3 m^2 (23.7) admits another configuration which is the negative of (23. the constant must be zero.10) is given by (23. (23. We also observe that the expression is singular for g2 —> 0.].10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .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 (23.13) into Eq.588 CHAPTER 23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26. Inserting the result (23. Path Integrals and Instantons The Euclidean action of the instanton solution (23.4).
#(r) = </>C(T .TO far apart and T'Q 3> TQ.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. 0c(r) = &(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. 23.5. as we can see e.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.r0).15b) since we require configurations from vacuum to vacuum. .g. from the sketch or from the variation of SE.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 .15b) is not exactly a solution.15b) has the shape shown in Fig.5. and an antiinstanton localized at (say) TQ. 23.TO) + ^ c ( r . (23. b u t if TQ. The sum (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. Clearly we are here interested in configurations of type (23. 23.15b) In the case of (23.4 The antiinstanton.g. (23.15a) we have a simple superposition. / > Looking at the instanton (23.15b) a solution of the classical equation? Is it a topological configuration — if so.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). in the case of (23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g. e.23.
at oo the variation is <<> = 0). In fact.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.15a) is not exactly a classical solution. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. Finally.2 m ( T o . 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 . 5^ In order to answer the second question we observe that (23.CHAPTER 23. we see that in the domain [—oo. Path Integrals and Instantons '.15a) can be argued to be zero. #(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 . T\] with LECIT = 0.e x p { . vanishes as T\ —> ±oo (of course.15a) is a configuration which starts from 4> = —i^/g and ends there.5 Superposition of an instanton and a widely separated antiinstanton.16) . g cosh m(r — TQ) (23. since (23. 23. Thus it belongs to the socalled vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero.T 0 ) } ] . of course. Ignoring exponentially small contributions the Euclidean action of the configuration (23. it also does not exactly stationarize the Euclidean action.
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).
so that (with T *T) 1/2 •Dz N(T) (23..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. (23. [44]. Thus in our case one of the kernels is zero and we have to decide what we do when.exp • /_:T JV(T')" dr' = e x p [ . i. 192.67) i=l v ) = lim exp V In ( 1 + ' L K(T.64) D (23. Thus the factor a inserted above has to be taken as 1/2. ^This is the method explained by B. We observe that if we discretize Eq..23. H. . (23.T')dT' (23. 2! . 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.64). X(TI.j.66) •Dr) .7 Y. Cameron and W.TI) K(T2.T1)dT1 + J dn J dr2. the kernel K(r. Felsager [91].N{T) Finally we add the following remarks concerning the discretization (23.T2) I exp I K(n. are summed over all values from 1 to n. = 1 52 K(n. Thus D„ = l + 5 ^ K ( r i . r ' ) is to be evaluated at r = T ' .e. n )d? (23. Martin [43].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.n) o K(TJ.. In the limit S —* 0 . the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . 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. p. whereas the determinant arises only once in the original determinant Dn.. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.63b). (23. . r i ) + .n) z=l x (23.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .a { l n i V ( r ) . T. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. as in Eq.n) K(Tj. where a f 1 See R.TI) 0 K(T2.64) as above and consider the determinant of the expression. the value of the determinant is half (or arithmetic mean value) along the diagonal.63a).n) i=l N lim exp £*(T.Ti) i=l : exp \ £ K(T.68) D above. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). we obtain exactly the nonvanishing terms on the right of Eq.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i.64) Inserting the explicit expression for the kernel given in Eq.lnJV(T)}] N(T) N(T) (23. 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.
In = dzi dz2 • • • dzn e x p . With partial integration (in the second step) we obtain (since (cf. (23.73) . (23. f(T) = N{T) J^ dr'^^z(r') (23.59)) + f(T)] = — f dz(T) f°° dae^W+f™. represents a constant translation. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). N(T') .— .59) the endpoint conditions »)(±T) = 0 and evaluate the integral. In every integration / dzi the contribution €7. Briefly.^ . (23.70) and this now requires also integration over the endpoint coordinate z(T).(T) + /_ dr — .57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . Path Integrals and Instantons Example 23. (21.0. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23.61) 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).l + £7t) «i (23. . rn+i = T.Z i .72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.N(T) .3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. I2 and using induction one obtains (cf. Solution: From Eq.608 CHAPTER 23.J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ .L dT{ \ (i(r) exp T>z 1 N(T) + 1>n exp N(T)\2 ' N(T) J a2 N2(T) 2 JV 2 (r) We perform the functional integration as in Chapter 21. Evaluating Ii. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. Eq. and hence may be ignored. n J=.e—*0 In.( .2o) 2 rc^O.Z ( T ). T + N(T) / T' = T JT dr' N(T') . t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T.22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . (23.71) Hence (with h = 1) l = j . (23. Eq.
3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.11). This result verifies Eq.^ 23. Faddeev and V. D.e.10) and (21." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. i. / d£o <9r0 5[F(TQ. £n = ~2 dTT](T)7]n(l (23.74) where we reinserted h.. ^E. j / dTT]n(T)r]m(l r . a = const. Gildener and A. Eq. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. 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. / — / / We started off with V(T) = ^CnVniT). .76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. (23. Ignoring the remaining metric factor with the reasoning explained > between Eqs. (21. 1/2 T W 2 (r)J (23.59) and agrees with that in the literature.£Q) — aTo]dTod£o.12). (3.75) 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.4.4 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. 11 L. Patrascioiu [110].23. N. (23. J co § dr Recall that / f ^ doce^ = yfVfp. Popov [89].
demand validity of) the constraint £o = 0. / dTm{r)n{T). (23. . We establish the function ^(To>£o) with the following reasoning. In the integral of Eq. e.g.76) we set 4>c{r) = ^CnVnir).£Q) — aro] contained in Eq. in the direction of the zero mode.78) <t>ij) = Y.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T).79) (23. i.75) requires. In analogy to (23.J drm(r)Mr + T0) (23.81) 2 / dr»7o(r)0c(r) = .77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. Path Integrals and Instantons _^ ^ w .£o)/a. We consider the following integral / ( r 0 ) := . We now impose (i.^O) with dependence on ro which the replacement (23.83) **The collective coordinate in the present context is also called "instanton time".610 this means that CHAPTER 23.75) enforces the collective coordinate ro to be given by F(ro. The delta function 8[F(TQ. at r = ro.81) the rdependence is integrated out and hence the integral does not provide the function F(TO.e. (23. (23. 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. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. i. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . (23.^n{r). where L = in + cn. (23. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). The number £o is the component of the fluctuation rj(r).e. i /*oo (23. (23. Thus. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). Then cn = ~2 / dr(pc(T)rjn(T).e.
82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T). (23.(23.84) (23.23.y/SET0 = 0. 158.85) PJ W i t h Eq. (23.^) .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).89) Here the factor App is called the FaddeevPopov determinant.J dTrj0(T)[(t>c(T + TO) + ??(T)] i r /• I a rfT77o(T)—0C(T + TO) + sr — / dTrj(T)r)o(i # C ( T ) . jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0. We evaluate this quantity as follows in the leading approximation. (23.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o..75) 1 = AFP f dT06[{F{rQ. .88) ro0 Then in Eq.£O) OTQ = = £ " .Zo)y/SETQ We now define A FP '• =0 • (23. .91) " F o r more discussion of this point and related aspects see also B. Felsager [91]. so t h a t with F(TQ. We have AFP = /F(T0. p./ drri0(T)<pc(T) + PJ (23. 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) .y/S^ro] (23.87) l/M) (23.<t>c{r)} .86) = Jdrrjoir^Mr F(TO.
612 CHAPTER 23.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . 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. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. (23. Inserting (23.e.93) where AT = TT — 00. =~ (23. tangential to (J)C(T)) leaves this and hence the square of this quantity.90). i.93).77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. if the coefficient of 770(70) vanishes. if ^ + d£o = 0. 0 d r (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. 7 Thus there is no fluctuation along the direction of the zero mode in the sum. in d(j)c(To) + dr/fa). The factor p is a normalization factor which we > retain here only for ease of comparison with literature and may otherwise simply be put equal to one. i. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). P the relative sign being of no significance.0AFP8 4o H P . iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ).e. /(TO) ^/SE~T0} V~SET0 c d£.e. i. ~ There is a quick way to obtain the result (23. 770(7"). the "static energy" invariant. using the relation (23. P J AT (23./ dT0AFP.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.91) into Eq.e. . Consider. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.
(23.60a) in not involving the zero mode contributions.4 Field 23.23. We determine I'Q by observing t h a t this differs from IQ of Eq. (23. (23. (23.93). so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23.52a) and replace in this J d£o by the expression (23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo. (23. 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.A T A F P i " o e x p T—>oo p where I'o n^O. 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. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.4.96) does not yet allow a comparison with (23. Since w\ is for finite T given by WQ of Eq.97) is a constant still to be evaluated.43)).35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23. We observe t h a t the result (23.55b).60a) for I0 and Eq. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.96) lim .98) .55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.
. ^See E. Patrascioiu [110]. 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. the approximation is then known as the "dilute gas approximation".32).TiMiy+'Mr «>.e. Although we choose p2 = 1. we drag the factor p2 along for ease of comparison with literature.a n t i .55b)) w0 = v0e VQ 23. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23.614 CHAPTER 23.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .5 I n s t a n t o n . This means we are concerned with configurations like* 2n+l d(r . Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 .101) 2T > oo (cf. the result becomes h lim 2 m r ^ ^ 1 e .100) in such a way that it exhibits the crucial factors of which it is composed. In view of their normalization of the normal modes. *For the validity of the calculation the instantons or kinks have to be far apart. 0c M. (3.r). Gildener and A. they have to introduce the inverse of that factor on the far right of their expression (3.2 m T e x p 2 irgh ' T»oo (23.i n s t a n t o n contributions As already mentioned.* Finally we rewrite the expression (23.4. (23. their Eq.102) *For comparison with the parameters of E. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.4>C exp h (23. ~ rKo')e{Tx .20). Gildener and A. (23.100) We observe that the dependence on the normalization constant p cancels out. i.
4>i. . (23.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.Tl}{4>l.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .Tf\(p2.r / 3 .4>l)eSE{4>l. r 0 ( 2 ) = 0. Consider (with T 0 = T. the total Feynman amplitude is given by (^/. T3 = T) d(f>l((pf. Proceeding as before we now consider first in analogy with Eq. here (f>2. (23.<Pi)2 + • • • . (3) (23. 2 . T2 = T/3.36) (/.fo) = Id2 St SE(<Pf.r/l&.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.H)2 + . . are varied.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left). .2 T / 3 .<t>2)eSE{<t>2.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.r 0 ( 1 ) = . The factor ( . this means we have to consider the amplitude (23.23. Since these endpoint configurations.<t>ff + h .T2}((t)2. 7 i = .4>I) ld2SE SE(c/>i. (23. (n = 0 implies the single instanton solution). we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f.T2\(f)l. 23. Taking these configurations into account.106b) SE{<t>l.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.l ) i + 1 in the sum of Eq. r0(3) = 2T/3.Tl\(j)i.4 Field Fluctuations 615 for n = 1. ( 2 3 .<j>i) .104) Considering the case n — 1 of a kink with one additional kinkantikink pair.(l>i) (23. 1 0 6 a ) and SE{(J>I.6.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .
e.4>i) / d(f>2 e x p {<h .e. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.e.3 9 L 3m 2 and 9SE((/)) ± . the dilute gas approximation. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. each of the three intermediate kernels contained in Eq.e. (23.105) can in other respects be treated in analogy with the single instanton amplitude. the .106b) is to be substituted into the second relation of (23.4>i) { 2 " 2d SE . ±m/g (23. i. in the case of l2n+i therefore ry2n. The allowed positions of the individual kinks have to be such that their order is maintained. d2SE{4>) d(j)2 2m. i. rn ^2 J2H 9 1m' ±2gcj).107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.108) where the last expression is obtained as follows. each of which implies a multiplicative factor 7.2J.(01 . (23. (23.4>f) d4>\ 4>s72 (23.12) we have JM~°°) i. Path Integrals and Instantons where the first expression in Eq. On the assumption that the kinks of the instantons are widely separated.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations.106a). It has to be remembered.. i. dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .2d2SE ~c 4>i d(f)i e x p SE^f. that each applies only to a fraction of the overall Euclidean time interval of r = 2T. With our remark after Eq. of course.616 CHAPTER 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.105) t h a t each of the three amplitudes contributes a factor e x p [ . Patrascioiu JT/3 ° ir< >2T/3 1 dr.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23.(*) o ax = 2 T / 3 . (3) ° JrW: 2T/3 (2T)3 .6 the collective (i) coordinate TQ has to be to the right of TQ . In the case of (<^f.93) has been applied. then sign reversals TQ i ) . and in the case depicted in Fig.T/0j.101): (^/. a 2 0.(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23.a 3 [110].110) Finally we have to recall from Eq. we now have to integrate over three collective coordinates. Am3 2m J rW ( r ( 0 and this followed by setting aj.r.4 Field Fluctuations 617 replacement (23.)(3) 1 o A T 3 Ao lim . T ] with ] T 3 = 1 A » T = A T = 2T. Gildener and A.^ . i.23. (23.7 3 ^ — . 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) .V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.m A . 23.Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.r/^.2 T / 3 this product becomes that given by E.e.112) We now compare this result with expression (23.111) 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 .
W. 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. by either replacing A T by 2AT or doubling A*.113) is very complicated.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. MiillerKirsten [30]. (23. Gildener and A. Bose and H. 23. The difference between levels. MiillerKirsten [163]. l CHAPTER 23.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs. MiillerKirsten and A. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton). we obtain AE = 2A* and agreement with the result (18. 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. K.W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). J. AE. W. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — . J. J. explicit quantization).182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92. here as 1) is also taken into account. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . W. has been obtained without explicit use of canonical commutation relations (i. GP for those of E.' 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.^r .182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. (18. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S.618 and hence AE A. . whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state. even without explicit specification of the prefactor.113) for qo = 1: A1E „ . We observe one more aspect of our calculation.Q. .113). ^See P. H. Achuthan. To help the comparison we note that — with indices indicating the authors. Patrascioiu [110]. J. (23.With this correspondence A\E of Eq. Wiedemann [4]. Liang and H. Finally we realize that the computation of (23.e.
i. 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. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling). Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 .a 2 ) 2 + c^T2. however. ™=£('^° < 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). in a somewhat different formulation.2) . We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues.Chapter 24 P a t h Integrals and Bounces on a Line 24. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp. 619 (24.e. and their interpretation and uses. Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance.1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1.
J 4.a2)4> + 2cTz 2 0.1 The asymmetric double well.e. T = 2\(<f>2 . = 4(ACT2 cT 2 ).620 CHAPTER 24.A a 2 ) . We can destroy the reflection symmetry of V((f)) (i. > / V[<H 'true'vacuum 'false'vacuum Fig. but implies instability for T < Tc. at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) . We see that <fi = 0 minimizes VT(</>) for temperatures T > TC. Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool.2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. its symmetry under . a spontaneous symmetry breaking leads to the double well (i.e. A2 _ a — \„2\ u ) ~r u/\y> 2 (24. 24.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 . Path Integrals and Bounces on R1 Here dVT{4>) vanishes for .
tf(*) =^ 1 . the higher minimum is called a "false vacuum". and (c) determine the lowering of the energy to the true quantum mechanical ground state (i.Y.) .24. Cornalba. V(<j> = m/g) = 2e.1 with (24.g. Pi [126].t In the following we consider first (after some recapitulations) a simple model of a false vacuum. for instance.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . The Lagrangian density of the theory is given by C = dp*d»*U(*). We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory. T False vacua and bounces occur. M.g. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE). A. On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. False vacua play an important role in models of the inflationary universe* and elsewhere. 24. We calculate the classical configuration — which in this case is called a "bounce" — which corresponds to the soliton or instanton in our previous chapters (but is not topological) and we use this in order to calculate the lifetime of the unstable state (or equivalently the imaginary part of the energy). Guth and S. a linear term as in w^^'A1^'*0V(<l> = m/g) = 0.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e. see e.7) *See e. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions. H. and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only. Since the true vacuum minimizes the potential. . which (b) are saddle point configurations of E (or SE). perturbationtheory vacuum). (24. .g. S. the quantum mechanical vacuum).5) We see that one minimum of V now lies higher than the other. <44 2) ' This is now an asymmetric double well potential as in Fig. the vacuum being understood as the classical. Costa and J.e. L. (24. Penedones [59] and references cited there. also in contexts of string theory.^J >o.
small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention). dx2 + u"(<j>) 5(x y). Path Integrals and Bounces on R1 Static fields given by d$(x. which is an important point as we shall see.7) yields the following equation called stability or small fluctuation equation. (24. (24.10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24. (24.8). we see that (24.14) .622 CHAPTER 24. i.9) 9 for the doublewell potential given above. t) = <f>c(x) + Y^ iPk{x)ekt (24.12) Since <fic(x) solves Eq.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x).8) This equation possesses the kink solution <t>c{x) m tanhm(a. (24. The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c.e. We observe that this kink solution is odd in (x — XQ). — XQ) (24. Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24.t) dt 0 are written (f>(x). i.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.e.
15) cosh mx From our study of the solutions of this equation in Sec.2 The inverted potential.e. In the case of the double well potential (24. o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = .1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite. 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24. the zero mode). d \m . configuration (for our </>4theory).8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero. i. .24. eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue. i. We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. 3m 2 and w\ > Am2. 24. underneath the particle's velocity.17) . the other a positive quantity) and a continuum which are given respectively by w\ — 0. 22. . i. kink.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24.oo \ x== + CX) Fig.16) V[<H x= oo 0 d2 .e. thus the zero mode here is given by d .e. (24.
2.e. with Euclidean time. (24. by Q = \(°° dxf^eoid1^) = l[M°°) ~ M°°)] = 1(2422) . i. Integrating the classical equation (24.e. quantum mechanics. and — clearly — we can resort to a classical interpretation of this function. i.18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U. 4 TTT.21) The charge is defined as the space integral of the time component of the current. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function. The associated zero mode represents the velocity of the pseudoparticle as indicated in Fig. In this case the usual Newton equation reappears with an opposite sign of the potential. i. which has the solution 4>c(x). the associated zero mode. so that its derivative representing the pseudoparticle's velocity is odd. The kink solution can now be looked at as the classical solution of a onedimensional field theory. 24. We see that this pseudoparticle starts from rest at one peak of — V at x = — oo and in losing potential energy gains just enough kinetic energy to reach the second peak at x = +00.20) The number 1 on the right is the socalled topological charge which is obtained as follows. 4>"{x) = U'(4>). Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton.e. where it is then again at rest. F (24.8). We define the current k» := e^dvfc. In the pseudoparticle consideration the coordinate x plays the role of time. (24.e.3 (24. Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function.19) E((f>c) > / J00 dxV2U(f)'c(x x0) = —x &9 1. i.624 CHAPTER 24. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. 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. Below we shall see that the configuration described as a bounce is given by an even function of x. The classical configuration is then called a pseudoparticle configuration or instanton. Then the energy satisfies f°° .
A W + £/(A0C (24. 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. The theory does. J. 2. T. It is important here to realize that the nonzero value of Q res