INTRODUCTION TO OUHNTUM MECHANICS
•
Schrddinger Equation and Path Integral
INTRODUCTION TO QUANTUM MECHANICS
Schrodinger Equation and Path Integral
This page is intentionally left blank
*»«*?:#•
m
m
't
Schrodinger Equation and Path Integn
University of Kaiserslautern, Germany
\(P World Scientific
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • HONG KONG • T A I P E I • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
INTRODUCTION TO QUANTUM MECHANICS: Schrodinger Equation and Path Integral Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9812566910 ISBN 9812566929 (pbk)
Printed in Singapore by World Scientific Printers (S) Pte Ltd
Contents
Preface 1 Introduction 1.1 Origin and Discovery of Quantum Mechanics 1.2 Contradicting Discretization: Uncertainties 1.3 ParticleWave Dualism 1.4 ParticleWave Dualism and Uncertainties 1.4.1 Further thought experiments 1.5 Bohr's Complementarity Principle 1.6 Further Examples xv 1 1 7 12 14 17 19 20 23 23 23 29 29 31 34 38 41 41 41 49 53 54 55 59 59 60
2 Hamiltonian Mechanics 2.1 Introductory Remarks 2.2 The Hamilton Formalism 2.3 Liouville Equation, Probabilities 2.3.1 Single particle consideration 2.3.2 Ensemble consideration 2.4 Expectation Values of Observables 2.5 Extension beyond Classical Mechanics 3 Mathematical Foundations of Quantum Mechanics 3.1 Introductory Remarks 3.2 Hilbert Spaces 3.3 Operators in Hilbert Space 3.4 Linear Functionals and Distributions 3.4.1 Interpretation of distributions in physics 3.4.2 Properties of functionals and the delta distribution . . Dirac's Ket and BraFormalism 4.1 Introductory Remarks 4.2 Ket and Bra States v
4
VI
4.3 4.4 4.5 5
Linear Operators, Hermitian Operators Observables Representation Spaces and Basis Vectors
62 68 71 73 73 73 77 78 80 83 83 84 90 91 98 98 105 105 105 Ill 113 118 118 123 129 129 130 133 139 143 146 147 152 155
Schrodinger Equation and Liouville Equation 5.1 Introductory Remarks 5.2 The Density Matrix 5.3 The Probability Density p(x, t) 5.4 Schrodinger Equation and Liouville Equation 5.4.1 Evaluation of the density matrix Quantum Mechanics of the Harmonic Oscillator 6.1 Introductory Remarks 6.2 The OneDimensional Linear Oscillator 6.3 The Energy Representation of the Oscillator 6.4 The Configuration Space Representation 6.5 The Harmonic Oscillator Equation 6.5.1 Derivation of the generating function Green's Functions 7.1 Introductory Remarks 7.2 Timedependent and Timeindependent Cases 7.3 The Green's Function of a Free Particle 7.4 Green's Function of the Harmonic Oscillator 7.5 The Inverted Harmonic Oscillator 7.5.1 Wave packets 7.5.2 A particle's sojourn time T at the maximum TimeIndependent Perturbation Theory 8.1 Introductory Remarks 8.2 Asymptotic Series versus Convergent Series 8.2.1 The error function and Stokes discontinuities 8.2.2 Stokes discontinuities of oscillator functions 8.3 Asymptotic Series from Differential Equations 8.4 Formal Definition of Asymptotic Expansions 8.5 RayleighSchrodinger Perturbation Theory 8.6 Degenerate Perturbation Theory 8.7 DingleMiiller Perturbation Method
6
7
8
Vll
9
The 9.1 9.2 9.3 9.4
Density Matrix and Polarization Phenomena Introductory Remarks Reconsideration of Electrodynamics Schrodinger and Heisenberg Pictures The Liouville Equation
161 161 161 166 167 169 169 169 170 173 176 178 184 185 189 191 195 199 199 199 205 206 210 213 215 215 . . 219 223 227 234 237 239 243 249 249 250 251 254
10 Quantum Theory: The General Formalism 10.1 Introductory Remarks 10.2 States and Observables 10.2.1 Uncertainty relation for observables A, B 10.3 OneDimensional Systems 10.3.1 The translation operator U(a) 10.4 Equations of Motion 10.5 States of Finite Lifetime 10.6 The Interaction Picture 10.7 TimeDependent Perturbation Theory 10.8 Transitions into the Continuum 10.9 General TimeDependent Method 11 The Coulomb Interaction 11.1 Introductory Remarks 11.2 Separation of Variables, Angular Momentum 11.2.1 Separation of variables 11.3 Representation of Rotation Group 11.4 Angular Momentum:Angular Representation 11.5 Radial Equation for Hydrogenlike Atoms 11.6 Discrete Spectrum of the Coulomb Potential 11.6.1 The eigenvalues 11.6.2 Laguerre polynomials: Various definitions in use! 11.6.3 The eigenfunctions 11.6.4 Hydrogenlike atoms in parabolic coordinates 11.7 Continuous Spectrum of Coulomb Potential 11.7.1 The Rutherford formula 11.8 Scattering of a Wave Packet 11.9 Scattering Phase and Partial Waves 12 Quantum Mechanical Tunneling 12.1 Introductory Remarks 12.2 Continuity Equation and Conditions 12.3 The ShortRange Delta Potential 12.4 Scattering from a Potential Well
vm 12.5 Degenerate Potentials and Tunneling 13 Linear Potentials 13.1 Introductory Remarks 13.2 The Freely Falling Particle: Quantization 13.2.1 Superposition of de Broglie waves 13.2.2 Probability distribution at large times 13.3 Stationary States 13.4 The Saddle Point or Stationary Phase Method 14 Classical Limit and W K B Method 14.1 Introductory Remarks 14.2 Classical Limit and Hydrodynamics Analogy 14.3 The WKB Method 14.3.1 The approximate WKB solutions 14.3.2 Turning points and matching of WKB solutions . . . . 14.3.3 Linear approximation and matching 14.4 BohrSommerfeldWilson Quantization 14.5 Further Examples 15 Power Potentials 15.1 Introductory Remarks 15.2 The Power Potential 15.3 The ThreeDimensional Wave Function 16 Screened Coulomb Potentials 16.1 Introductory Remarks 16.2 Regge Trajectories 16.3 The SMatrix 16.4 The Energy Expansion 16.5 The SommerfeldWatson Transform 16.6 Concluding Remarks 17 Periodic Potentials 17.1 Introductory Remarks 17.2 Cosine Potential: Weak Coupling Solutions 17.2.1 The Floquet exponent 17.2.2 Four types of periodic solutions 17.3 Cosine Potential: Strong Coupling Solutions 17.3.1 Preliminary remarks 17.3.2 The solutions 259 265 265 265 266 270 272 276 281 281 282 286 286 290 293 297 301 307 307 308 315 319 319 322 328 329 330 336 339 339 341 341 350 353 353 354
ix
17.3.3 The eigenvalues 17.3.4 The level splitting 17.4 Elliptic and Ellipsoidal Potentials 17.4.1 Introduction 17.4.2 Solutions and eigenvalues 17.4.3 The level splitting 17.4.4 Reduction to Mathieu functions 17.5 Concluding Remarks 18 Anharmonic Oscillator Potentials 18.1 Introductory Remarks 18.2 The Inverted Double Well Potential 18.2.1 Defining the problem 18.2.2 Three pairs of solutions 18.2.3 Matching of solutions 18.2.4 Boundary conditions at the origin 18.2.5 Boundary conditions at infinity 18.2.6 The complex eigenvalues 18.3 The Double Well Potential 18.3.1 Defining the problem 18.3.2 Three pairs of solutions 18.3.3 Matching of solutions 18.3.4 Boundary conditions at the minima 18.3.5 Boundary conditions at the origin 18.3.6 Eigenvalues and level splitting 18.3.7 General Remarks 19 Singular Potentials 19.1 Introductory Remarks 19.2 The Potential 1/r 4 — Case of Small h2 19.2.1 Preliminary considerations 19.2.2 Small h solutions in terms of Bessel functions . . . . 19.2.3 Small h solutions in terms of hyperbolic functions . . 19.2.4 Notation and properties of solutions 19.2.5 Derivation of the Smatrix 19.2.6 Evaluation of the Smatrix 19.2.7 Calculation of the absorptivity 19.3 The Potential 1/r 4 — Case of Large h2 19.3.1 Preliminary remarks 19.3.2 The Floquet exponent for large h2 19.3.3 Construction of largeh 2 solutions
361 363 371 371 373 375 377 378 379 379 382 382 384 391 393 396 402 405 405 407 412 414 417 424 427 435 435 436 436 438 441 442 446 455 458 460 460 461 464
X
19.3.4 The connection formulas 19.3.5 Derivation of the Smatrix 19.4 Concluding Remarks 20 Large Order Behaviour of Perturbation Expansions 20.1 Introductory Remarks 20.2 Cosine Potential: Large Order Behaviour 20.3 Cosine Potential: Complex Eigenvalues 20.3.1 The decaying ground state 20.3.2 Decaying excited states 20.3.3 Relating the level splitting to imaginary E 20.3.4 Recalculation of large order behaviour 20.4 Cosine Potential: A Different Calculation 20.5 Anharmonic Oscillators 20.5.1 The inverted double well 20.5.2 The double well 20.6 General Remarks 21 The 21.1 21.2 21.3 Path Integral Formalism Introductory Remarks Path Integrals and Green's Functions The Green's Function for Potential V=0 21.3.1 Configuration space representation 21.3.2 Momentum space represenation Including V in First Order Perturbation Rederivation of the Rutherford Formula Path Integrals in Dirac's Notation Canonical Quantization from Path Integrals
466 468 470 471 471 476 479 479 486 493 494 495 500 500 501 502 503 503 504 510 510 513 514 518 524 533 537 537 539 544 549 554 557 564 570 574 579
21.4 21.5 21.6 21.7
22 Classical Field Configurations 22.1 Introductory Remarks 22.2 The Constant Classical Field 22.3 Soliton Theories in One Spatial Dimension 22.4 Stability of Classical Configurations 22.5 Bogomol'nyi Equations and Bounds 22.6 The Small Fluctuation Equation 22.7 Existence of FiniteEnergy Solutions 22.8 GinzburgLandau Vortices 22.9 Introduction to Homotopy Classes 22.10The Fundamental Homotopy Group
XI
23 Path Integrals and Instantons 23.1 Introductory Remarks 23.2 Instantons and AntiInstantons 23.3 The Level Difference 23.4 Field Fluctuations 23.4.1 The fluctuation equation 23.4.2 Evaluation of the functional integral 23.4.3 The FaddeevPopov constraint insertion 23.4.4 The single instanton contribution 23.4.5 Instantonantiinstanton contributions 23.5 Concluding Remarks
583 583 583 592 596 596 603 609 613 614 618
24 Path Integrals and Bounces on a Line 619 24.1 Introductory Remarks 619 24.2 The Bounce in a Simple Example 625 24.3 The Inverted Double Well: The Bounce and Complex Energy 631 24.3.1 The bounce solution 631 24.3.2 The single bounce contribution 635 24.3.3 Evaluation of the single bounce kernel 637 24.3.4 Sum over an infinite number of bounces 641 24.3.5 Comments 644 24.4 Inverted Double Well: Constant Solutions 644 24.5 The Cubic Potential and its Complex Energy 645 25 Periodic Classical Configurations 25.1 Introductory Remarks 25.2 The Double Well Theory on a Circle 25.2.1 Periodic configurations 25.2.2 The fluctuation equation 25.2.3 The limit of infinite period 25.3 The Inverted Double Well on a Circle 25.3.1 Periodic configurations 25.3.2 The fluctuation equation 25.3.3 The limit of infinite period 25.4 The SineGordon Theory on a Circle 25.4.1 Periodic configurations 25.4.2 The fluctuation equation 25.5 Conclusions 649 649 650 650 659 663 664 664 667 669 670 670 671 673
Xll
26 Path Integrals and Periodic Classical Configurations 675 26.1 Introductory Remarks 675 26.2 The Double Well and Periodic Instantons 676 26.2.1 Periodic configurations and the double well 676 26.2.2 Transition amplitude and Feynman kernel 678 26.2.3 Fluctuations about the periodic instanton 679 26.2.4 The single periodic instanton contribution 684 26.2.5 Sum over instantonantiinstanton pairs 688 26.3 The Cosine Potential and Periodic Instantons 690 26.3.1 Periodic configurations and the cosine potential . . . . 690 26.3.2 Transition amplitude and Feynman kernel 693 26.3.3 The fluctuation equation and its eigenmodes 694 26.3.4 The single periodic instanton contribution 696 26.3.5 Sum over instantonantiinstanton pairs 700 26.4 The Inverted Double Well and Periodic Instantons 702 26.4.1 Periodic configurations and the inverted double well . 702 26.4.2 Transition amplitude and Feynman kernel 705 26.4.3 The fluctuation equation and its eigenmodes 706 26.4.4 The single periodic bounce contribution 708 26.4.5 Summing over the infinite number of bounces 710 26.5 Concluding Remarks 714 27 Quantization of Systems with Constraints 27.1 Introductory Remarks 27.2 Constraints: How they arise 27.2.1 Singular Lagrangians 27.3 The Hamiltonian of Singular Systems 27.4 Persistence of Constraints in Course of Time 27.5 Constraints as Generators of a Gauge Group 27.6 Gauge Fixing and Dirac Quantization 27.7 The Formalism of Dirac Quantization 27.7.1 Poisson and Dirac brackets in field theory 27.8 Dirac Quantization of Free Electrodynamics 27.9 FaddeevJackiw Canonical Quantization 27.9.1 The method of Faddeev and Jackiw 28 The 28.1 28.2 28.3 QuantumClassical Crossover as Phase Transition Introductory Remarks Relating Period to Temperature Crossover in Previous Cases 28.3.1 The double well and phase transitions 715 715 717 720 723 726 727 734 736 740 740 745 745 753 753 755 756 757
Xlll
28.3.2 The cosine potential and phase transitions 28.4 Crossover in a Simple Spin Model 28.5 Concluding Remarks 29 Summarizing Remarks A Properties of Jacobian Elliptic Functions Bibliography Index
759 760 771 773 775 779 797
This page is intentionally left blank
Preface
With the discovery of quantization by Planck in 1900, quantum mechanics is now more than a hundred years old. However, a proper understanding of the phenomenon was gained only later in 1925 with the fundamental Heisenberg commutation relation or phase space algebra and the associated uncertainty principle. The resulting Schrodinger equation has ever since been the theoretical basis of atomic physics. The alternative formulation by Feynman in terms of path integrals appeared two to three decades later. Although the two approaches are basically equivalent, the Schrodinger equation has found much wider usefulness, particularly in applications, presumably, in view of its simpler mathematics. However, the realization that solutions of classical equations, notably in field theory, play an important role in our understanding of a large number of physical phenomena, intensified the interest in Feynman's formulation of quantum mechanics, so that today this method must be considered of equal basic significance. Thus there are two basic approaches to the solution of a quantum mechanical problem, and an understanding of both and their usefulness in respective domains calls for their application to exemplary problems and their comparison. This is our aim here on an introductory level. Throughout the development of theoretical physics two types of forces played an exceptional role: That of the restoring force of simple harmonic motion proportional to the displacement, and that in the Kepler problem proportional to the inverse square of the distance, i.e. Newton's gravitational force like that of the Coulomb potential. In the early development of quantum mechanics again oscillators appeared (though not really those of harmonic type) in Planck's quantization and the Coulomb potential in the Bohr model of the hydrogen atom. Again after the full and proper formulation of quantum mechanics with Heisenberg's phase space algebra and Born's wave function interpretation the oscillator and the Coulomb potentials provided the dominant and fully solvable models with a large number of at least approximate applications. To this day these two cases of interaction with nonresonant spectra feature as the standard and most important xv
To what extent the two methods are actually equivalent. Thus this first part . The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. Excluding spin. with degenerate vacua) were first and more easily derived from the Schrodinger equation. we recapitulate the origin of quantum mechanics. With the growing importance of models in statistical mechanics and in field theory. whereas for the calculation of discrete eigenvalues the Schrodinger equation. 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. but also about complex energies that he encounters in a parallel course on nuclear physics. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. basic postulates and standard applications. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. 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.e. screened Coulomb potentials and maybe singular potentials. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. Chapters 1 to 14. These basic cases will be dealt with in detail by both methods in this text. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. its mathematical foundations. in Feynman diagrams in quantum electrodynamics). the aforementioned exactly solvable cases. but arise also in recently studied models of large spins. Our approach to quantum mechanics is through a passage from the Poisson algebra of classical Hamiltonian mechanics to the canonical commutator algebra of quantum mechanics which permits the introduction of Heisenberg and Schrodinger pictures already on the classical level with the help of canonical transformations. With various techniques and deeper studies.g. again point the way: For scattering problems the path integral seems particularly convenient.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. numerous problems could. be treated to a considerable degree of satisfaction perturbatively. 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. like periodic potentials. in fact. However. even though the expansions are mostly asymptotic. In the first part. has not always been understood well.
and the elliptic or Lame potential — here introduced earlier as a generaliza .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 potentials with degenerate minima will be seen to reappear throughout the text. In the final chapter of this part we discuss the large order behaviour of the perturbation expansion with particular reference to the cosine and double well potentials. Chapters 15 to 20. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). The most prominent examples here are the double well potential and its inverted form. we deal mostly with applications depending on perturbation theory. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. 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. bounces and sphalerons are introduced and their relevance in quantum mechanical problems is discussed (admittedly in also trespassing the sharp dividing line between quantum mechanics and simple scalar field theory).e. In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. the chapter thereafter deals with Yukawa potentials. i. and their eigenvalues. In the second part. 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. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). periodic instantons. Thereafter the concepts of instantons. which is presumably the only such singular case permitting complete solution and was achieved only recently. After a treatment of power potentials. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. Using perturbation theory.
cubic). the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. however. for instance. This puts Schrodinger equations with e. quartic. We then illustrate the relevance of this in the method of collective coordinates. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. 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. With a fully systematic perturbation method and with applied boundary conditions. In fact. This method is therefore more complicated. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. from our precise reference to unavoidable elliptic integrals taken from Tables). Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. an approximation whose higher . anharmonic oscillator potentials on a comparable level with. we can make the following observations. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. 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. this comparison on a transparent level being one of the main aims of this text. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. All results are compared with those obtained by perturbation theory. it is. the Mathieu equation. and whenever available also with the results of WKB calculations.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. 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. such as limiting procedures. for instance. An expected observation is that — ignoring a minor deficiency — the WKB approximation is and remains the most immediate way to obtain the dominant contribution of an eigenenergy.g. except that these are no longer of hypergeometric type. we consider the topic of transitions between the quantum and thermal regimes at the top of the barrier and show that these may be classified in analogy to phase transitions in statistical mechanics. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions.
g. K. The author is deeply indebted to his onetime supervisor Professor R. Liang (Taiyuan).g. are relegated to separate subsections which — lacking a better name — we refer to as Examples. Andrews). not the least for permitting a more detailed and hopefully comprehensible presentation here.Q. N. For ease of reading. D.XIX order contributions are difficult to obtain. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. B. Throughout the text some calculations which require special attention. . Tchrakian (Dublin) and Jianzu Zhang (Shanghai). 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. as well as applications and illustrations. J. Park (Masan). Since this comparison was the guideline in writing the text. like E. Dingle for paving him the way into this field which — though not always at the forefront of current research (including the author's) — repeatedly triggered recurring interest to return to it. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. Nonetheless. The line of thinking underlying this text grew out of the author's association with Professor R. B. also for the verification of results. T. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). Whittaker and Watson [283]. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. W. the references referred to are never cited by mere numbers which have to be identified e. we also consider at various points of the text comparisons with WKB approximations. 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. at the end of a chapter (after troublesome turning of pages). MiillerKirsten *In the running text references are cited like e. Their deep involvement in the attempt described here is evident from the cited bibliography. thereafter University of St. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years.g. Watson [283]. As a rule. Dingle (then University of Western Australia. in particular Professors J. Whittaker and G. e. shows him that each such formula here has been properly looked up).* H. D. H. with the source given in the bibliography at the end. Instead a glance at a nearby footnote provides the reader immediately the names of authors.
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. very few explain in this context what he really did in view of involvement with statistical mechanics.Chapter 1 Introduction 1. 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. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. We do not enter here into detailed considerations of Planck. the amount of radiation absorbed by a body is equal to the amount the body 1 . 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. Instead.2 cm at moderate temperatures T) is the radiation emitted by a body (consisting of a large number of atoms) as a result of the temperature (as we know today as a result of transitions between a large number of very closely lying energy levels). Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien.
1 Absorption in a cavity.e. i.3 4 J s.T) = dv3eC2U/T.1) ' y Here u(v. energy per unit volume) of the radiation (i. u(u.1) c is the velocity of light with c = u\. radiators.e. T) were known and tested experimentally. v + dv in equilibrium with the black body at temperature T. These expressions are: (1) Wien's law. and then measuring the increase in temperature of the heat bath. and the other in the region of large A (or AT).2) and the . 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. The parameters k and h are the constants of Boltzmann and Planck: k = 1. T)du is the mean energy density (i. kT kXT (1.l ) where x = ^ = ^ .626 x 10 . It was found that one expression agreed well with observations in the region of small A (or AT).38 x 1(T 23 J K'1. 1.T) = 2*?£(?)kT. In Eq.1) containing the constant h by treating the radiation in the cavity as something like a gas? By 1900 two theoreticallymotivated (but from today's point of view incorrectly derived) expressions for u(u. (1. Fig. Let us look at the final result of Planck. Introduction emits. y J c 3 \ex .. i. h = 6. (1.2 CHAPTER 1. How did Planck arrive at the expression (1. Black bodies as good absorbers are therefore also good emitters. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. the formula (to be explained) u(u.e.e. A being the wavelength of the radiation.
2) and (1.47TZ/ 2 {x small). (1. "small" (i.T) 2^^kT. To this end he considered Boltzmann's formula S — klnW for the entropy S. (x large). in the first place Planck had tried to find an expression linking both.2 Distributing quanta (dots) among oscillators (boxes). Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. Considering Eq. C2. e xhv.T) = 2^C3T. but — here the discreteness appears properly — only in elements (quanta) e. which are indistinguishable) among the N indistinguishable oscillators at .e. he searched for a derivation.T) = av e6"/Ti' where a and b are constants.3) are contained in Eq. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. (1. and thus over a discrete number of admissible states. Indeed.1.1) in regions of a.1) as approximations. When Planck had found this expression. we obtain: u(i/. . and he had succeeded in finding such an expression of the form u(v. We see. 1. where the Fig. discretization begins to enter. that the formulas (1.3) Ci. C3 being constants. 3 (1. This is the point. T) u(i/. exp(x) ~ 1+x) and "large" (exp(—x) < 1). Here W is a number which determines the distribution of the energy among a discrete number of objects. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons".
3 Comparing the polarization modes with those of a 2dimensional oscillator. M. .1)!P! (1. as for small values of e). . = 1/T). with riydu — 2 x —w—dv. e. (1. 1.2) requires that e ex is. Gradshteyn and I. Magnus and F. as in most other Tables.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.e.g.2. Ryzhik [122]. (1. i.7) *See e. with the quanta represented schematically by dots as indicated in Fig. U{T) being the average energy emitted by one oscillator. h = const. We visualize the iV oscillators as boxes separated by N — 1 walls. i. S.4) IniV! ~ JVlniViV + O(0). formula 8. Introduction temperature T. 1.4 CHAPTER 1.e. N * oo. and the second law of thermodynamics ((dS/dU)v Example 1. Then W is given by w = (N With the help of Stirling's formula* {N + piy. v + dv. Agreement with Eq.6) Fig. e = his. there not called Stirling's formula. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. u(i>. We now obtain the energy density of the radiation. I. Oberhettinger [181]. (1.g. W. 940.3. The Stirling formula or approximation will appear frequently in later chapters.T)dv. one obtains (cf.343(2). p. p.
Then L^j = nrii.2.7) for instance. We obtain therefore u^T) = Unv = 2^fJ^—i.e.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. (1.V 2 ) E = 0..8) This is Planck's formula (1. v + dv.UJ = 2KV.2. (1. nvdv per unit volume.965 and xmin = 0.2. . dj\l dM . as in electrodynamics. 2 2 2 r2 L K — 7T n .T) has a maximum which follows from du/dX = 0 (with c = vX).1)) The solutions of this equation are ^max = 4... so that (lvL\A 0 I I = rr. 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 = 0.3. where^ 2 [2KUY . '''From the equation I \ JW . We observe that u(v. i.UJ2/C2 + K? = 0.. as indicated in Fig.1.i = 1. rii = 1.7). The number with frequency v in the interval v. We obtain the expression (1. is given by .3..1..3.3 (as for ideal conductors). (1. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. In terms of A we have u(X.1). KT = I J . so that the derivative of u implies (x as in Eq. so that .T)dX = ^ehc/*kT_idX. . L for i = 1. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.
. which can assume the discrete energies en — nhv. and from which the constant h can be determined from the known value of k. which can not be eliminated without a different approach. (1. . i. of course. Thus we have a rather complicated system here. we then have at T = 0 independent oscillators. . that of an oscillation system at absolute temperature T ^ 0. but one can expect an analogy. We are not dealing with the linear harmonic oscillator familiar from mechanics here. 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=—. 2 .1. We therefore examine such contradictions next.e. . that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. which had also been known before Planck's discovery.10). Since temperature originates through contact with other oscillators.e. the behaviour of the system at zero absolute temperature. 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 . Later it was realized by H. x — oo) the mean energy vanishes (0 < U < * > oo). Lorentz and Planck that Eq. . One might suppose now. Introduction he 4. This is Wien's displacement law. n = 0.6 The first value yields AmaxT = CHAPTER 1. However. A.8) could be derived much more easily in the context of statistical mechanics.l.—v <L9) We observe that for T — 0 (i. that it is easier to consider first the case of T = 0.10) Thus here the socalled zero point energy appears. ra = 0. which did not arise in Planck's consideration of 1900.965K = Const. One expects. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.2 (1. such a procedure leads to contradictions.
1.1)! .k dU 1+U\ f ln(l T + U\  U U k ( e . with Heisenberg's discovery of the uncertainty relation.e. 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. kT/2.1)! .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later. 1. In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").= .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV.In . i. . so that 1 f ds\ T — \dUjy . for 2 degrees of freedom. d .2 Contradicting Discretization: Uncertainties 7 Example 1.kT This means U is then the classical expression resulting from the mean kinetic energy per degree 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. 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". we obtain T V S =fc[ln(TV+ P .1: Mean energy of an oscillator In Boltzmann's statistical mechanics the entropy S is given by the following expression (which we cite here with no further explanation) S = fcln W. around 1926.ln(TV .1 ' u~ e .i n . {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) .
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% are therefore again observables.1) as d . (2. whereas the velocity requires observations of space coordinates at different times. (2.t).2) If we have only one degree of freedom (i = 1).pi) of qi.g. since . Goldstein [114] remarks at the end of his chapter VIII. Pi has to be considered as an independent quantity. We can therefore consider Eq. It suggests itself therefore to consider more closely the properties of the symbols (2. which all together describe the state of the system.P. (2. the entire timedependence of observables u(qi. (2.1). (2.e. 6t>0 5t Real quantities which are directly observable are called observables. the equation of motion of the observable u. The total time derivative of u can therefore be rewritten with the help of Eqs.4) contains as special cases the Hamilton Eqs.) = £ [wm + WiK) = £ [WiWi .n n. which can be observed directly at time t. M.1). qi(t + 6t) . Rather. A system consisting of several mass points is therefore described by a number of such variables. H.24 CHAPTER 2.Pi) is contained implicitly in the canonical variables q^ and pi. Hamiltonian Mechanics is wellknown.. (2. i. The following properties can be verified: *See e. as mass times velocity. S »(«.4) as the generalization of Eqs. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.^ ^ j . chapter VIII. .qi(t) qi = hm f . x^fdu. one obtains the Hamilton equations* • OH . in analogy with Eqs.3). It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. (2. du \ x^/dudH du dH\ . One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq.2) as This equation is. this expression is simply a functional determinant. the standard reference for the application of Poisson brackets is the book of P.1). Dirac [75]. (2. Compared with an arbitrary function f(qi. A. . chapter VII.Pi. Goldstein [114]. All functions u(qi. One can verify readily that Eq. ^As H.
we wish to evaluate {A.5b) (3) complex conjugation (note: observables are real. A}} + {C. {A. These are {Qi. (2.A}.B}* (4) product formation: {A.B} = {B.5d) above. Property (2.2 The Hamilton Formalism (1) Antisymmetry: {A. Later we shall consider noncommuting quantities. {B.Pk} = 0. B2}.5c) The first three properties are readily seen to hold.5d) = {A*. (2. it is irrelevant whether we write {A. a i S i + a2B2} = ax{A. that the very general Eq.6). If. that of the linear harmonic oscillator. B}} = 0.B}C + B{A.e. which combines the Hamilton equations. where A and B are arbitrary observables. (2. Since Eqs.5d) is useful in calculations. {pi. the Poisson bracket {A. {qi. i. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. As long as we are concerned with commuting quantities. then the ordering is taken as in (2.5e) = {A. for example.6) We can now show.6) give the values of these. {C. but could be multiplied by a complex number): {A. we obtain the fundamental Poisson brackets.Pk} = 5ik.3) of the Poisson bracket.B}. (2. (2.2.B}C or C{A. (2) linearity: {A.5a) (2.C}.4). The original definition of the Poisson bracket will not . can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. 25 (2. As an example we consider a case we shall encounter again and again. (2. Bx} + a2{A.Qk} = 0.Pi. B} is completely evaluated. If we evaluate the Poisson brackets for qi. like here. in other words without any reference to the original definition (2.B*}.BC} (5) Jacobi identity: {A. C}} + {B. B}.
(2. we consider as Hamiltonian the function H(q..9) Similarly we obtain from Eq. These are transformations qi—>Qi = Qi(q.6). and so q = q.H}. (2.p) is an ordinary function is also irrelevant. (2. and Pi are ordinary real number variables and that H(q. Pi—> Pi = Pi(q.26 CHAPTER 2.8a) and use the properties of the Poisson bracket and Eqs. (2.qot2 .P2} + {q.7) into (2. (2. (2. According to Eq. Since constants are also irrelevant in this context..8b) p=q. q = {q. In the evaluation one should also note that the fact that g.p. Then we obtain: (2. "4' = q.q2}) = = 2i{q.p}p + p{q..7) q = [q.8b) We insert (2. (2.11b) (2.t).yPot3 + •••• (2.t).10) In classical mechanics one studies also canonical transformations.H}.p\) P.p..8a) (2.4) we have for u = q.12) . q + q = o. or q(t) = qo+ Pot . From Eqs.p) = ±(p2 + q2).\(p2 + q2)} = l({q.11a) 'q' = q. and P={p. from which we infer that q(t) — qocost + posint.p.10) we deduce q = p = q.9) and (2. (2. Hamiltonian Mechanics be used at all.
those at a time t — 0. i. Of course. verify that the Poisson bracket of two observables A and B is invariant. {Qi. i.Pk} = 0.t). i.p <> Q.e. Example 2.12) Qi—>qi = qi{Q. .Pi back to their constant initial values.B}Q>P.e.3) we can now express the Poisson bracket {^4.2. .p = {A. {PhPk} = 0. (2.14) With the help of the definition (2.P.qk} = 0.P One can then show that {A. B} of two observables A and B in terms of either set of canonical variables. and Example 2.15a) °raS {AiB}Q.2 below contains a further illustration of the use of Poisson brackets. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations. chapter VIII.pk} = 5ik.P provided the transformation q.Qk} = 0.e.15a). (2. i.P is canonical in the sense defined above.13) We write the reversal of the transformation (2. (2.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. {QhPk} = Sik. which means that a Hamilton function K(Q.1 we verify only Eq.e. {qi.t).15a). which transforms qi.B}q.3 deals with the relativistic extension. Eq.e. for which Hamilton's equations hold. (2.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P). _ax p__dK_ (2.B}q.P. (2.* Hence in Example 2. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here. as {A. Example 2. Goldstein [114]. *H. that (dropping the subscripts therefore) {Pi. P) exists. Pi^Pi=Pi(Q. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation. {qi.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. i.
Qj}q.F}r ~di d~E ~ d~E~8t Consider F = H(q. § See P.B}q. OQk Pk}q.p = J2 d~Fk Replacing in the above A by P and B by A.p) . we obtain dA {Qk. we obtain analogously dA dQk' Inserting both of these results into the first equation.15a) r„ ™ v . § E x a m p l e 2. p.p^v dP.+ dQ. Thus whenever q and p are multiplied.p = dB dA dB \ rA 1 {A'B^ E b r s r . V V °Qk dPk dPk dQk J E x a m p l e 2.p(0) = poSolution: The solution can be looked up in the literature. their product Et — qp being relativistically invariant.q/c) and (E.B}QIP.Pj}q.( 9A = {Pk.E(t).s F 5 7 r ={A.15b) dA {Qk. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. 236. (2.2: 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). we obtain as claimed by Eq.A}q.p^ "Pk.A}q. Thus we extend q and p to spacetime vectors (t.P^pr. Qk}q.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.p^— + {A. Solution: Relativistically we have to treat space and time on an equal footing. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets.3: Relativistic Poisson brackets By extending qi. Mittelstaedt [197]. dA {Qk.pc). . Replacing here A by Q and B by A. Hamiltonian Mechanics S o l u t i o n : U s i n g t h e definition of t h e P o i s s o n b r a c k e t a p p l i e d t o ^4 a n d B w e h a v e r „ „•. {A. we have —Et. (2.28 CHAPTER 2.
g. This probability is evidently proportional to AqAp. if qo. c dt 2 2 w^ du du dr .2.p) to G\{q. and the space spanned by the entire set of canonical coordinates is described as its phase space.Pi are the coordinates of some mass point m. Since Hamilton's equations give a continuous map of one domain onto another. i.po. We consider first the a priori weighting or a priori probability.. (1.. we have A = (p apaq — J .y.e.t). Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi. rrr . i. and E as a function of t). Probabilities 29 (This is._.t). where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du .p).)•'=(*)**£.dt y/l  qZ/c2' 2. of course. Instead we assume a case in which we know only that the system is located in some particular domain of phase space. so that Go(qo. g oc AqAp. Let us assume that at some initial time to the system may be found in a domain Go(q.po is a point on Go. We distinguish in the following between two kinds of probabilities.3. but partial derivatives of F do not vanish.p).16) For example. Of course. But now we consider a system whose phase space coordinates are not known precisely.3 Liouville Equation. 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. one obtains Gi with q = q((lo. also boundary points of one domain are mapped into boundary domains of the other.to.17a).p) . (2. Then . du . dudH s l u . I . and at time t > to in a domain G\(q.Po. it is the equations of motion which lead from Go(q.p). in the case of the linear oscillator with energy E given by Eq.1 Liouville Equation.. (*.t0.po) = Gi(q. 2TTE UJ and hence g oc — . p= p(qo.. H(q..' ±=j1V c \dtj' u^. to .14) and area A of the phase space ellipse given by Eq.e.p) around some point qo. dE . du q\ P = —. . .E(t)}r = Hence at Relativistically we really should have clu/dr. (1. since H is expressed as a function of q and p. numerically zero.3 2.Po.
as is demonstrated by Liouville's theorem in Example 2.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E.4. dpdq and similarly dt = — Ap. Example 2. Pi 2 / \. Hence from the difference: — .p^.15) — an ellipse of area § dpgdp.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . this has the same value at a time to. dt dq dq dp dp d2H dqdp d2H = 0. which means. J0=0 Hence g oc 8n2IdE. in view of this independence it can be expressed in terms of the conserved energy E. Show that the total volume of phase space covered for constant E is 8n2IE.eded<t> = 8TT2IE. — dp A A — In(AqAp)s = — + — = Example 2. and hence g oc %ir2IdE.1): d . Example 2. (1. dt dq and with with Hamilton's equations (2. Thus g must be independent of t. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . = 2TrIEsm6. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin.— = — Aq.)curve for constant E and $ is — as may be seen by comparison with Eqs. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle.. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left./. the (pg. Je=0 .14) t o (1.4: Liouville's theorem Show that A q A p is independent of time i.30 CHAPTER 2. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. 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.
by jr = P(9»P.e.e.p.p. i.1 The system moving from domain Go to domain G\.q + dq. Probabilities 2. d d 1 P = JJdqidpi. Thus dn is that number of systems which at time t are contained in the domain q. leaving the basis of classical mechanics! Hence we set / «w)^ =. dn — p(q.3.t).2. 2.20) .p + dp. (2.3 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.p.p.Thus we assume a large number of identical sytems.19) Thus W is the probability to find the system at time t at q. i.p. Since W has the dimension of a reciprocal action.Po. it is suggestive to introduce a factor 2KK with every pair dpdq without. however.17) F ^GT^^ 0 Fig.*). (2.18) With a suitable normalization we can write this / W(q. (2. W=p(q. whose positions in phase space are characterized by points. The total number of systems N is obtained by integrating over the whole of phase space.t)dqdp= 1.2 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.t)dqdp = N. (2.
The equation of motion for n or W is the socalled Liouville equation. * q Fig.t)dp{ jt Q. 2.a t dt + q+dq.p. In order to derive this equation. i. — if vq(q. t + dt)dqdp. p + dp G P O q qndq Fig.p / . is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.e. t + dt)dqdp — p(q.32 CHAPTER 2. we take into account. The number of points at time t + dt in domain G.p. t)dqdp p(q.p) and vp(q.p) denote the velocities in directions q and p — p{q. We are now interested in how n ov W changes in time. In doing this. that in our consideration no additional points are created or destroyed. i.P dt p(q + dq.3 and establish an equation for the change of the number of points or systems in G in the time interval dt.e.3 The region G. i. 2.t)dp( — I .2 The ensemble in phase space.p.p.p. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space.e. how the system moves about in phase space. p(q. 2. we consider the domain G in Fig.
P)]K(Q. .p)dtdp .p + dp. t)dqdp = p(q.p(q.3 Liouville Equation.t)vq{q.p)=p = 3H —. 33 Dividing both sides by dqdpdt this becomes p(q.4) shows that p and u satisfy very similar equations.p.p.t)vq(q + dq. = K(9.P.2.21) we can also write dW(q.21 ) This is the Liouville equation which describes the motion of the ensemble or.p)dtdq .t) dt p(q. put differently.p(q + dq. (2.t)vq(q. (2. Probabilities and thus p(q.p.p + dp)dtdq.p. vp(q. Comparison of Eq.P *' % = iH>P} ( 2 . the probable motion of the system under consideration. t)vq(q + dq. (2.t)^ = 1.p.22) The generalization to n degrees of freedom is evident: The volume element of phase space is .p(q + dq.20) and (2.p).t) dt {H(q.P)].p) .p) dq p(q.t + dt)dqdp — p(q.p(q. dH Mq.p.t + dt) . p.p.t)vp(q. . . p.t)} with JW(q.t)vp(q.p(q.p) = q = g^. With Eqs. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .t)vp(q.p)dtdp +p(q.19).p.p) .p.21) with Eq. (2.p + dp) dp = or  However. (2.p.t)vp(q.p + dp.W(q. .
Thus in Fig.p. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.p.t)(^J. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. (2. that dtj\y v if qo. i. since no systems are created or destroyed.4). We now inquire about the time variation of the expectation value (it) of u. (.24). (2. We shall see that we have two possibilities for this.p)W(q.that the density or probability W(q.p) be an observable. We deduce from the Liouville equation the important consequence that ^ M = 0.1 the area Go is equal to the area G\. p. With Eq. i. (2.po are the initial values of q.p) the following expression: (u)=Ju(q.26) we described the time variation of the observable u(q.34 CHAPTER 2.e.24) implies that p is a constant in time.p). We define as expectation value of u(q. Example 2.p (cf. for i<«>!/«<*p>"w>(^)".24) since the total derivative is made up of precisely the partial derivatives contained in Eq.e.4). and hence that equal phase space volumes contain the same number of systems. p+dp. (2. The first and most immediate possibility is — as indicated . Equation (2. t) depends explicitly on time t (if determined at a fixed . q+dq. 2.4 Expectation Values of Observables Let u = u(q. 2.
Goldstein [114].28) = Ju(q.29) The distribution of the canonical variables is given by W(q.po.t) W(q0. we can express these in terms of their initial values qo. (2. Observables 35 point in phase space).33) po = f(q.p. ie. However.W(q. where we used Eq.t)W0(q0.t)}(^y. We expect.po.t).Po.i).t) = = W(g(q0. Sec. (2. since W oc p is constant in time according to Eq. Then Eq.32) In this expression the time t is contained explicitly in the observable u(q. (2.t).0) =u(g(qo. We verify this claim as follows.Po. 8. (2.^ Solving the equations of motion for q.po.t).t).p) = uo{qo. of course.34) S e e also H.t).0) = W0(q0. so that Q = g(qo.p0)(^^J. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t).30) p = f(qo.p).p.po.24): W(q.po. . we can also employ a more complicated consideration.0) = u0{qo. Reversing Eq. (2. (2. (2.f(q0.t). (2.29).f(qo.Po) at time t = 0.Po.p.p.p.Po.p.4 Expectation Values.t).p){H(q.31) i.t).p.p) assumes certain values) that depends explicitly on time. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.8. and hence u(q. Thus we can write.p) = u(q. at t = 0. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.2. 1 (2.t). we have Qo = g(q.p0.27) becomes !<»>  /«(*P>!"W>(^)" (2.e.22). (2. that (u) = (u)0.Po.
.f(q. (2.t).p.36 CHAPTER 2. [ du0{q0. (2. (2.30): duo(qQ.M 27Th i .(dq0dp0\n We deal with the partial derivative with the help of Eq.Po)(^^)n.p. (240) . Eq.p.o.0)=u(q. d dt {U)0 = dt. (2.t) 0H_ 0 q =  {{H(q.25) into Eq.f({H(q.30) Moreover. p = f(<j(q.p).p.t).30) u0(g(q.p.f(q.) f / u (q0.PO.p. i. (2. Eq.(2.t).Po.t) = (2 30) (2.p.t) v{g(<j(q.t). (2.28).36) = (2. we obtain («)o = / u{q. / ^fdq0dpc .t)W "(q0.t). Hamiltonian Mechanics so that on the other hand with Eq.P)}).32).p.t) d t _ (du\ \ d<i dq / P=f(qo. (2.36) and (2.t).po.p).35).t)J(q.t) (2.p.) Wo(qo.t) dB__ (du\ °P \°PS P=f(qo.P0. Taking now the total time derivative of (it)o.p)})q=g(QO.t).u(q.t).P0. (2.f(q." V " " ™0' > • • » « » 0 .t).0) u(q. (cf.t) dp M du\ ° q 'P=f(qo.31)) W0(q0.P.Po.t) .f(g(q.\„.p.t).38) we obtain £<«)„ = .p)W{q. Inserting Eqs.29) q = g(g(q.37) as had to be shown. (2.t). (2.t).p0) = W(q.t).35) Inserting these expressions into no we obtain uo(qo.39) Substituting this into Eq.P.u{q.p.e. (2.t).POit)i p=f(qo. we obtain an expression which is different from that in Eq.37) and (2.Po.Po.p).t) M f 9u\ V °P / P=f(q0.32).t)\—^\ = (u).po)(~ .po.=9(.f(q.po.
W} in Eq.t) dt {H(q.t)} .p). (2.28). (2.45) the Liouville equation. uW} = {H. we obtain zero after partial integration of I and hence from Eqs. Consider {H. described as "Schrodinger picture". This timedependence is described by the Liouville equation dW(q. lim ——uW — 0.t) of u assuming certain values q and 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..p)) is attributed to the probability W(q.31). W}. (2.p)}W(q.4 Expectation Values. The phasespace integral of a Poisson bracket like I (2.p.34) and use (2. .p. ^ „. However. / d q d p \ n 2irh J (2.p). x r „. u}W + u{H.u(q.p.t) dqdp\n 2irhJ (2. The considerations we just performed demonstrate that we have two ways of treating the timedependence: The explicit timedependence can either be contained in the probability W or in the (transformed) observables.p.(2..p). from the properties of the Poisson bracket. .25) and (2.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). dt <«)o v f u(q. so that ~d~t <«>0 = J{H(q.2.43) ——uW = 0.42) the relation d .45) in agreement with Eq.p).W(q. we obtain {H.W(q.p){H(q.41) This expression contains {H. Observables 37 Here we perform the transformation (2.p.. u}W instead of u{H. the observable u is treated as a function u(q. Alternatively we could deduce from Eqs.41) and (2.28). and the time variation of (u(q. (2.t)}.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. In the first case.28) and (2.
t) dt __ ~ (2J30) ( 4) du0(qo. (2.H(q. .e. which describe the state of a system.p0. does not have to be commutative.46) We thus also recognize the connection between the Liouville equation.4) of an observable on the other. (b) As usual a muliplication by a complex number is defined.p). Koppe [152] at the university of Munich around 1964.F(«. The time dependence of the expectation values can be dealt with in two different ways. 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. i.p) It = {u(q. as the equation of motion of an ensemble or of a probability distribution on the one hand.** 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.p)}. called u Heisenberg picture". for which the axioms of a commutative group apply.38 CHAPTER 2.po) is assumed. and the equation of motion (2. the reason being that — since qo. the probability of the initial values Wo(qo. 2. Hamiltonian Mechanics In the other case. (cf.po.po are constant initial values — we have du0(q0.po. Eq.t) dt du(q. (2.t).. and probabilities W.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. which.rt. but does satisfy the usual associative and distributive laws. " T h e author learned this approach from lectures of H.p)}.4)) ^ = {«(. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. The equation of motion is then that of an observable. as we observed. This formulation deals with observables u representing physical quantities. " This means: Only in the Heisenberg picture dW/dt = 0. (c) A multiplication of the quantities among themselves is defined." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.
2. which all apply to (2. it is helpful to introduce already at this stage some additional terminology.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x.p} = l {q. Eq. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. ox In view of our later considerations.p] = l. also called "density matrix". if u ^ 0.p} = 0 • — ^ ~[q. That this is the case. Moreover. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2.6) a socalled "commutator''' [A. can be deduced from the following characteristic properties of a trace.48): (a) TV u* = (Tr «)* = TV^. for a phasespace integral we define the word or symbol "trace".q}=0. and to be able to correlate these with the above classical considerations. Introducing it here assumes.) = TV (vu). Thus we can write the expectation value of an observable u (cf.48) In matrix theory the symbol "trace" has a welldefined meaning. y:\p. (2. by Traceu:=Ju(q. B] := AB — BA in the following way: 'i {q.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. ~[q.26)) (u) = Tr(wW). In Chapter 9 we attempt a corresponding approach for classical systems with . also written "TV". so that we consider this next. therefore. that its use here implies the essential properties it has in matrix theory.q} = 0 {p.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. px > in—. (b) TV(cm + j3u) = oTV u + /3Tr v. (d) Tr(ut. (2. (c) TV (u*u) > 0. (2.p)(^y.
e. . However. since gauge fixing (i. it will be shown that in electrodynamics.e. and. These aspects will be considered in Chapter 27. the Poisson brackets require modification to Dirac brackets. electrodynamics. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. Hamiltonian Mechanics a wavelike nature. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. excepting the Poisson brackets. i. a constraint) has to be taken into account.40 CHAPTER 2.
Building upon this. i. 3. if the elements ipi of M satisfy the usual axioms of addition and K 41 . We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters. we can define the Hilbert space as the space of states of a physical system. A set M. 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.e.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. We found that the Poisson algebra permits extensions to noncnumber formulations. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.e.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. of measurable quantities. which turn out to be those of the theory today known as quantum mechanics. i. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. and selfadjoint operators in this space as representatives of observables.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. with the canonical commutation relations or Heisenberg algebra defining the basic product relations.
aiV>i + a 2 ^ 2 ) = ai(ip. ( n (3.~ip2) '• M. f > 0 if V T ^ O .ipn are said to be linearly independent. if any two elements ip\. the vectors il)i. (3 5c) + a2{ip. E (3.ip2. .3) If all on = 0. (a.. n. . with the properties (a* G IK): (^2. are linearly dependent.4) The vector space . (ipi. A + tpj = tpj + i'i.ip2). . if numbers eti. . fa. (ipi + ipj) + ijjk = ipi + (ipj + ipk). such that n ^ = YJCi^i. and n is the smallest such number.1) aipi + aipj.^1) = (01. i. ip2 of this space can be associated with a complex number (V'l. —»• IK. . Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. (3. . .tp2. 2 . .^) called inner product. (V>./?eK).5b) . 2 . x M.e.02)* (hermiticity). n is called the dimension of .  . so that n 5 > ^ = 0.42 CHAPTER 3.5a) (3. ^ i = 0 if ^ = 0. (aptyi.^i) . . .i = 1. if every vector ip E M can be associated with numbers Q . (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M).M.2) Vectors ip\. If n + 1 elements ^ € M.M is said to be a metric vector space or a preHilbert space. In each case n linearly independent vectors are said to form a basis. (3. i=l (3.ipn are said to be linearly dependent. • • • . n exist (not all zero). i = 1.
9a) we start from if) = ipi + \if>2 6 M.7) In addition the following relations hold in a metric space M. A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i.9a) (3. (3. if Wi.. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2. meaning oneandahalffold linearity).tp2 £ M.2. which we can write 0 < ( ^ i .9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.9d)   V i   = sup  ( ^ i .i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).^1) 2 h 11 0 <%«e + M .V') +"2*(V. linearity in the second component.)> (36) i.2 Hilbert Spaces 43 where the asterix * means complex conjugation. for arbitrary A and ip2 ^ 0: (V>i + AV>2.*h)=\\AH\(38) (3. (310) (^2. »2 2 .rh) (3.9b) = 0 (Pythagoras theorem)..^) = ai*(V'i.is defined by d(ipi. (3. In order to verify Eq.3. The distance between two vectors if>i. antilinearity in the first component (also described as sesquilinearity. for ip\. (3. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). tp2 £ M: 1(^1. ^ i ) + A * ( ^ . ^ i ) + A ( ^ i .9c) HV'i + tp2\\2 + \\1p1 .V>i + A</>2) > 0 . ^ ) + A 2 (V2.V. (3.^2)1 < H^ill • 11^211 (Schwarz inequality). WA+Ml2 = ll^il 2 + ll^2 2 . The first two properties imply ( a i ^ i + a2ip2.^ 2 ) =   ^   2 + 2 ^ i . ^ 2 )  .e.
10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. (3. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. ll^ill + llVdl. (3.* Two vectors i f i . Examples of metric vector spaces: (1) Let M.9a). so that   ^ l + ^ 2   <   ^ l   + ^2.9c). for which CO 2 := V^l^l 2 < 00. beginning with A = 1 in the second line of Eq.V2) IIV'l2 + V'22 + 21(^2. (3.\\ih\\. ^ e M are said to be orthogonal if (</>!. o ^ V i = o.^1)1 Vi 2 + ^2 2 + 2   ^   . HIHI (3.44 or CHAPTER 3.ii) thus verifying Eq.13) the vector is said to be normalized. (3. (3. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. V 2 ) = 0 .9e).9d).12) We omit here the verification of the remaining relations (3.^)1 <IIV>il.  ^ l l = (HlMI +   « 2 .11). . (3. hence verification in each case is necessary. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . In verifying the triangle inequality we use this result (3. Mathematical Foundations of Quantum Mechanics 1(^2.
with inner product oo 45 (v. and one defines addition and multiplication by complex numbers with respect to these classes. i.5b).5a).[$]):= / JM /(x)* 5 (x)d 3 *. .w) :=J2v*Wi. which satisfies relations (3. */ x } = / /o for x = 0. is defined by ([/]. Then L2 is the space of all these equivalence classes.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi).3. In order to avoid this difficulty.g)= [ /(x)^(x)d3x. for which the scalar product. Elements of the classes are then called representatives of these classes. •'^ \ 0 otherwise. (2) Let M.e. 2 and so on.5c)) / ( x ) = 0 = > ( / . all squareintegrable functions / which are "almost everywhere equal". i. With the scalar product (f. Eq. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. etc. = C? be the set of all complexvalued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 /(x) 2 d 3 x < oo. which differ solely on a set of measure zero. Js the space C? is not yet a metric vector space although for (cf. / ) = 0. (3.5c). (3. (3.e. But this applies also in the case of any function which is nonzero only on a set of measure zero. are combined to an equivalence class [/] (with space L2).
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.
5) 4. Dirac's Ket.5c)). Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. It follows that we can now write the Schwarzian inequality (3. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. The dimension of V is the same as the dimension of V. if x\u) = 0 V kets \u) E V. or (4. This expression is linear with respect to \u) and antilinear with respect to \v).and BraFormalism Furthermore we have: ( x  = 0.4c) (44d) (v\ = J\*(0(m One defines as the scalar product or inner product of \u) E V and \v) E V the cnumber (v\u).62 CHAPTER 4. Two ketvectors \u). if (u\v) — 0.e. (3) (u\u) = 0. V \u) (in the case of £) with infinite norm!).4b) (4. (4.4a) \v) = J\(m<% and (4.3 Linear Operators. the norm squared. that the vector \u) E V is associated with a vector in V. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. (2) (u\u) > 0.\u) = (u\0*. which we write (u\.(t. i.9a): \{u\v)\2 < (u\u)(v\v). \v) E V are said to be orthogonal. if \u) = 0 (see (3. 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.
. This does not always exist.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). and one writes: (V\ = (X\A. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2.e. If Vi has the dimension A/i.3 Linear Operators.or Kronecker. i. AB^BA. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. defines the unit or identity operator. Then one says: (771 results from the action of A on (x. Two operators A and B are said to be inverse to each other. Finally we define the tensor. denned as a linear operator on V. This product is commutative. ltt) = \u). if \v) — A\u) i. as we can see as follows. the space Vi <S> V2.4. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . on bravectors are similar to those on ketvectors. Furthermore linearity applies.e. Furthermore. The inverse of a product is (AB)'1 =B1A~1.. Let the following bravector be given: (x\ £ V. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). 1 and \u) = B\v). The operator A. Then the following rules apply: A + B = B + A. The inverse of A is written A .e. It follows that (r. B. let (771 6 V be the bravector defined by this functional.. BA = 1. also acts on the dual space V. i. Multiplication by unity. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). In this way the operations of operators A.or direct product of vector spaces. The product vectors l?^ 1 )^ 2 )) span a new vector space. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. if AB = 1. The product of such vectors is written: \u^u^) = \u)^\u)^. the space Vi <S> V2 has the dimension .
i. Every operator A^> commutes with every operator A(2\ i.e. AW\U)W = \V)W. also Sec.7a) (4.7C) («)(u)t = \v){u\. \t) G V. in the product space. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. If (v\ = (u\A.and BraFormalism an operator in the Mi A/2. we obtain the conjugate relation <tAt«> = (u\A\t)* (4. here designated with the same symbol. i.3).7b) (4.6) for all \u). AW\uW)\uW) = \vWUW).e. Let A^> be an operator in the space Vi and A^ space V2.e. Dirac's Ket. i.g. Every such operator is then also associated with an operator. As a consequence we arrive at the following properties (e. we write \v) = A*\u).7d) . Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.e. Since we demanded the scalar products to have the property (t\v) = (v\ty.A^] = 0. [A^. (v\t) = (u\A\t). Next we construct the following scalar products: (t\v) = (t\Ai\u). 3. and (v\ = (u\A. then \v) is a linear function of \u). since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW).64 CHAPTER 4. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4.
e. but only if these commutators commute. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. as is also (u\u).4. Obviously the quantity \a)(a\ is an hermitian operator. for we had (\u){v\)i = \v){u\. or (u\A\u)* — {u\A^\u) = (u\A\u). i. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H.K\. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. 65 The linear operator H is called hermitian. . The operator H is said to be positive definite. so that (n^4n) is real. Verification: Since A — A^ and A\u) = a\u). An hermitian operator A has the properties (a) that all its eigenvalues are real. [H. An important theorem is the following. The commutator of two hermitian operators is antihermitian: [H.K}. if H — H' and antihermitian. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian .K]^ = = [HKKH]* KHHK = = K^H^ [H. It follows that a is real. i.3 Linear Operators.e. K] = 0. it follows that (u\A\u) = a(u\u). Analogous considerations apply in the case of bravectors (u. so that (a)(a)t = a)(a. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. hermitian The product of two hermitian operators is not necessarily hermitian.
Ps is idempotent.Au) = a(v\u) — b(v\u) — (a — b)(v\u). we have 0 = (uArt) — (u. Next we introduce the very useful concept of projection operators. Assuming A\u)=a\u). since a ^ b. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. which includes vectors of infinite norm. If these form a complete system. i. (u\u') = 5{v . since a is real.9a) . (4.66 CHAPTER 4.e. a ^ b. Ps2 = Ps(4. 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. The entire set of ketvectors spans the extended Hilbert space. Dirac's Ket. ( n  z / ) = 0 . (n\n')=5nnl.u').e. as we can see as follows. i. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S.) = 0. i. (v\A = b(v\.9b) (4. Let S be a subspace of the Hilbert space !K. so that (u\Ps = (us\.and BraFormalism Moreover. and let S* be its complement. Ps\ua.8) The projection operator has the following important properties: Ps is hermitian.e. but the continuum vectors are normalized to a delta function. {us\us*) = 0. \u8*) E S*. The above theorem remains unchanged. A continuous spectrum can immediately be included by passing to the extended Hilbert space. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). (u\Ps = {us\. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). i. so that (v\u) = 0.e.
Then 0 = (P2P)\p) and hence p = 0. Let p be an eigenvalue of the projection operator P. iV) is a set of orthonormalized states.4. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. 2). i. 67 (4. \ua) = c\a). = (p2p)\p). Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. i. P\p) =p\p). so that (Ps2 ~ Ps)\u) = 0. \u) arbitrary.3 Linear Operators. i. (1 — P)\u) are orthogonal. so that \ua) = (a\u)\a) = \a)(a\u). where £ is a continuous index.e. Ps is projector onto S.9d) Thus the vectors P\u). Let a continuum state be written £). The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace.e. p 2 .. (1 — Ps) is projector onto S*. Very similarly we can construct operators which project onto the subspace spanned by the continuum states. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. . Set \u) = \ua) + \ua*) with (a\ua*) = 0. if the set of vectors 1).9c) This property can be seen as follows.p = 0.e.e. i. (4. The quantity \a)(a\ is called elementary projector..1.. as can be seen as follows. . c = (a\u).
which are orthogonal to each other (i. which we introduced earlier.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.114c). Coulomb potential.e. In general it is possible.6. i..68 CHAPTER 4. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. Observables are representatives of measurable quantities. Examples 8. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.4 Observables Operators which play a particular role in quantum mechanics are those called observables. that one and the same eigenvalue Aj is associated with several eigenfunctions. u2). 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. Let \u\). see Eq. one says the degree of degeneracy is r .g.114b). and Eq. are linearly independent).6.. 4. However. See e.r)(ui. If there are r such vectors.r\ "An example is provided by the case of the hydrogen atom. (11. Dime's Ket.. 8. . the operator P is correspondingly P2\u)= f2\0m\u). in this case the spectrum also has a continuous part. Degeneracy will be discussed at various points in this text. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator).e. be a system of basis vectors in this space. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K.1 and 11. (11. P = £n)(n. Sec.
r \(ui. the operator A is completely determined by specification of the eigenvalues A. and APi = XiPi.10) gives the linear combination it) = ^ i.4 Observables The dimension of this subspace is that of the degree of degeneracy. Let us set {A . Together with the orthogonality condition.r\u) i. in the case of degeneracy with (ui. The uniqueness of the expression (4.e.10) follows from the fact that i Applying the operator A to the projector (4.12) .r.r)(ui. The operators Pi are linearly independent. i i (4.r\ = l. the relation (4. we obtain i i i i i.r)(ui.r) (the convergence.10) expresses the completeness of the orthonormal system.e. Applied to an arbitrary vector \u) G "K Eq. PA = y£pi i = Yt\ui.r\u).r (4. (4. is always implied).r'} = Sii>6rr>. then PiPi' = 0. or also as subdivision of unity or of the unit operator.11) i. 69 i If A is an observable and if the spectrum is purely discrete.r =^ i.r\ui/. i. and the eigenvectors \ui.e. i.10) This expression is known as completeness relation or closure relation. (4.4. the projection of this operator is the entire space. i.e.r \ui. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r>(ui. If Aj ^ Aj'. ^ A i P i = A = ^Aiui.\i)Pi = 0. which we do not enter into here.r)(ui.r\u)\2.10).
> rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. (4.e. i.and BraFormalism This is the expression called Parseval equation which we encountered with Eq. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. for instance. we have the corresponding generalizations.70 CHAPTER 4. The ketvectors \uv) are orthonormalized to a delta function. as explained earlier. for instance exponential functions. the eigenvector \ui): f(A)\ui) = f{\i)\ui). (4. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). Dime's Ket. i. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). One defines as action of f(A) on. If the spectrum of an observable A consists of discrete as well as continuous parts.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). 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.v').e. Let v be the continuous parameter which characterizes the continuum. (3.17) previously. (u„zv) = 8{v . now however. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. In a similar way we can handle functions f(A) of an observable A.14) . written in Dirac's notation.
if and only if they possess at least one common system of basis vectors. We shall return to this theorem.e. their commutator vanishes.B. (4.. again later. (3.5 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. [A. Extending this we can say: A sequence of observables A.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. / dx\x){x\ = 1.17) The Fourier transform provides the transition from one representation and basis to the other.16) represent subdi . where the symbol ip is already indicative of the Schrodinger wave function ij. x) € Fx. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. if (a) they all commute pairwise. First of all we can rewrite the integral representation of the delta function.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. (4. which all commute with one another. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.and bravectors.e. (4. 4.. In the onedimensional case we write {x\x') = S(xx'). B]. For many practical purposes the use of the Fourier transform is unavoidable.59). i.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. \k)€Fk.. form a complete set of commuting observables. with energy E. Since both expressions (4. and (b) if they possess a uniquely determined system of basis vectors. i.. for which the completeness relation is f dk\k)(k\=l.58) in terms of ket. In the following we consider more generally ketvectors \ip) as representatives of the physical states. Therefore we want to reexpress the Fourier transform (3.4.C.15) and (4. Eq. as a formal orthonormality condition. here presented without proof.
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). Dime's Ket.2a) etc. (4.and BraFormalism visions of the unit operator. (4. (4.72 CHAPTER 4. .18) According to Eq.1a).59) or Eq.19) Comparison with the orthonormalized system of trigonometric functions (4.e. Rather \x) and \k) serve as basis vectors in the representation spaces Fx.Fk.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. (3. (x\k) = ^=eikx.e. i. which are representatives of the states of our physical system. called wave function. V 27T (k\x') = . <a#) = [ dk{x\k)(k\ip). The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. il>{x) := (x\i/}) : "K * C. Obviously we obtain this by inserting a complete system of basis vectors of Fk.space representation. we can rewrite Eq.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. i. (4.L e " * * ' = {x'\k)\ V27T (4.21) provides the ketvector \ij)) in the A. (4.e. this expression has to be identified with — / dkeikxeikx\ i. ${k) := (k\if>). shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).15): 5(x . The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K.
e. the calculation of which for specific cases is the subject of Chapter 7. which is the quantum mechanical analogue of the classical probability density p(q. 1. i.t). 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.2 The Density Matrix We shall establish a matrix p. We mentioned earlier (in Sec. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. Schrodinger [244].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. with the complementary system or *E.Chapter 5 Schrodinger Equation and Liouville Equation 5.1 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.p. 73 . Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). if we do not take into account the complementary part of the universe. 5.
.. (a\i)=0 = {j\b). Schrodinger Equation and Liouville Equation the measuring instruments. where i represents collectively all quantum numbers of the state of the system under consideration.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. \a). as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. 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).i. These are the quantum mechanical states which are also referred to as microstates.b. The actual and real state of a system is then a superposition of two types of states. (52) (5. called the density matrix. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. a (5. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows.3b) The quantum mechanical expectation value of an observable A. we have to deal with a socalled "mixed state".3a) and N J  a ) ( o  = l in the complementary subspace. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. that is. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe... but also of the states of the complementary set of the universe.i. are called "pure states". where T means transpose: p = CTC*. we have (a\b) = 5ab. \a).b.74 CHAPTER 5. of an operator that acts only on the states of the system we are interested in. i.^ Since p is hermitian. \i).e. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. The states \i) £ "K. 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. i. T ..e... p" ( C C * ) t = CTC* = p.
(5. This ensemble is interpreted as a large set of identical systems.6) (i'\p\i") = Wi'Si'vi = real. (5. which are all subjected to the same macroscopic boundary and subsidiary conditions.e.e. Gibbs introduced the concept of ensemble. in the position state representation or xrepresentation. i. This transition to a new set of basis vectors in which the matrix becomes diagonal. which today is a fundamental term in statistical mechanics. a complete orthonormal system with properties (5. In this sense we have p(x'. or (5. but occupy in general different microscopic. .2) and (5. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. i i (511) In Eq.\p\i")(i"\ = Y. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). e. where x represents collectively the entire set of position coordinates of the (particle) system.. In the following we write simply \i) instead of \i'). a projection operator. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i). If the ensemble corresponding to the system under consideration consists of Z elements (i. in a specific representation. P ^ ) = i)(#) = (#)i) (5. as we discussed previously. as we saw earlier.) = YJc t .3a).x) := (x'\p\x) = yju.4) we had the expression (A) = Y/Pij(j\A\i) i.7) p = ^2\ii)(i.* We can specify the state of our system. quantum states. i.3 with Pi^YtCaiCaj*.5. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.10) and is therefore.g.2 The Density Matrix 75 \i') 6 "K. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. (see below) eM.9) is diagonal.j(x'i)(ia.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. (5..e.ji(x / )i*(x). a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.. The ensemble can be represented as a set of points in phase space. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. seen macroscopically of Z copies of the system).e.
To this end we set A —>• Ai' := \i')(i'\ (no summation over i').4)) (5. (5. Hence Wj/ > 0. (A.18) Substituting this into Eq. (5. (5. (5. In particular for A = 1. (5.17) Hence the sum of the classical probabilities is 1. (5. On the other hand. Thus Tr(j>ii)<i)=l. 3 (5. as claimed by Eq. according to Eq. Eq. i t (5.16) (5. We can also show that Ui > 0. Eq.14) which we can write (A) = Tr(pA).18). Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).76 CHAPTER 5.4). We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l.3 = pvv = uv. we have Tr(p) = l.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.) := (mm = (#') W> = l(#')2 > 0. Then the expectation value of the observable A is (cf. (5.5) above) (i\p\j)=Pij.
3 The Probability Density p(x.21) p=\i){i\=Pi. We observed above.t). i. (5. The effect of the interaction of the system under consideration with the surroundings. not in a mixture of states. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. the system is in a pure state. defines the statistical operator. (5.t)\2.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. We now consider the latter expression. i. Thus we have a system in a pure state. so that p{x. i. that these are actually not sufficiently general. by representing the vector \i) by the wave function i(x. but we see here.e. t) p= i The expression (5.t) and instead of (x\Pi\x) we now write p(x. thus enters these states. one says. (Question: Is the state of the universe a pure state?) 5. in all other cases the state of the system is a mixed state.e. For the case n we have TJ=J (520) (5. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights.20) expresses that the system can be found with probability 1 in the state \i). that — different from u>i — P.22) In the following (x\i) is always to be understood as (x\i(t)}. (5.3 The Probability Density 77 We see therefore that p defines the mixed states. which is contained in the coefficients Cai. If all ui but one are nought.5. since they do not take into account the (interaction with the) rest of the universe. and in the other states with probability zero. the only remaining one is 1.t) we now write tp(x.23) Instead of i(x. .24) s (x\x') = 5{x . We see this more clearly by going to the position space representation. Equation (5.t) = \^{x.t) = {x\i).x').8). represents a quantum mechanical probability. ^2ui\i){i\.e. In the following we are mostly concerned with considerations of systems in pure states.
The generalization to a threedimensional position space is selfevident: JK 1 / p(x. (5.27) is known as the Schrodinger equation. the expression (5. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).78 CHAPTER 5.27) is iHJ\ jt= m . We shall convince ourselves later that the postulated equation is sensible (in fact. (5. We therefore postulate first an equation for the time dependence of a state vector.27) where H is the Hamilton operator of the system.t)\2.3.28) Differentiating Eq.t) = / dx{i\x)(x\i) = (i\i) = 1. Equation (5.25) J Thus the particle whose state is described by \i). and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. (5. 2. t).8) and multiplying by ih. can be found with probability 1 in the space R1.8). 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. i.e. which is chosen in such a way that the desired analogy is obtained. (5. . the Schrodinger equation is always postulated in some way. We now want to obtain the quantum mechanical analogue.22) can also be written (xi) 2 = (i\x)(x\i).t) = \rp(x.26) 5. or the wave function tp(x. (5. (5. This differentiation requires the time derivative of a state vector \i) E "K. and hence the integral over all space dxp(x. but the linear operator states are moving — even if the system is observably stationary).^ The equation which is the conjugate of Eq. with respect to time.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. To this end we have to differentiate the density matrix p. Schrodinger Equation and Liouville Equation The relation (5. it is not "derived").e. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).
(3. we observe that here on the left hand side we have the total time derivative.24)). Eq. (5. We can now consider the Schrodinger equation (5.21). such a formulation is described as Schrodinger picture. i.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.e.p}. (5.32) Comparing Eq. both uji and i)(i).28) we obtain (see also Eq.p} + ihJ2^\J)(J\.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H. the relation ih^ = [H. p also contains "quantum mechanical probabilities" (i.34) . The reason for this is that in addition to the "classical probabilities".28) the states are considered as timedependent.32) with its classical counterpart (2. (5. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.27).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). not the partial derivative.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.37)) (5. With Eqs. dp/dt ^ 0. We thus obtain the operator form of the Liouville equation. (cf.5.27) and (5. Eq.Xi)\j). (5. i.e. (5. (5. (2. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.
**R.4. the state \i) is still timedependent. T meaning temperature. to) = exp[—iH(t — to)/H\. (5. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. i.1 Evaluation of t h e d e n s i t y m a t r i x As a sideremark with regard to statistical mechanics we can make an interesting observation at this point. In expression (5.and temperaturedependent Green's functions.** In view of the close connection between time.. i. (5. (5. which we shall need later. we write V>(x. „ / ith—. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x.36) in in The initial wave function or timeindependent wave function ^(x..80 CHAPTER 5.38) (x\j)t=o = Yt(x\E^(En\J)t=o. 10. P. a (5. in Sec.t) = exp ip(x. Feynman [94].36) however.8) for p. with respect to time t. n (537) This equation is described as the timeindependent Schrodinger equation.x 1 ^ ( x . . in p = ^2i^i\i}(i\. since the solution of Eq.t): > ih—</>(x.35) can be written'! ip(x. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. 0) or \il))t = \il>)t=o exp (5. it is plausible to refer to this equation at this stage.e. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. appearing in the Boltzmann distribution.e.Xi){x\En) = En(x\En). we shall describe as "time development ator given by U(t. or H\En) = En\En). since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed. With the help of Eq.4. t ) . however.t) = Hld . we can replace these states by corresponding timeindependent states. but with respect to the parameter /3 = 1/kT.0): where H(ih—.35) Here we can look at the Hamilton operator as a "time generator3''. operator" the exponentiated oper . 5. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1..
40).40) 1. we can rewrite p. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. (5.41 ) or with 4>i(x) := (x\Ei) this is p(x. Eq.x'(3) = Si e ~^W*V). (5. (1.^ n tne position space representation Eq.5.39) Without proof we recall here that in the socalled canonical distribution the weight factors uji (similar to those of the Boltzmann distribution) are such that (cf.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. i (5.e. (5.4 Schrodinger Equation and Liouville Equation 81 Example 5. i.44) r (TrePH)' (5.45) since on the one hand •PEi i i i and on the other hand .1). or ^ = ^ e 0Ei _m. (5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). we obtain p(x. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. x') = ^ i LUiMxWix'). (543) Since H(f>i(x) = Ei<f>i(x).x')^p(x. i ( 5 .42) Inserting here Eq.
X'.45) without normalization as PN(J3) := e~pH.51) where the subscript x indicates that H acts on x of PN(X.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.52) Hence this expectation value can now be evaluated with a knowledge of pN. [A. With Eqs. whose calculation for specific cases is a topic of Chapter 7 (see Eq. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).82 CHAPTER 5.51).51) is seen to be very similar to the Schrodinger equation (5.x. we obtain = HxPN{x./3).45).x'. Example 5.x'. (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.46) is pN(x. (7.55)).47) with respect to (3.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.49) In the position or configuration space representation Eq. Equation (5. (3). . Wilcox [284].35).46) we can write the expectation value of an observable <^Tr<„.47) (5. (5. This solution is the Green's function or kernel K(x.48) = 1. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5.L [B. B}} . (5. Differentiating Eq.x'.50) Differentiating this equation with respect to j3.4> = ^ g ^ . (5. (5. (5. A}} + • Solution: The relation can be verified by expansion of the left hand side.(3).45) and (5.(3) := (x\pN(P)\x') = (x\e~0H\x')./?). (5.tt tt For a more extensive discussion see R. Tr(pN) = / dxpN(x. [B. (5. (5. B] + L [A. M. Schrodinger Equation and Liouville Equation so that p is given by Eq. We now rewrite the factor exp(—(3H) in Eq. (5.
The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs. its quantization here in terms of quasiparticle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles. (8. that the oscillator. we shall see on the way.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. 83 ..53)).. Kim and M.51) to (8.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. 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. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. Noz [149]. like the Coulomb potential. E. Since the spectrum of the harmonic oscillator is entirely discrete. can — in fact — be quantized in quite a few different ways. 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). S. This is a fundamental topic since the harmonic oscillator.the present form of quantum mechanics is largely a physics of harmonic oscillators. * Comment of Y." Quite apart from this basic significance of the harmonic oscillator. However.
However. so that there is no ambiguity due to commutation. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). 2mo dx \ 2 2 (6. of which the only nontrivial one here is [q.p.p]=ih (6.84 CHAPTER 6. p^ih—.' ^mgLOh by setting and *~. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. like taking half and half. which is — in the first place — representationindependent. we can also proceed in a different way.1) for the hermitian operators q. Also observe that H does not contain terms like pq or qp.2) which for mo = 1/2.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. This equation would then have to be solved as a second order differential equation. hence we do not use a distinguishing notation.to = 1 reduces to the parabolic cylinder equation. For instance we can go to the special configuration or position space representation given by q^x. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . . the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. which is called Weyl ordering.2 The OneDimensional Linear Oscillator In the case of the onedimensional. h = l. in the present case + [EmooJ2x2)ip = 0. When such a term arises one has to resort to some definition.^ The first problem is to determine the eigenvalues and eigenfunctions of H. This can be done in a number of ways. To this end we introduce first of all the dimensionless quantities rriQUi fl q. Quantum Mechanics of the Harmonic Oscillator 6.
6) and p = y/motoh AAi =^.2 The OneDimensional Linear Oscillator 85 (p.10) then a = (a\A^A\a) = \\A\a)\\2>0. (6.14a) we deduce for an eigenvector \a) of A^A.C]. (6.34d): [A.13) ( ^ A ) i t = A\A^A From Eq.14b) [A* A. . C}B. Prom these relations we obtain ( ^ = ^ .1 ) . With the help of Eq.A^] = 1. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . (6. and [AB. i.12) in order to obtain the following expressions: [tfA. C] = A[B. (6. i.8) The eigenstates of H are therefore essentially those of N := ^ f A (6. q are hermitian).e.e. + 1).14a) (6.5) Reexpressing q and p in terms of A.6.7) (6.B}C + B[A.11) Thus the eigenvalues are real and nonnegative. if A*A\a) = a\a).9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. (6. A\ we obtain h A + A^ rriQLU y^ (6. C] + [A. A] = [A\ A]A = A.1) we obtain immediately the commutation relation [A. (6. (6. A*} = A^A. for ^ A  a ) = a\a). We now use the relation (3. A^] = A\ (6. (6.BC] = [A.
l)A\a) = A(A^AA . However.  a) is eigenvector of A^A with eigenvalue (a + 1). since this equation implies that the eigenvalues cannot be negative.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1).18) = '= a ) A(A^A . unless A 2 a) = 0. (6.17) (6. but (b) An\n>\ llA An+1a)=0. (6. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a.e. A 2 a) is eigenvector of A^A with eigenvalue (a — 2). (6. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). (6. In this case we have (A*A)A2\a) (6 (6. we must have (a) Let \an):= Ana)^0.86 the relation CHAPTER 6.2A)\a) = A(Aa .2)A a).l)\a) = A(a . in view of Eq.. or Ata) = v ^ T T . Similarly we obtain from Eq. (6. unless A\a) = 0.16) This means.19) i. This would mean that for sufficiently large values of n the eigenvalue would be negative. unless A^\a) = 0.21) .l)A\a).l)\a) = (a . or \\A\a)\\ = Va.A)\a) A{AAU . Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .2A)\a) 2 (a .20) . Next we consider the vector A 2 a).11) this is not possible.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). (6. If we continue like this and consider the vectors An\a) ^ 0 for all n. (6. Thus for a certain value of n > 0.
26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2. v.. \ I 2 2 x 87 Replacing in Eq. (6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.22) With relation (b) of Eq (6.n ) = " A»a>=» ' = A«a)F = 1 .I n U .2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n). I „ .. \ 9 \ I / II 4 T ) . (6. Moreover. The state 0) is called ground state or vacuum state. (6.' " /.27) According to (b) of (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. In the following manipulations we always use the commutator (6. (6.20) we obtain from this that the right hand side vanishes.17) a by (a — n).6.5) to shift operators A to the right which then acting on 0) give zero. i.20) we have for a = n = 1: A2\1) = 0. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. that an = 0. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.25) (6. For a — n = 0 we deduce from (b) of (6.28) .18) we obtain for a = n: p t  n )   2 = n + l. or a = n > 0. (6. From relation (6. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.n  a . (6.20) that A0) = 0.24) This is a very important relation which we can use as definition of the state vector 0).e.
i.e.11).Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.34) (6.^ ( O l A ^ n o ) . A^A* + A*[A.32) (6. A^} = 2A\ and [A. In general we have \n) = ^(A^n\0) (6.88 and CHAPTER 6.26) the equality l) = A t 0).33) . (6.11) we have [A. Similarly we find 2)oc^Ut0). The states \n) thus defined are orthonormal. as we can see as follows.5) and again (6. l>ocAt0). (6.29) Hence we obtain and in view of Eq. (6.32) we have (nm) = . (0^ 2 (At) 2 0) = 2. 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. (6.29). and in general [A. (A^)3] = [A. we have 2> = ^=dAi\0).30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). and in view of Eq. According to Eq. But using Eqs. (A*)2} = [A.(6.(A*)n] = n(Ai)n\ (6. (At)2]A+ + (A*)2[A. (6.31) (6.
37) n^)711}^) = v n  n . Here.'n\ and = v n ![ ( A t ) M i= + (6. . •(» (6. (6. . we obtain (n\m) = 5nm We also deduce from Eq.34) A\n) = ±=A{Al)n\0) .1 0).41) The contribution hu/2 is called zero point energy. With Eq. whose number is given by the integer eigenvalue of the number operator N = A^A. .1) (6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.2 The OneDimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An1A^mA\0) + (0\An1m{A^)ml\0} 89 0 + m(0A n . In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.8) we obtain therefore H\n) = hu (A^A + ]. in quantum mechanics. .1 ) . chosen in close analogy to the postulated "second quantization relations" of field theory. the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . 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 . i. (6.34) and (6.35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. l . we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle").38) (6. (6. In view of the properties (6.1) = n\n).35) Inserting this result into Eq. however.1 (A + ) m . (6.39) A^A\n) = y/nA^\n . The terminology is. .33).6. n = 0 . . . ldnm = n\5nm. 2 .40) (6. (6.J \n) = hw (n + i J \n).36) + 1) (6.e.32): rf\n) = .
and — in fact — we have used this already previously (see e.'\n) i'\A*\n) (n + l\A'\n) and similarly (n'An) (n — l\A\n) In matrix form this means 0 0 0 0 0 0 0 0 0 A/4 = = Vn + l(n'\n + 1) = \/n + 15 n / iTl+1 . (6. 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 (6.7) the energy representation of the operators q and p.e.43) = = y/n(n'\n — 1) = y/n5n^ni. (6. En = hu(n + ).42) We can deduce the energy representation of the operators A. Eq.90 CHAPTER 6. ( 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. A^ from the relations (6. Quantum Mechanics of the Harmonic Oscillator 6.g..6) and (6.2. i. (6. other elements zero..38) and (6. n = 0.1. (6.46) v^ .39): {n\A.. Vn + 1.45) Correspondingly we obtain with Eqs. also called Heisenberg representation..3 The Energy Representation of the Oscillator The energy representation. other elements zero.44) yfn. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn.. is defined by projection of operators on eigenfunctions of energy.39)). (5.
by (cf.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. i. mow / (6.6. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.1) and (6. A\Q) = 0. (6.47) LP 0) = 0 . The ground state wave function (/>o(x) is defined by Eq. 2 . Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w . (6.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. (6. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n).e. (6.5) are also satisfied as matrix equations.3)) ^ (q 2n \ + (6.32). 6. This is a simple differential equation of the first order with solution (x0) = The normalization constant C is determined by the condition* OO /C 1 = (00) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / OO J —< emou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero.24). Eq.4 The Configuration Space Representation We saw that the eigenstates are given by Eq.
56) < ^ = We have as an operator identity the relation . they are called Hermite polynomials. (6. it suffices according to Eq. Vn! Now {XlA (6.92 CHAPTER 6.dx'{X] so that <*'>^(irrv^i« mQU dx Setting a we have 7T 1 /4 v ^!2«/2 \a Setting a and ff„(0 we have = ^ / 2 u^n^m^''^ h mouj dx) K ' ( f .32) to apply the appropriate number of creation operators A^ to the vacuum state 0).l ) n f f „ ( 0 . (6.L ( x  ( A t ) »  0 ) . ) e"^2. so that <t>n(x) = ( . i.e.l ) > n ( x ) . c/>n(x) := (x\n) = . In order to obtain the wave functions of higher states.51) \l 2h {X q \ m0uP)\l 2h \X m0u. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. These polynomials are real for real arguments and have the symmetry property # n ( . (6.2 n n! The functions Hn(£) are obviously polynomials of the nth degree.54) (6 55) i/4 rzri7r 1 / 4 vo.0 = ( .
gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt. . — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.g.58a) ' T h i s definition — apart from the usual one — is also cited e.2)m = (±t)m for some function VKO.4 The Position Space Representation i.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .en_ dt. d£" (6. (6. Magnus and F. Oberhettinger [181].12. p. 93 .e.58b) by W. e. 81.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. (6. 2£.12(20 . (202"2. ( 2 0 4 . (2036(20.6.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.57b) • Generalizing this result and inserting it into Eq.
(6. Magnus and F. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. . (6. Oberhettinger [181]. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0.94 CHAPTER 6.58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists. (6.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. £.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.(e^/2) or Hn(0 = 2n/2Hen(V2ti).6 £ 2 + 3.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . £ 4 .3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) .59b) = (l)^2/2^. 80.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. £33£. (6.60b) J—< "See for instance W.^ + n) He n(0 = 0 (6. (6. pp. = nHe^iO.He'n{£). 81. = ffen+itO+ntfeniCO. £2l.
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£) = ( .. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.s) = eM?(Zs)2}(6. Their normalization is therefore given by OO /"OO / OO J —OO z (6. Magnus and F.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.l ) ^ e^HeniO 2 / 4  .61) e?l*2nl2Hn{Z). .63) The meaning is that if we expand F(£.s) = J2 n=0 (6.64) n\ See e. 93. W. p. s) as a Taylor series..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(0.6. and so on. Oberhettinger [181].g.s)=F(£. Thus the generating function can be written F(Z.^ = e^/42~n/2Hn(aV2). F(S. (6.
W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). for x(t) = 0 and p(t) = mox(t) — 0.e. 6. Quantum Mechanics of the Harmonic Oscillator Fig.e. In fact we can use the uncertainty relation AxApx ~ H/2. ground) state is fko/2. 2 . has parity + 1 .1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. 6. We also see from the eigenvalue t h a t the energy of the lowest (i. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. We see t h a t the lowest state is symmetric. i. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. i. in order to estimate the lowest energy of the harmonic oscillator. and we see t h a t even and odd states alternate.1.e.96 CHAPTER 6. In classical mechanics the energy of the oscillator can be zero. mow' Apx = ^m0Luh/2.
to 0. classical orbit i. n*h. The kinetic energy is always positive or zero. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q).. 27rmoi/ <f> V'a 2 — q2dq = n*h. According to "old quantum theory" with integer n. to —a.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. back to 0. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. which is the complete orbit. 14. In Example 6. Sec. In the plane of complex q the integral has no poles for finite values of q. E . Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways.plane pole at z=0 ^ .4) to obtain by contour integration the eigenvalues of the harmonic oscillator. But it has a pole at infinity. 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)). original path q .e. Example 6.1.6.plane Imz z . like twoatomic molecules.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf.1/a O l/a Rez all at infinity Fig. pdq. 6.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". and is directly observable. The analogous Example 6. /classical orbit Here q goes from 0 to a. in corrected version instead with n* = n + 1/2. Thus. mass mo. we obtain the integral .: 27 T mov2a2. This observability can be taken as direct proof of the uncertainty relation.2 The original path and the transformed path at infinity. setting q = 1/z.
6. are the most basic and exactly solvable cases in quantum mechanics. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .1.1 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.2: Contour integration along a classical orbit Setting q = 1/z. we start with a Laplace transform ansatz for the remaining part of the solution.V62 a2). Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. 1 .65) f ^Va2z2 . mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points).2b = Ze2. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. With a2 = E.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . hence we consider the following differential equation in which n = 0 . / " ^ . (11. in q at infinity).5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials.5.1)! (6. After removal of a Gaussian or W K B exponential.1. Example 6.5 . .114a)). but here we leave this . for instance — without deriving all properties of the solutions.3.( 6 .2TTI a ^ = .) hv. i.. harmonic oscillator and Coulomb interaction.e. z = 0 Vi 2Kmov(Ka2) = n*h = E/v. 2 .98 CHAPTER 6.e. . . E = Z2e4/4(l + n+ l ) 2 (cf. £ = n*/w = ( n + . and obtain with this all properties of the solution. i. (m . Eq. since these two cases.e. a J \d Solution: The solution proceeds as in Example 6.a) m = 2 . 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). « qdq + I2 = f dq . 6.c2 = (I + 1/2) 2 (natural units. In the following we investigate in more detail the important differential equation of the harmonic oscillator. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z .
(cf. (6. Eq. fp{ps{p)) = {v2 + n)s{p). (6. Oberhettinger [181]. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). W.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. This has to be true for all values of t.g. 80.6. p. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. so that the integrand of the remaining integral vanishes. Substituting these expressions into Eq.68) § = J p2s(p)e*dp.J ±(ps(p))e*dp. / dpf(p)e~px the Laplace transform of f(p).p t dp = \ps(p)ePt} . i. Magnus and F.66) First we have to remove the £2term. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. (6.e. . Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .g. and / dpf(p)pn the Mellin transform of / ( p ) .69) **See e.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n. (11. (this determines the limits).111)) the Mellin transform of e~p is n!. E.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . We demand that C(p) := — \ps(p)e~pt] = 0. (6. Then (with partial integration) *dt = ~*/Ps(p)e.i ^  (ps(p)) P n P Integrating this expression.t*ij& + ny{t) = O. .
= / • Im p Rep n not an integer Fig. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer.. (1) This will be satisfied if \p\ — ±oo. The value of the exponential at this point is ~ exp(£ 2 /2). 6. (6. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. as indicated in Fig.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. We now investigate possibilities to satisfy the boundary condition C(p) = 0.n+l between the two limits of integration. This can be attained if p is allowed to be large (t can be of either sign). 6. (3) We can choose a contour once around p = 0 in the complex pplane. n must then be an integer. This exponential assumes a maximum value when pt \—p2 = minimal. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6.71) . For singlevaluedness. so that y(t) ~ exp(i 2 /2). consider the exponential in the integrand.70) = 0 P.3. i. when p = —t (which is possible. Eq. we can choose a path from zero to infinity (the condition will vanish at p = 0).3 The path when n is not an integer. The quantum mechanical wave function is then (cf. Hence a possible path of integration > is from — oo to oo. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. then for large t we have y(t) ~ exp(i 2 /2). To see this.100 and hence CHAPTER 6. so that p cannot go to infinity. of course). We can show that if \p\ is allowed to be large. (2) If n < 0. Therefore we take ptp2/2 y(t) .e.
n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . ... 4 ..6. . pt p2/2 (6.P 2 / 2 00 C—tY ( . this becomes jfj. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt). We can also obtain the derivative form of this Hermite polynomial.if n is odd.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e .5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.3. . (676) . 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.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. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. 2 .5. if n is even and v = 1. We have 27T! Setting q — p + t. . In y(t) above.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( .l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.„y2 v l { ' ^ r (^) \ 2 ' where v = 0. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).
Pptp 2 / 00 /2 dp.3 and 6.102 CHAPTER 6.77) = t.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Starting from Heo(t) = 1.4. 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).75) as Hermite polynomials.1 '2dt using Hen(t) = (l)n27 r . Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. ( n .)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . Hen(t).59a). n + l) = l.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. (n. n — l)Heni(t).3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. one can construct the tower of polynomials Example 6. Solution: We have to evaluate °° 2 .it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.n— 1) = n. and elsewhere. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6. Example 6. in radiation problems.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. as in the normalization of asymptotic expansions of Mathieu functions). (6. (6. we obtain tHe„{t) (l)"n! ~ i. Reinterpreting the remaining integrals with Eq.e.n + l ) f f e n ^ i ( £ ) + (n. Hei(t) Henj. (6. 8. for the evaluation of expectation values.70). 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+<.g. Example 6.
Hence the normalized wave function of Eq. zero otherwise (6.q).q) = = e .78) \/27rn!.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2)./(n __ 2r — /i)!.n)\ i s an+2rij.82) ^ .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. (6. 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.^ + 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 .5 nm . g) is n! Next. so that /2 Q2(n+rrf in / ( p . and 0 otherwise \/2n(—l)n+mn! if n = TO. (6.80) f(p. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6.6. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.66) is \Zn\V2w Example 6.jQ) n n!I ^—' ^Q fi\(n .
v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n . Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).v)\(s .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . 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.2s _oo ( a 2 / 2 ) i . /i = (n — v) _ / r o o e .n + 2) + (n.(t)e«2/2dt Now.e. n + l ) ( n + 1.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . n + l ) ( n + 1.l . n + l)(n + l .A i ) ! (6. n — 1) are given by the first index. 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) + .79). the coefficient of T 2s »{sr u)\ As we observed. n + 2)]ife„ + 2 (t) + ••• . we obtain the result (6.e.e '' ^ o^M ! ( n . .e.' . i. the relation (n. n + 2)(n + 2. The result then follows with half the normalization and orthogonality integrals (6. Prom Example 6. n)(n.l ) ( n .3 we know that upgoing coefficients are 1 and downgoing coefficients (n. i. . n + 2) + ( n .u)\ 2s<—" v\(n . n + 2){n + 2. n ) ( n .77).r . n + 2) + (n.^ ) ! ( n + 2 r .104 CHAPTER 6.r . n .. n + l ) ( n + 1.ra + 3)(n + 3." ) a2^) r coefficient of (2s)! in f^ (s .A o?(. by n!(7r/2) 1 / 2 . 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 « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ .( n + Jr). n + 3)(n + 3.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . n + l ) ( n + 1.— — ^ .n+ l)(n + 1. i. i / ! ( n .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.i / ) ! ( 2 r + i/)! > . n + 4)_ffe n +4(t) + [(ra. 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 . K is the coefficient of a / ( 2 s ) ! in the expression J .v)\ ^ Multiplying both sides by the value of the normalization integral.^ / 2 * . n. so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.83) = (n + 2r)!e a / 2 V „ „. n + l)(n + 1.t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . n + 2)(n + 2.78).
x'). 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.Chapter 7 Green's Functions 7.1) be the timeindependent Schrodinger equation which has to be solved. in Feynman's path integral as the kernel or free particle propagator. This example also offers a convenient context to introduce the inverted oscillator potential. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. With reference to this case. in Chapter 21 (with a different calculation).1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. Let H^ = E^> (7. which is a point of unstable equilibrium for a classical particle. 7.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. which we write G{x. We consider first timeindependent Green's functions. and is thus an important point to be noted here. of a differential equation of second order. we consider the sojourn time of a quantum mechanical particle at a point. which is normally not discussed. These are the Green's functions of the timeindependent Schrodinger equation or.x'. more generally. and those which are timeindependent. The first of these cases will reappear later. We 105 . which we write here K(x.t).
e.g. the relation X') = 5{x . The timeindependent Green's function G(x. (7. i.x'). x') has to satisfy correspond to those of * (to insure e. (7. for instance.x') can be written t G (7. 'The case of £(°) equal to some Ei is considered later.106 assume a Hamiltonian of the form CHAPTER 7.x') or ^2^(x)^*(x') i = S(x . (7. since the generalization to higher dimensions is selfevident. (7. although the explicit calculations can be much more involved. The boundary conditions which G(x.g. Eq. of the kinetic part p2/2mo and some part of the potential.x').x') is defined by the equation* (Ho . a perturbation part like (3q4. where £ ( 0 ) = lim E. e.EW)xG{x. cf.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. i. We can readily see that G(x. and Hi is some other contribution. square integrability. We can obtain an expression for G(x. aq2.e. .4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . appropriately decreasing behaviour at infinity).x') by recalling the completeness relation. Green's Functions H = H0 + Hi.g. e.13).2) where Ho consists.
x')(E .10) The contribution on the right hand side is the socalled perturbation contribution. We can see the significance of the timeindependent Green's function for the complete problem as follows. The timedependent Green's function K(x. We see that the initial condition (7.0) = S(xx').E^ We can check this by considering (H0 .x'.t).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.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').x'.Hi{x'))^(x').9) = i H0YieE^i%(x)^(x') = H0K(x. (7. (JL x'\ t) = H0K(x. i (7.(*'))*fr') Hi{x'))^(x') f dx'5{x .E<®)xG(x. (7. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.£(0) .EM)X*(X) ( ..x'. x'\ t) (7. x'){E .11) = = 3) = fdx'(H0 . We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.x'.E^ of the eigenvalue. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E . (7.7.x'){E .) = f dx'G(x.t) since ih^K(x.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x). .x.E® (EE{0) Hi{x))m{x).t) (K for "kernel") is defined as solution of the equation ih—K(x. .#.7) with the initial condition K(x.
14) We can now see the significance of the timedependent Green's function K(x. see E. W.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).. (7. Wiedemann [183]. 2nd ed. . Instead of Eq.2 (not contained in the first edition!). (7.11) with a system of linear equations of the form Vi = Mijyj. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. Bleecker [33]. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). 17. that (E . J. MiillerKirsten and A. Green's Functions Can we add on the right hand side of Eq. Maharana. Merzbacher [194]. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. In general.108 CHAPTER 7. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation).11) are called (homogeneous or inhomogeneous) Fredholm integral equations.t) for the complete problem as follows. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. H. if we assume that E^ = Ej.e.15) ''This problem and its circumvention can be formulated as a theorem. The theorem is also known as Fredholm alternative. then at best such a contribution would be something like cipi(x). This will be different. Booss and D. (7.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. But then the Green's function (7. All such methods of solution are based on the analogy of Eq. integral equations are more difficult to solve than the corresponding differential equations. however. Examples in various contexts have been investigated by L. Sec.12) Equations of the form of Eq. D. But if we assume that E^ ^ Ei for all i. (7. In cases where an inhomogeneous contribution is given. see B.6) we would have (7. (7. i.6) is not defined.x'.
x'.158.g.t) = ^ e £ " f / « V n W < ( ^ ) . 2nd ed. 7.22) According to Eq. (7.21) dx'K(x.7) with Eq.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. *>0.The solution of Eq.' which obviously satisfies the initial condition K(x.7. Eq.0)eEnt/in. 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. E.16) As usual. 109 (7.. Merzbacher [194]. (7.t) = J2 fdx/^n(x)^*n(x')^(x'. t) with the Note that when K(x. (7. e. for the oscillator potential. x'. \En)(En\ip)t=oeE^\ n (7. p. i. in different formulations. this relation provides the probability density \ip(x. We use this in Sec. (7. We can write this expression also as§ ip(x. Comparison of Eq.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.x'.t)il>(x'.17) is the initial condition of \4>)t. . 1 See e. t > 0.2 for the computation of the sojourn time.15) which contains (7.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7.19) i/>(x. (5.t) is known.x'. t)\2.g.17) as initial condition is obviously W)t = Y. (7. (7.0).5. (7.Q) = 6(xx').t) describes the evolution of the wave function from its initial value ^(a^O). H\En) = En\En).20) is the timedependent Green's function.t) = where if(x^'. (6.18) or. x'.e.20) the Green's function K(x.
We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.x'.x') and K(x.ie (7 . In the following we shall derive the Green's function for the case of the harmonic oscillator. 24 )  ReE Fig. between (with E^0' = E) G(x.t) K(x.23) along the contour C in the plane of complex E as shown in Fig.x'. we obtain / W : = .«^«2 W .t).1. = J2 ^„(s)C(*') En — E We see that G = GE depends on E. 7.e. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. t > 0.i /f e «/. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. (7. we obtain then also the corresponding density matrix. x')6(t) (7. 7.25) in agreement with the timedependent Green's function . i.110 CHAPTER 7.E .1 The contour of integration. x'.x') = GE(x. Green's Functions difference that (5 = 1/kT plays the role of it. Inserting for GE+ie the expression above. t) = J2 e £ " f M i ( x ) < ( x ' ) . Jc 27T „ hn .
(7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7. It is clear that it is nontrivial to solve an equation like Eq.7. v ' 2m 0 v h i.31) The constant A has to be chosen such that K(x.e.30) into Eq.x>.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.7). A and B being constants.29) and (7. K(x.26) This is the case of a free particle with mass mo. B m0 n _ 2ih 2iK (7.e. .t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.28) B(xx')2 t2 DB(xx') 2 /t (7.x'.3 The Green's Function of a Free Particle 111 7. which is moving in one space dimension.X>. i.— d2 —2K(x. we obtain *I£I = h2 —(2AB).t). (7. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.x'. / dxK(x.t) t h2 = .29) tJJ 2 2B(x .0) = 5(xx'). and identifying coefficients of the same powers of t on both sides. (7. 2mQK h h2 ih(AB) = —{4B2A). In this case the Green's function is the solution of the equation d hK(x.x'. (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. Thus we try the ansatz.27).0) = 1.30) Inserting Eqs.
(7.zEnt.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. (7. we have This is 1 provided A=JB= ^ (7 . 2m0 (7. we > have to make the replacements V ^ fdk. from K{x.x'. (7.38) ..t) into Eq.oo 27T (7.t) = — / dkeak2+Pk 2vr J.(7. v27r En^^. "See the excercise in E. (7.21). (7."737) \27rii/i in agreement with Eq.e. 1st ed.34) K(x.H We can insert the expression for K(x. (7.t) — for instance for a wave packet given at time t = 0 of the form ^(x. *>0.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. Merzbacher [194]. 158. Mx)^M^) „ n so that J =^ . 32) It follows that x'.0)oce"Qx2+ifcox.ihMxWn{x'). i. Green's Functions For parameter values such that the following integral exists.35a) K(xy.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.112 CHAPTER 7. p.20) and can then obtain ip(x.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS .33). j3 = i(xx'). n For a free particle moving in the onedimensional domain \x\ < L — oo. t) = /jJo_ e mo(««') 2 /««.x'.x't) = Y.
HQ = ^—p2 ZrriQ + \m0uj2q2.t) We now set h2 d2 = .x'. mQLu .t) 1 + m0u2x2K(x.x'.7) is the solution of d ih—K(x. i. x'. We rewrite this initial condition in terms of £ and use for a = const. (7. in this domain *Recall that S(x) = ± f eikxdk.e. / ) = ~^K(x.e.0) = 6(xx') at / = 0.4 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.8).37) will later be obtained by a different method — see Eq. x'. i. —ih—K(x.33). §jK(x.. (21. . K(x.x'.x'. 1 2mo u>n ox n with the initial condition (7. / ) (7.e. (7. x'.x'.40) becomes 2 .39) In this case the timedependent Green's function K of Eq. ./ ) + fK(x.——K(x.40) Then Eq.X .t).d r. . . (7. 5{ax) = ± J eikaxdk = ArS(x). . I (7. . —KK(X..3 The Green's Function of the Harmonic Oscillator 113 The result (7. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.e.7. i.42) .x usn ot i. the relation* 6(x) = o<y(os).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.25) — in the context of Feynman's path integral method. z ..t). (7.t) + ~x K(x. (7.t) = h2 2 d2 TW . 7.
4 a 2 .114 CHAPTER 7.Z'.e. it • is suggestive to attempt for K the following ansatz^ m. 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.47) becomes (7. (7.48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7. Identifying coefficients on both sides.b2.4a 2 )£ 2 .t) become similar to expression (7. we must have /o = 0. W ) .4a&£ + 2a . so that for / — 0: > a^—.42) and obtain the equation (with a1 = da/df etc.46) If we interpret Eq.f) with a(0)»y.45) in this sense. (7. (/ = 2a. Integrating Eq..e. 50. (7. Feynman [94]. b' = Aab.^ . i.b2.48). and we expect K(x. .c o t h 2 / .47) We insert this ansatz for K into Eq. that is.50a) (7.) a'£2 + b'i + c' = (1 .33). (7. p.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7. Green's Functions to the particle behaves almost like a free particle. (7. we obttain a' = l . (7.50c) (7.51) See R.50b) (7. as the limiting case / — 0. c(0)^.x'.45) in accordance with (7.50a) we obtain (7. a = . f i.
55) 'For an alternative derivation and further discussion see also B. if we return to x. we can use K(x.48). (7. and to ensure that we obtain the prefactor of Eq. p. x'. (7. we must have (besides A = —£') = /m 0 u.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq.x'.x'. K(x.b(f).£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. as we shall see in the following. / —————rexp y 27rmsin(u. as one can verify. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. Felsager [91]. 2TT^ Inserting a(f).7. (7. (7. 174.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. (7. B independent of / .5.i) (7.27). . i.t) to obtain this element (x. (7. (7. t. (7.53) or.0).50b) Kf) = —r^77) smh 2 / A independent of / .48) 6(0) = — £'/2f. Finally Eq.45).e.54) For t —• 0 this expression goes over into the expression (7.50c) yields for c(/). c(0) = £' / 4 / . Comparing the Eqs.33) for the Green's > function of a free particle.* With this result we have another important quantity at our disposal.c(f) K = 5 =exp Vsinh 2 / into Eq. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . i.X'.In B.2. with A. 115 To ensure that in accordance with Eq. x .e. in particular for the derivation of the sojourn time in Sec. (5. we must have A = ('. In order to satisfy Eq.40) of the timedependent Green's function with Eq.t) = . 7.51) for the density matrix PN(X./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.
For instance we have with Eq. Eq.. ^—• Irpiv J pN(x. (5. (7.P) 2.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.x.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. at temperature T): {A) = Tl{pA) = ^ > . Eq. we obtain § < Z> = ^ c o t h ^ ^ ° — . eP E i = i + y.p)dx W ) = —^r = —p .x.55).e.uij>0 >• 0. (5. we skip the algebra here.x.52)) the expectation value of an observable A in the canonical distribution (i.e. P. 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"}.52): Tr (PNq2) Jx2pN(x. (5. <2 (7. R. = jdxp(x.40)) 1 epEi 1 Ui = y..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. )x Inserting into Eq. [ [ fdxdx'dx"(i\x}(x\p\x')(x'\q2\x")(x"\i) Idxdx'dx"{x"\i){i\x){x\p\x'){x'\q2\x") J2 [ [ i = fjf^»(x>)(xW)(xW) dxdx'dx"5{x" i. . 52.» l. Thus the system is in a mixed state and the expectation value "Cf.56a) {q2) = = Y. Green's Functions With this expression we can evaluate (cf. p.116 CHAPTER 7. 11 For T ~+ 0: u>0 . Feynman [94].
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. In this case the Green's function K(x.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. e > 0. (6. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . If we consider the system in the pure state \i).J (w — ie)< e (n+l/2)Et e i(n+l/2)u. t) of Eq.56b) is that with respect to this mixed state (whose cause is the finite temperature T).QUJ (7. We return to the Green's function (7.x'. 2 + x'2)le^2xx' to—>w—it t—>co 2 . /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . and we replace to by LO — ie.50) we obtain. x'.21). i. (7. 1 mo^ exp lmow / 2 2 ft X (7. We assume t > 0 and t —> oo. of K(x.54). This is the first and hence dominant term of the expression (7.7. x'. setting a = mooj/h.59b) n+ l \ ftwt here 2hf exp _ i ( n + .3 The Green's Function of the Harmonic Oscillator 117 (7. the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m. which means in the oscillator state \i) with eigenenergy fku(i + 1/2).59a) E 0 = ^fiw.54) is K(x.t "With the normalized ground state wave function of the harmonic oscillator given by Eq.e.
In Example 7. the particle will stay there only for a finite length of time T.59a). which are planes. In a very analogous manner we obtain the solution of the equation for the density matrix.5. (7. the velocity of planes of equal phase. k = k. i.e. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. and with this we estimate the sojourn time T.. 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. (7.118 CHAPTER 7. lie on surfaces.x'. to infinity) the contribution with n — 0 dominates. i. is defined as vv = . (7. 7. of dpN as pN(x. The wavelength A is given by The phase velocity v<p.62) .(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. Considered classically.1 we estimate T semiclassically.e. Green's Functions For t large (i. However quantum mechanically in view of the uncertainties in position and momentum.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0.e.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)].61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. as in Eq.
k ') 2 .7. (7. ^ ( r . ' t) dk'. We now ask: How and under what conditions does the time variation of the function ^(r.)¥'(*:")) +ism(<p(k') .)ll/(fc")l[cos(¥'(fe. (7.<p(k"))}dk'dk". i. i ) = /*/(k')e i(k '" r '' . t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. the fundamental postulate on matter waves in Chapter 2): E = hu. to the function ip given by oo / fik'y^'x^dk'.e. The wave packet describes a wave of limited extent. t) is the spatial Fourier transform of this Gauss distribution — as we know. / oo J—oo . essentially again a Gauss curve. a > 0 .e. duo' da . oo (7.63) The relation u = w(k) is known as dispersion or dispersion law. If we assume for / ( k ' ) a Gauss distribution. or dco' da The centre of mass determines the particular phase. e . One defines as centre of mass of the wave packet that value of x for which d(p — = 0.Jt + a.64) where / ( k ' ) differs substantially from zero only near k' = k.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. then ^(r. for which \ip\ assumes its largest value: OO . xt— + — = 0. i. e. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. i. u = w(k).Q ( k .65) Let f(k') = \f(k')\eia and <p := k'x .e. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k.
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. = — (768) It is instructive to consider at this point the following examples. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v.fc w> vv = ~a^ = S r a d P E «i7 Sd u . Example 7. .e. Solution: The function to be calculated is the integral dxeax OO ezkx. 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 = . . p = hk the group velocity is equal to the particle velocity v.. The threedimensional generalization is evidently du . we can use a simpler method.— g(k). Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. In the present case. (7. (7. For E = hio. however. i. a > 0.120 CHAPTER 7. as claimed for =0.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . dE v 9 = .^ g=r a rad. we obtain the deBroglie relation p — hk. Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const.
(7.r „• With the help of Eq.72) Solution: From Eqs. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. Example 7. (7.7. or see H.73) 27T 7 . XZ (e > 0).7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.. 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^.Afc = 81n2. (7. W..70) we obtain 6(h) = lim — f°° dxeax2eikx = .1 verify the uncertainty relation Aa.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .. with partial fraction decomposition.0 0 7T 7 0 7T e From Eq.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. (7.71) (7.74) Example 7.. Appendix A. (7.2: Representations of the delta distribution Use Eqs.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a . where c = 9(0) is a constant.75) ' E ..73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . (7. J.69) and (7.&)] for a + b. g . We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 .a) + 6{x . \a — b\ (7. 2 f + . (7.69) and (7.
2 The Gaussian curve. In quantum mechanics the square of the modulus of the wave function ip(x. A simple calculation yields A i = 2v/2~In2a.77a) g(k) = e . According to Eq. 7. i. Thus a slim maximum of the curve of /(a. The breadth A/c of the curve g(k) around k = kg.70) the Fourier transform of f(x) is (7. (7.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7.r)1/2a Re f(x) Ax Fig. requires according to Eq. at time t = 0). Physically . where g(k) = max<?(fc)/2.77b) (7.e.« 2 ( * .t) is a measure of the probability to find the particle at time t at the position x. is Afe = 2 V 2 1 n 2  (7.77a) small values of a. The wave function corresponds to the function f(x) in the above considerations (e.e. (7.) leads to a very broad maximum of the Fourier transform \g(k)\. when Ax is very small.78) It follows that the product of the uncertainties is AxAk = 8 In 2. i. Green's Functions 1/(27i)1/2a — 1/2(2.CHAPTER 7. and hence implies large values of Ak. Solution: In Fig. 7. the width Ax is a measure of the uncertainty of the probability.* o ) 2 / 2 . where  / ( x )  = m a x  / ( x )  / 2 .g.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. Thus a sharp maximum of the function f{x).
81) *G.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.". The result (7. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. 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.71) > and (7. Thus G.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} . This means the particle is localized at x = 0. Barton [15]. the less precise is the determination of its associated momentum. For reasons > of simplicity in the following we set in addition t mo = l. ..e. (7.54) these parameters appear in the combinations moui/h (dimension: l e n g t h .* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. 7.h = 1. nr = S(x). (7.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . Barton mentions at the beginning of his paper.2 A particle's sojourn t i m e T at t h e m a x i m u m The very instructive topic of this subsection has been explored in detail in a paper by Barton. In the same limit the momentum uncertainty Afc grows beyond all bounds.u> = l. which means all values of the momentum are equally probable.5. O ) .x'. E. 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.«> = / ^ C ' W .2 ) and U)t. ^In Eq.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. so that Ki0(x.7. i. <7*» We assume that initially. at time t = 0. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. The function g(k) is therefore described as momentum space representation of the wave function f(x).79) According to Eq. In the limit a — 0 we obtain from Eqs.. (7.54) with the substitution ui — iu. Barton recalls that W. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. (7. (7. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.
. Green's Functions Inserting (7.t) = J oo P ..84) we obtain* ip(x. (7. = .(i/2) sinh 2t 2 2 2C 2 (i) + *c^.82) i/f(x. ' 2\Aswht % cosh t (7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = . 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 .83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.2 (7.124 CHAPTER 7.86) Evaluating the Gaussian integral of Eq..81) and (7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.cosh t sinh t' Wisll2b ^ (7.83) ip(x. 9 (7.t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.88) . A .80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .79) into (7.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.85) = —r= o2 7 2 sinh 2i. + cosh 2 1. (7.
since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.t)\2 Ji == 71 / V " Jo dO ? 6' (7.t)= where Co(t) / dx\^(x.92) The sojourn time T is now defined as the mean time T given by T:= r*° ioo 2 (M)s r*(f4 d^e.94) / dxf(x)=g'(y)f(g(y)).t)\' 2 125 (7.£ 2 (7. C2(t) (7. •&(t)=//c(t) = / or.7. We have therefore . t) the phase ip drops out and one has (x. _2 1 exp C 2 (i) = l + 2sinh 2 i.90) with the expected limiting behaviour \im \ilj(x.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .4 The Inverted Harmonic Oscillator The real part of Eq. (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.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.89) = exp[x2/C2(t)} 7rV2C(t) (7.95) .91) C(t)' (7.. In taking the modulus of ip(x.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).
V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. p. 136. Dwight [81]..From this we obtain C(t) C'(t) Z2 ^(r?2_.=0 V^ C'(t) I Jt=0 V^ (7.98) We insert this into Eq.99) .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. C(t) = . £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.(4 2 .1 ) 2 \n J ^"•""ff^^V^K?.126 We set I C(t)' so that dT ~d~l Now.= \ / l + 2sinh 2 t. B.. (7. we obtain dT _ 1 ~dJ~l 'See for instance H.2)(p_7?2)(7.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.96) °° l Jr.97) sinh 2 1 = . + 0(e'2) (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. It follows t h a t T~lnZ. U) (7. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down.4. this is T ~ ]n(ly/mou>/h). Fig. p. 7. See e.7.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.100) Reintroducing the dimensional parameters we had set equal to 1. also motivated by this paper. 238. Explain the parameters entering the calculation. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon".102) 'As a matter of interest we add that in the theory of a free scalar field x. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. this equation is the equation of motion of this scalar field with negative masssquared.g. .3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. The equation there describes the classical "rolling down" of this tachyon. T h e following Example 7. Zwiebach [294].4 The Inverted Harmonic Oscillator 127 for I sufficiently large. B. (7.
(10. i. (x 2 ) h 2mou 2mou> and so l^J^ULl). (p2):=mW(x2).g. 2V ' Let A and Bio be the values of x and x at time t = 0. / 7 ^ 2 =2J(x }. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. the smallest macroscopic length. A p are defined by the mean square deviations given by (see Eq. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. (7. For h —> 0 the time T —• oo in agreement with our classical expectation.128 with x2 = LO2X2 + const. v A x A p = \ (x2)(p2) v = m0u(x2). (7. Thus quantum mechanically x(0). but instead are subject to an uncertainty relation of the form AxAp > h. CHAPTER 7. Eq. so that e. and for simplicity we take (cf. i. so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*).p(0) cannot be determined with arbitrary precision.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. and solution x = A cosh Lot + B sinh cot.e. Ap= v /(p 2 )<p)2. 2 10 x(0) = B u . Green's Functions x = Aco sinh cot + Bco cosh cot. .e. 2 x(0) ^ where A i . For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0).102)) p(0) = ^/(p 2 >. Quantum mechanically x(0). x(0) = A. ^(x2)(p2). Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T.5)) Ax = <J(x2){x)2. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p).
For mathematical purists the question of convergence of the series is an even bigger challenge. In general perturbation series do not converge. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. 129 .e.Chapter 8 TimeIndependent Perturbation Theory 8.2h2 cos 2 a # = 0. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). in closed form) only for very few potentials.E^ is already a bigg problem. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.e.(i)+/^(2) + ". one would set tt E = = ^(°)+/ty. 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.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. i.
h hl are in the strict mathematical sense divergent. see e. . the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. E = E^ + h2E^+hAE^ 71 72 7 + . 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.• + (ir~ X X* 1' 2' r?l Xn + Rn(x).3) *See e.. It will be seen later (e.2) n—»oo n It is interesting to compare the behaviour of the series (8. Whittaker and G. In the following we explain the difference between convergent and asymptotic series. ^For further details. TimeIndependent Perturbation Theory In the case of this equation. p. (8. 581. W.130 CHAPTER 8. which is known as the Mathieu equation. T. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x.* i. This convergence can be shown with D'Alembert's ratio test since''' lim . Cll E = a_ 2 n + a_i/i + a0 + — + —r \ .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 expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence.g.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 .g. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. Meixner and F .. E.= 0 < 1.2 However. 8. at the end of Chapter 26) — and by comparing the results of WKB.g. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation.+ . e.e. . 0.. the expansions in descending powers of h2. concerning also the uniform convergence of the exponential series.. Schafke [193].g. J. (8. Watson [283]. N.
.000 000 06 + • • • . .0.e. . the better the approximation obtained. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. Whittaker and G. We observe in the first place that since lim n—>oo co (8.5) Cf.8.0. T.. after reaching 6/27 = 0. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. N. we observe that the individual terms of the asymptotic series have the form of a factorial divided by the power of some parameter which is large.2.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.. begin to increase again.3) with the convergent series (8.... . +0. + •••..1). the series (8.0. and we see that the moduli of successive terms first decrease and then..29629. + 0. . / ( I ) = 1 — 1! + 2! — 3! H • However. e. Comparing the asymptotic series (8.001 + 0. and in fact increase indefinitely.4) for every value of x.000 002 00 . E.9876. for x = 3.. .222.3) diverges for every value of x. Watson [283]. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. i.4938.g.33333 .l n ( l + ^ ) n=l L ^ ' (8. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . p. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20.376 . It is evident that the larger the value of \x\. However. 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. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .3) can still be used to obtain almost correct values of f(x) for x sufficiently large.22222 .g. /(1000) = 1 .22222 .. + 0.0. We can see this as follows. .. 235. Thus e.. for larger values of x.0. in spite of this the series (8.
1 between Eqs. Whittaker and G. it is sensible to make the assumption. 371. that E and \ijj) (i. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. meaning convergent. In many respects Messiah aims at more rigour in his arguments.41) and (16. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. § However. p.1. In fact. ^E.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. Watson [283]. TimeIndependent Perturbation Theory where 7 = 0. N. ip) can be represented by rapidly converging power series in e. Schiff [243]. p. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small.5). Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems)." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. Merzbacher [194]. although this has not been investigated except for a few simple problem^. the dominant approximation is. Vol. 236. Messiah [195]. Sec. 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. I. p. This latter observation hints already at the importance of asymptotic series in applications. 11 L. in fact. T. 16." These "rapidly converging power series" in physical contexts are most likely very rare cases. . On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. 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. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. This is not appreciated by mathematical purists. the leading term of an asymptotic expansion. E.132 CHAPTER 8. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z.J In2. 152.5) are practically unknown. (16.e. whereas applications of the convergent series (8. It is interesting to observe the comments of various authors on this point. **A. 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.
Dingle [70]. II.. which have been studied in great detail." Of these three authors t h e first.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . In fact.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. (8. B... (8.7) as / e'* dte* dt = 1 .2. 8.^=e~x2 / e^^dt. Replacing t h e exponential in Eq. It is therefore suggestive to write iy poo <t>(x) = 1 . Vol.* and we follow some of t h e considerations given there. p.7) by its power series and integrating term by term one obtains t h e absolutely convergent series Considering for t h e time being only real and positive values of x. (8. Messiah [195].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. He says.7) This function has been considered in detail in t h e book of Dingle.11) A . ft Jx du = 2tdt. there is no need for such bias. .= / e~t2dt.10) Vn Changing t h e variable of integration to u = t2x2. we can rewrite Eq.8. *R. Merzbacher. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. 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. 198 (German edition). (8. (8. We know from books on Special Functions t h a t all of these functions.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. 2 (8. solutions of second order differential equations.
^ X e U ~ ( 1 + * \ 2 . (8. « \~2 =E n=0 l 1 V5F(ni)!(l)n nlTT (M x2) / « r oo 1+ For the domain \u/x2\ <f>(x) U \2 1 \/K n=0 2< n! V (8. since (—4)! = y/n. 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.^e~x \Ar i. Using the reflection formula (*!)!(*)! = for z = n + i .' we have 2 nH 7T 7T (8. du.14) sin7T2.12). 7T (ni)!sin{7r(n+^)} ( „ . TimeIndependent Perturbation Theory <P(x) = 1 .134 we obtain CHAPTER 8. (8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.i)[(_!)«' Then.13) presupposes t h a t < 1.16) . we obtain 1+ i.e.e. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .15) < 1 we can insert this expansion into Eq.
We now want to relax this condition and allow for phases argx ^ 0.l)! = / Jo Then 2 cHnxdt.e.e. We now see how the asymptotic expansion (8. (8.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. and in fact so slowly. (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.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. It is fairly clear that the above considerations remain unaffected for argx < 7T.8. the asymptotic expansion of the error function is.e. much more useful than the convergent expansion. convergent) only in the restricted domain u < x2.19) originates. ' arg.15) into (8. (8. In an asymptotic expansion the first few terms successively decrease in magnitude. /•oo T(n) = (n . We can actually understand the deeper reason for this in practice. In the foregoing discussion of the error function we assumed that x was real and positive. Frequently "=" is written instead of "~".19) approximates <f){x) the better the larger x is. the fact that we effectively use a binomial expansion beyond its circle of convergence implies that the resulting series is divergent so that even if the first few terms decrease in magnitude the later terms will increase eventually as a reflection of this procedure.r<7r.17) 0(X) = i. whereas in a convergent expansion the terms first increase in magnitude. Whichever way we look at the result.18) or that we insert the binomial expansion (8. As in the case of the gamma function. The expansion (8. . argx = 0.12) ignoring the fact that the latter is valid (i. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. i.—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^. i. (8. in general.
22) we obtain 2 4>{iy) = . dv = 2sds = .^ = . (8. (8. VK Jo V^ Jo (8.21) = y2s2. we can try to proceed with Eq. (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). thus with \v/y2\ < 1. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. we set x = iy in Eq. The situation becomes critical. i.vds.26) or else return .7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.2 y V . We therefore consider this case in detail. (8.15).^ f V e ^ . V7T Jo Changing the variable of integration to v (8.23) Throughout the range of integration 0 < v < y2 the integrand is real. when argx = 7r/2. however. (8. (8.136 CHAPTER 8.24) We can read off the binomial expansion of the factor in the integrand from Eq. =dv.e. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 .20) where we set t = is.
We have therefore <j>(iy) = ^e^TL fV e~v ( l .24). If we proceed with Eq.8. Suppose now we consider Eq. (8. (8.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv.28) n=0 We observe that this expansion differs significantly from (8.28) because then the correction term oc f°£ becomes smaller and smaller. (8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . i. ey is much larger than something of order 1/y.. (8.2 yir (8. Eq. e ^ ( i y ) ^ ^ .e.27) Then . i. (8. the better the approximation expressed by Eq.2 Comparison of Asymptotic Series with Convergent Series to Eq.28) ignores the integral contribution of Eq. y y 2 ( n _i)! y arg(* = iy) = i * .27). and the larger y. Returning to Eqs.19). . — (8. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 .28) we observe that the expansion (8.29) where 5t means "real parf. However.27) and (8.^ ^ p .24).^] * dv.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . i (8. (8.30) yvn f Jo V y J Cf.e. for large values of y. (8. and so the integral is real.17): J£° e~vvndv = n\. which will be studied in more detail below. 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. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. We know that for v < y2 the integrand is real. (8. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions.
is obtained for <f>(x) with arg x = —ir/2. and • higher order terms of the associated series all have the same sign. . B. e xZ ^ (n — £)! „ ^ ) ." — E ^ ^ forarg. 7r = .33)) would have • a maximally increasing exponential e+x .2 OO rz=0 V .e. We have therefore found that XTT n=0 ^ ' and . TimeIndependent Perturbation Theory If we insert here the expansion (8.e. y.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. (8..25) and integrate from 0 to oo.31). The corresponding phase lines are called Stokes rays. Looking at Eqs.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. . R. Eq. With a similar type of reasoning one can show that the same expansion. (8. (8.31) and (8. Chapter 1. (8.26).28). Since (cf. This result is identical with that obtained previously. since (8.3!) which is (8. *Cf. (8. Dingle [70].33) The sudden disappearance of "1" at argx = 7r/2 hints at something like a discontinuity of (j>{x) at arg x = n/2 which was discovered by Stokes and is therefore known as a Stokes discontinuity.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. i. (8.138 CHAPTER 8.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.
Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o .rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8.8. though. 1/1 ipv{x) v1 vl (±ix). —1 cancel out. i. 9 .13. Magnus and F.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. .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.34) the contributions 1. 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. Dingle [70]. W. (6. in the sum of the right hand sides of Eqs.2. and so may not be universally applicable. These rules. pp. Chapter 1.37) Cf.32) and (8. p.e. for arg x = 7r/2. apply predominantly to asymptotic series of the solutions of second order differential equations. i. Eq. (8. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. '"Cf. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. 5. See also Tables of Special Functions. B.2). which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). Equation (8. e. R.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. Thus if we set x = XR + ixj.35) 8. Oberhettinger [181]. 91.g.e.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x.
(8. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. 8. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a.7. Chapter 17. > > The equation (8.e. Chapter 1. with a real proportionality factor. R. 0 < argz <  (8. Eq. TimeIndependent Perturbation Theory i. i.e.140 where^ CHAPTER 8.36) is invariant under these replacements. Dingle [70].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. . (8. Dingle [70]. B. Chapters I (in particular p. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign.ei<v+VxvHv{x\ argx = £ . Hence Dil){x) = xv(j>v(x). We consider Dv (x). 1 1 Cf.x 2 oo «*)  — E^fi=0 v .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. R.40) § Cf. (14). We also observe that the late (large i) terms of the series in (8. v — — f — 1. Thus D{J\x) (i.39) (as in the example of the error function). = 7r/2. i. The continuation of (8.37). B. x~l/~1ipu(x). 9) and XXI. ti asymptotic series (8.38) are of the type (2i)! i\ i\(2x Y 2 (2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. the series of D ^ x ) develops an additive contribution (the discontinuity) which is IT/2 out of phase with xu4>v{x) and proportional to the associated function.e.
41) xuct)v{x) + a{x)~u~lipu{x).42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). sm(iri/) = —a/3/2 or a/? = 2sm(irv). e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. i. = ir. we must have at arg x = 7 in the dominant factor on the right of Eq.8. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a.41) the Stokes multiplier a is.e.e. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real. On reaching ir the part containing ^ ( x ) . = ir. another half of the discontinuity appears on leaving the Stokes ray on the other side.e. < 7T.41) to arg a. (8. where in the first line on the right hand side the second term contains the real proportionality constant a/2. (8. In (8. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. as yet an unknown real constant. (8.40).44) + ^iap) = 0 . i. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. < 37r/2. (8.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. with \x\ = (—x) for arg a. Since Dv (x) is real when x is real. for xj = 0) and therefore possesses a Stokes discontinuity there.e. The value of Du (x) on the other side of the Stokes ray. beyond arg x — 7r/2.42) T ^(eiwu 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). We therefore proceed to continue (8.
(z)\(ziy.v .42)) for argx = ir (in the second line of Eq.1) = ±a(y).45) that af3 = 2sin(7r^)) D^\x) = (8. i. . a{y)P{y) = a(u .e. We can see that the equation is satisfied by (3{u) = ±a(u . Hence the left hand side must have the same property.47) Equation (8. 8. (8.e.l)0(v .50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. (3{v .51) . Dl \X) is (cf.1) = 2sin7ri/. (8. (8.45) is therefore given by a{v)a(v .48) We compare this with the reflection formula (8.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .42) we insert from Eq.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . Eq. (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. TimeIndependent Perturbation Theory Now a and (3 can still be functions of v.142 CHAPTER 8. (8.—^ ON • — ^ > + ^ ^ L .1) (8.46) For a given function a(y) this equation determines /3(u).42) that a multiplies ipv(x).e.1) (8. (8. i. i.1)! TT. V i=0 \U0 v 2w .14). a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . i. argx = vr.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1. (8.1).e.
' ^ (i/l)!z! ^ . 8. Eq. (8.e.1 ) ( . i. (8. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.7r).e.35). ' (X) (cf.52) (argx = 7r). However.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. u 1 y> {2i + v)\ 1 n 0. (8. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.36)) in the domain 0 < arg x < IT. (8.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8.A / ± sin^Cx)"M* 2 V ^ .e.8.. 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.50a) because this enables us to define normalizable parabolic cylinder functions. the second contribution vanishes and we obtain £>«(*) ..53) We have therefore obtained in a natural way the quantization of the harmonic oscillator..1.. (8.*"ei°> f ( ^ _  _ L _ = (!)««(.„ _ix2^(2ivl)l e 2 > —.2. solutions of Eq. i. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.I / . Chapter I. the error . It should be noted that in Eq.7).)(.) (8.e^ .36).1 ) ! 4~L i\{2x2Y i=0 V2^(x). which vanish at x = ±oo. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0. For details we refer again to the book of Dingle.. —f.. i. (2x2)1 = COSTTU(X) v .
0. Solving these equations we obtain = = 0..54) we consider again the important equation of the harmonic oscillator.54) (8. (8. x—>±oo (8.59) and equating to zero the coefficients of powers of x2. (8. and demonstrate. x.60) 2 x x l Inserting (8. (8. \ + U (8.36) in the form (coefficient of — x2 here chosen to be ~.36)) 0 where + x2Q(x)y = 0. Instead of dealing with Eq (8.38) can be obtained.36). (8. i. Eq. % = {.55) The large x asymptotic expansion can also be obtained from the differential equation.e. S2.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.e. we obtain equations from which the coefficients So. x°. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on.37). • • • can be determined. ^ + 2 . i. 5„=4 Sl=o.. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. + — + ^ + • • • . cf. (8. S\.144 CHAPTER 8. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . We write Eq.. Eq. previously we had ^.60) into (8.^ + 1 ) . how series of the type (8.
» 1.z) is a confluent hypergeometric T. function. Magnus and F.b.62) 4 \ 2 XiFi 2'2 . The derivation of the asymptotic series from integral representations (cf.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. (8. T h e expansions (8. In the following we are mostly concerned with asymptotic series in some parameter.91. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. 2 X 2Wi.38) in a variable for the solutions of approximated differential equations.38). therefore. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before).+ X M W = Q. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ .19). / i N . (8. do not enter here into their derivation and simply quote the result: W i t h \x\ . i.37) with (8. are asymptotic series in a variable x. and it is clear t h a t higher order terms follow accordingly. We. p.e.35) and (8. 4~l~2 ' i i _i. n 22e"T JziL (8. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  .61) (the dominant terms of (8.8. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. . Nonetheless we shall need asymptotic expansions like (8. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains.37).63) **W.36) have correspondingly e±x / 2 ) . etc. Oberhettinger [181].
E. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . . 8.0. N. whereas e~x '4 increases exponentially.+ — + ••• of a function f(x) for a given range of a r g x .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. T.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. TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.67) Cf.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially. 4 (8. then the series is said to be an asymptotic expansion of f(x) in t h a t range if for a fixed value of n lim \x\—>oo ft xn\f(x)Sn{x)\=0 3. 349.4x2 (8. Whittaker and G. Thus there is no contradiction^ between (a) and (b). 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.146 CHAPTER 8. top of p.1 a2 an a0 + — + ^ + . Watson [283].
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).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. lim xneW x—>oo = 0 (8. (8.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . in particular with regard to possible exponentially small contributions to the expansion for which e.68) is (effectively) simply a statement of the observations made at the beginning.69) and the consequent nonunique definition of the expansion.g.e. (8. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.68) The definition(8.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis".73) Vn + q/#> + e2T/i2) + • • • . 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. we set E^En * > ^ n = En°1+eEnV = + e2EnV + . A critical discussion of the definition can be found in the book by Dingle [70]. Chapter I. .8.Sn{x))  = oo n—>oo (8. i. (8. which is assumed to be small. The definition is due to Poincare (1866). first published in 1902. possesses a whole chapter on asymptotic expansions. In fact. 8.
Ho^+V^n = EW^+E^n. . (8. i.75) HoW + vW = E^+E^+E^. TimeIndependent Perturbation Theory We insert these expansions into Eq. (8. J f°° dxrn E(^ 0) .4 0 ) M = / ^C(4 1} . ( ^ .w » . 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.e.i4 0) )V4 1} = {EM .As a consequence the state vl/) is a vector in this space. (8.148 CHAPTER 8. we write in the position space representation ^ ^ E ^  (878) A contribution to ipn can be combined with the unperturbed part of the wave function \P.78) into Eq.77) The states \tpn) of the unperturbed problem span the Hilbert space"K. (8. i.74) (8. ^ ) = 0.n. The second equation can be rewritten as {Ho .72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n. We insert the ansatz (8.70) and equate on both sides the coefficients of the same powers of e. With the help of the orthonormality condition (8.V)i.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .76) The first equation is that of the unperturbed problem. so that we choose / • dxrpM^O. (8. (879) We multiply this equation from the left by ip^ and integrate over x.e. We obtain then the following set of equations: Hrj.n = 40)V>n.e. °° i J or 0 = EWJdx{rnVil>n).
We therefore consider Eq. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates. S ' ti^+Q^)• We observe that Eq.m ^ n and integrate: oo /•oo / oo dxrmY. i.e.e. (il)i. (8. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n. We multiply the equation by ipm.Viprt e (8. (8.V)^ Setting + 42tyn. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. with En . we return to Eq.i ^ 0. (8.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).79).5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. i.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. multplying Eq. (8.83) makes sense only if Ef] ^ EnV for all i + n.Vil>n) (880) In order to obtain ipn . ^ n ) + 0(e 2 ).82) *> = ^ + E .8. the coefficients a\ .4 0 ) M 2 ) = ( 4 1 } . (883) Thus the procedure forbids the equality of any E\ '.^(EW ~ V ) ^ + E& / J—oo dx^miln . (8.76) which we rewrite as (H0 .
Vlpn)(lpn.^w .V'>pTl (E^Eny V'fc (8. (8. j. ipn ) = 0 i.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.81) we obtain (2) _ _ {lt>m.^j)(V .V^j)(^j.Vlpn] M _ P (0K 2 ^r .4 0) ^+ e 2 E (lpk. Inserting the expression (8. FVn) + e 2 ^ ( V > n .79) •/—oo oo / Using (8.^ ) + $& w £ (^fc.£i°») 2 Hence +^£o (£«? .V4>n){lpn.2 (V> m .86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq. ^ n ) j&iift + ^n .^(8.Vlpn)E.V^n) .(1) (i&0) .86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .85) together with (8. ~Ej E E. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.VV .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .88) (Q) Inserting Eq.87) (i>k.B i 0 ) ) ( £ f .89) • • • where we guessed the term of 0(e 3 ). (8.Vlpn) E^ . n) £f) (8. (8.e.150 CHAPTER 8. V ^ i ) X (Q) (Vi. En0) (8.
. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i.Vlpn) = y^Ci(lPk. (8.n(x) = J2crtn+ii (8.7 is based on this procedure. i. An ugly feature of the expansion (8. One would prefer to lump it together with the other contribution. 8. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations.e.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w.lpn+i) = ^ Cj5k. .89). We can therefore rewrite tpn ': The perturbation theory described in Sec.g.8. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. We then obtain (lpk.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.90) is the term with only one summation J2 m the contribution of 0(e 2 ).n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).91) The coefficients Cj follow from the recurrence relations. The functions ipn(x) are eigenfunctions of some secondorder differential equation. Merzbacher[194]).90) can be found in just about any book on quantum mechanics.
. (8..e. m = I. (8. are allowed to be equal since otherwise the expansion becomes undefined.94b) with (ipn.96) are effectively the known equations of ipnitpm and so yield no new information. (H0 + eV)^n and (HQ + eV)Vm = Emym.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. so that En = 4 ° ) + eE™ + e2EP + ••• . TimeIndependent Perturbation Theory 8.97b) with ipn y£ ipm. .4 0 ) ) ^ } = (Em] .Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . I — 1 . Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8.96) (Ho . (8.97b) are . Em = 4 ° ) +eE$ + e2E$ + . .93a) (8. is not uncommon. For example the magnetic states enumerated by the magnetic quantum number m. Equations (8. + c22^m). i. . (8.93b) The lowest order eigenfunctions belonging to En.We now insert Eqs.94a) into Eq.93a). Equations (8. . (8.93b). however. . (8.ipm) = 0 = (^m. i. (8. —I. (8. i.e. that of double degeneracy En = Em .4 0) )</4 1} = (En1] ~ V)(cu^n (Ho .ipn).95a) (8.97a) and (8.70). all n.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.152 CHAPTER 8.e.97a) (8. (8. degeneracy. Em can now be linear combinations of ipni'ipm. in a central force problem all have the same energy unless the system is placed in a magnetic field. In order to deal with the problem of degeneracy we consider the simplest case. Equality of eigenvalues.V)(c2l^n + ci 2 ^ TO ).94a).
99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) . ^ ) = ( V .ci 2 (V„. Then (^./>£>) = C2l[E£> .98a) Setting I = n.EJ®(lPn. H0lP$) = C22[E$ ~ (Vm.m).H0il>W) = cn[EW .H0xl>$) . and ( ^ m . But (8.6 Degenerate Perturbation Theory 153 inhomogeneous equations. (^m.(^04°))41)) = EM^CuTpn (ipl.ipm ) the equations (iPn.(lPn. V^n)\ .J#»a&> = 0j since ii4i = En (degeneracy).y>m.C 21 (^ m . c2iV^n + c22V7pm).(Vn. V^n). Vlf>m)] .C22(Vn. Wm.£ < « = ag)^) .8.Vil>m)] V^m). Vl/. (8. H0lpV) .H0^) . Similarly Wn. t f ) = a k 1 } 4 0 ) " Eg>aU = 0. . We now multiply these equations from the left by ip* and integrate. cn(ipm.^ ) ( ^ m . ^ ) ) = CU[EW .cnVipn and tyi.V^n).^ ( ^ .m. ^ o ^ ) ) = 0. Equations (8. (^n.ipn n) = 0 = (ipm.98b) + Culpm) + cuV^m). £ aW^°Vi) . (8. we obtain with (ipn.99) and ( ^ ^ O ^ ) " ^ ^ . Vi/>n)] .
101) This secular equation determines the first order correction En • The equations (8. (V>V</>V>) = const. Then / > i>" 2— = V ip Hence 0 = V ' V . (8. 9} ip. At x = ±00 the constant is zero.E™ (Vn.Vlpn) CHAPTER 8. i. if)' (/>' — = —. the secular determinant.ci2.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate.^Vm) 0. i.Similarly Eqs. d>" E=^. In this particular case therefore nondegenerate perturbation theory can be used (cf. Example 8. If one wants to perform a perturbation calculation in such a case.6. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . i> 4> Hence there is no degeneracy. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8.154 and V (^m.e. It follows that ip and <j> are linearly dependent.E)tp = 0.e.3).1. A onedimensional system. \nip = ln(j> + c o n s t . the degenerate perturbation theory must be used.^ ' V = — (•>!>'<t><t>'i>). However.5).C22. .1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. as we saw. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. It is then possible that a special perturbation. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. 4. dx i.e.100a) determine the coefficients Cn. like that providing the Stark effect (proportional to rcosO). splits the problem into two independent onedimensional systems.Vrl>n) . Example 11. TimeIndependent Perturbation Theory (l/>m.. 11. . in particular in parabolic coordinates (see Sec. (8.100b) determine C2i. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. ip = c<p.
of course. J. The subdivisions (8. Muller [73].104) "This method was developed in R. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. Dingle and H.103) such that if E is written correspondingly E = E0 + e. the method has to be applied in each domain separately. a potential. Applications of the method are. See also R. Muller [210].* In various versions this method is used throughout this book.7 DingleMuller Perturbation Method 155 8. Eq. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.e. . J. and the solutions then have to be matched in their regions of overlap.104) are chosen such that the equation Dcf) + (E0 . and such that one can even obtain a recurrence relation for the coefficients of the expansion. in particular in applications to periodic potentials.8. anharmonic oscillator potentials and screened Coulomb potentials. (8. We assume that V can be expressed as the sum of two terms. with some restrictions on the function V. W.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. also possible in areas not of immediate relevance here. It is immaterial here whether D does or does not contain a first order derivative — in fact. H. T For instance spheroidal functions can be studied with this method. B. Muller [216]. cf. (8. W.t Since in general a solution is valid only in a limited region of the variable. Thus the method is a systematic method of matched asymptotic expansions.V0)<t> = (v.102) where D is a second order differential operator and E a parameter. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. V = V0 + v.V)(p = 0. Dingle [72] and H. We consider an equation of the form Dcj) + (E . the considerations given below would be valid even if D contained only the derivative of the first order. i. (8.105) (8.106) (8.102) can be written D<f> + (EQ .V0)<f> = 0 (8. J.e)<t>.103). B. (8. W.
Em+s)(/)m+s. m + s)0m+s. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.108) with coefficients (m.+s ' (8. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .m + s)" in order to relate them to "steps" from m to m + s.e ) ^ = 2 ( m . s an integer: (ve)(pm = 'Y^{fn.V0)4>m = 0. and so v (8.e\ < \V0 .111) s then (see below) 0(1 = E (m. m + s) which are determined by the recursion formulae.m + s)4>m+s s (8. We write the coefficients "(m.Vo)(j)m+s = (Em .109) D(j)m+s + (Em .107) represents a first approximation to the solution (f> of Eq.) This result is obtained as follows.V0)<f>^ . (8. 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. v = {D+EmVo)<t>m. since D<f>m + (Em .110) If we write the next contribution <f>W to <p(°> in the form m+st (8. Then if ]e < \E\ and \v .113) . Considering (frm+s we have. the function 4>(0) = <t>m (8. (8. where m is an additional integral or nearintegral parameter (depending on boundary conditions).VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .102) with E ~ Em.E0\ (the latter in a restricted domain). the equation Dcf>m+S + (Em+S .m + .V0)<f>m+a = 0.156 CHAPTER 8.
117) „ . We observe that the coeffcients of expansions (8. .113) must vanish.n\ ^119) + .102) to order (1) of the perturbation (v — e). £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.119) follow a definite pattern and can therefore be constructed in a systematic way. (8.m) 1^1^ IW .114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.m + r)(rn + r. .8.TP \(T? . V ^ V ^ {rn.118).115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.m) + > ^ .Em+S) = 0 (second approximation). (8. Eq.. (8.m + s)(m + s. the coefficient of the term (m. Thus (m.m) — 0 (first approximation).m)4>m in (8. (8.^ (Em . (8.110) and are therefore related to the inverse of a .119) is effectively the secular equation). .116) m ^ v (m.e.m) — —.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .T? S The first expansion determines the solution 4> (apart from an overall normalization constant). and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. /01_.m + s)(m + s. (8. .m + s)(m + s. . We then find altogether ^  l^m  Em+S) + y^y^ SJ.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.m + s)(m + s. (8.m) E J^ ( m ~ Em+S) ' • fa.„N This procedure can be repeated indefinitely.Q (m.7 DingleMiiller Perturbation Method and so (m.
Other applications can be found in the literature. . 2 .1)( 2 + 64J3] — .e. i. ^The potential A r 2 / ( 1 + gr2). .[q(llq2 + 1) + 2(33g 2 . 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. and is typical of a large number of cases.2). as already mentioned after Eq. for the investigation of the behaviour of the late terms of the perturbation series. § H. S. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. for instance.9) + 4096J4] + 0 ( l / g 3 ) . has been treated by R. Miiller [211]. TimeIndependent Perturbation Theory Green's function. . does not occur in any of the later contributions.[ 4 ( 8 5 g 4 + 2q2 . 1 . in which q = in + 3.120) n4 .I2q + 64) + 256Z 3 (41 g . This means the solution is not normalized. Kaushal [146]. The eigenvalue expansion obtained is an asymptotic expansion. .* Example 8. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r .38). The following Example 8.423) + Z(2720 9 3 .6q + 1)1 + 24(5? .158 CHAPTER 8. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. this is — so to speak — the price paid for obtaining a clear systematics which is essential. (f>m in the above. for instance. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. W. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. J.e. as will be seen in later chapters. n = 0 . since it will be used extensively in later chapters. as will be seen later.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. (8. (8.2: Eigenenergies of the Gauss Potential^ Use the above perturbation theory for the calculation of the eigenvalues E = fc2 of the radial Schrodinger equation dr2 + k2 1(1 + 1) V(r) ip(r) = 0 with Gauss potential V(r) = g2e~a r for large values of g2. This can in fact be done and will be dealt with later.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . The result is the following expansion. An explicit application of the method (not including matching) seems suitable at this point. i. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory.
2 '(l + l) . . We now insert this equation into Eq. multiply the equation by onehalf and set h = —a/g. (8. Sees. .123) The first approximation (note the last expression as convenient notation) ^ . Setting q = in + 3 implies k2 + g2 ga(2l + q). <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.1).b. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. 1 1(1 + 1) 1 .2) where 1/ .a + l)if>(a + 1) + (a.123) the contribution fl<°> Ah + .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 .2a2 A. . Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . (8. 1 . 2 .l)ip(a . Therefore in the general case we may set k2 + g2 = ga(2l + q) .b.• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a.6 and 16. (8.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq.8.121). \2 . „2 J„2JL. (fc2 + <?2) 4ga and b = I \—. a)ip(a) + (a.0(z) = zl+1ez2^fa. b. i/> with Dq = d2 . We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf.V ! 4 I z S *! V2 ipq(z). 11. (.z2 is a normalizable function if a = —n for n = 0 . .. The solution i. 2 The solution XQ(S) = $ ( a . . 3\ 2 V ' 2. a .122) where A is of order 1/g. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2._ „!2 ^ ( " « 2 r r 2 > . . The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a. S) is a confluent hypergeometric function.
fa. Now we make the following important observation that Dqip(a + j) = jip{a + j). when j = 0. a ] i + 0(/i3) [a.120).a]i = . „ . (a. o + r) + S m _ i ( a . fa.a + 2]i P .3). a + r — l)(a + r — l . a]i + h fa. In fact. As a check on the usefulness of the perturbation method employed here. a + r) with the boundary conditions Sn(a. a) = b .2.l . The coefficients S m ( a . (8.a]1=0. of course. This equation shows that a term fiip(a + j) on the right hand side of Eq.a — 211 .^ 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. TimeIndependent Perturbation Theory (g . a + r) satisfy the following recurrence relation: Smi(a. a) = 1.S m (a. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps".2a = I + q. a + r + l)(a + r + l . 0 = 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.Z.1) = a . the latter determining to that order the eigenvalue.b = (g + 3) .a]2+L ^ J1[a + 2 . where [a.e.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. a + j)i/>(a + m). i. a + 1) = a = In general CHAPTER 8. (a.a—lli. a + r) = 0 for r > m. . a + i) = 0 for i ^ 0. a + i)(a + r. a ] x + L _ 2 J 1 [ « . a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a .120). a ] i + L ' ^ J 1 [ a + l . If we now evaluate the various coefficients in the last expansion we obtain the result (8. Hence the next order contribution to tp^ becomes oo i+2 r [a a . a + r) + S m _ i ( a . _ J [a .i ^(i)=£V+i i=0 J2 j = (i+2).160 where (a. So (a.j^0 ' + jU+1 J ^(a + j) with h[a. a . ( 5 m ( a . The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T (i+2) [a. a + 111 r [a. . one could now use the RayleighSchrodinger method and rederive the result (8.a + j}i+1^(a + j).
9. 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. the generalization to many particles.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. Regrettably this has to be left out here in view of lack of space. although not exactly straightforward. 161 . except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. and this means electrodynamics. as we shall see in Chapter 27.e. In continuation of our earlier search for an approach to quantum mechanics as close as possible to classical mechanics by looking for the generalizability of classical mechanics — as already attempted in Chapter 2 — it is reasonable to consider also classical systems with a wavelike nature. proceeds along similar lines. light. and we shall then see that these lead to analogous results as in Chapter 2. We therefore begin with the appropriate classical considerations with a view to generalization.Chapter 9 The Density Matrix and Polarization Phenomena 9. i.
When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.i). (9.3) (9. In the latter case one has the possibility of {A0R x A 0 /) • k > 0. (9.1) Fig. (9. 9.1 The vector potential ellipse. It follows therefore that A(r. This is the equation of an ellipse as indicated in Fig.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0.2) (9.6) .e.4) (9. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0. A 0 = A0R + iA0I. t) = A 0J R cos(cji) + A 0 / sin(u.t) = 3ftA. 9. In general the constant Ao is complex.162 CHAPTER 9.1. The Density Matrix and Polarization Phenomena potential A(r. t) given by A(r. i.
A2.A(°)). From the vectors A(°). i. (9. AW = £ fc=l. however. i.2 *i*4 0) (98) with coefficients F^. See for instance A.e.g. Here we are interested in observable properties of light waves — like.A« we can construct only the following scalar products with this invariance: (A(°\A(°)). (AW. the vector potential A.k (910) We give the initial wave the label "0" and subject this to an experiment. which leads to an outgoing wave which we give the label "1". with k .A 2 .A i = 0 = k . 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. (AW. the intensity of a wave. Rae [234].2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.e. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. represented by a filter "F". (9. (9. We set therefore A 0 = Ai + A 2 . the connection must be linear. as e. the direction of polarization. a calcite crystal or a film of nitrocellulose.AW). . Observable quantities can only be those which are invariant under translations and rotations.A«). Then their connection is given by Eq. (9. i.7) We consider now measurements with polarized light.e. (A(°). for instance.t in which k and ui remain unchanged. the fields E and B are observables.y)) components A i . The intensity 7 of a light wave with amplitude coefficients A\' . determined with the help of a polarizer. ie. I=<l2x2>:=E40)*l*40)i. In classical electrodynamics.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. not.2) the vector potential A is completely determined by its two (say (x.9.11) 'This means we consider the polarization.A2 is the sum of the moduli.8). p. we recall. 17.
(9. Then according to Eq. if (F) is an observable quantity.e.A( 1 )) = ^ ^ F i. (A(°). But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. pki = AkA*.164 CHAPTER 9.13) the quantity F must be an hermitian matrix and thus representative of an observable. i. We write the observable quantity as (F):=Y. the connection between the initial polarization and the final polarization.15) in order to rewrite Eq. i.k i (9.e. (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.A*FikAk.13): j2^FikAk = Y^AiF*kAi = Y. so that (F) can be looked at as the expectation value of an observable F. (F)=Tr(Fp). (9. (9. To this end we observe that as a consequence of Eq.k l f c 4 ° \ (9. i.12) i.k (9.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. it must be real and hence (F) = <F>*.17) . Thus (F) = J2FikPki = ^(Fp)a.13). We use definition (9. (9.e. (9.A*kF£A* = E and hence F = Fl A p i tkAk.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). Thus.16) i.
e. b = 0 and o = 0. 165 (9. <»^> (9. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. i.( J o ) . . The density matrix for the pure state (9. i. is given by a single polarized wave.d "»=(o ! ) " In the case of an admixture with equal portions of 50%.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). i.* For a lightwave travelling in the direction of z.22) For a — 1. like that of a bulb.19).18) The vector A describes a particular ray of light. The socalled pure case. is in general unpolarized. that of a pure state. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. A2 = b as </*) .9.21b) This state is still a pure state. 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. In the present case of a polarized beam of light we have two possible. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .(AAt) = ( £ £ ) . b = 1 we obtain the expressions (9.21a) follows from Eq. This admixture is described by introducing matrices p. (9. (9. But light. For a 45° polarized wave we have 1 .e. "incoherently. we have (9 19) ' In this case there is no preferential direction.15) with A\ = a.e.
We set in analogy to Eq. (9. Equation (9.™ V 0 n kmAm / ^^jPim^mk' . in which the intensity I = Tr(p) is not conserved (e.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. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.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. (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".8) A(1)=E^40)> k R = R(<*). Pi35° = \ \ 2 ? )• (9. as a consequence of absorption).166 and CHAPTER 9. (9.23) 1 . l l\ " i °\ ) ' ^925^ 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. 1 = .Z^ 3. Then according to Eq. b = —=.g.15) Pik — A i A k . (9.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M .
the dependence on a is contained in F' = R^(a)FR{a).31).pW).4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.e.32) (9. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. the observable F remaining unchanged.4 The Liouville Equation 167 pW = RpMRl (9. on the other hand.e.q. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)). The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.34) .30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. and p(0' remains unchanged. (9. i. here this contains the dependence on a. In the case of Eq. in the state functional) in analogy to the time dependence of W(p. in other words without creation or annihilation of systems.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. (9. the observable is transformed. i. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .e. that the initial intensity Jo is equal to the final state intensity I\.33) i.t) in Chapter 2 in what we described as the Schrodinger picture there. The corresponding condition here is that no absorption of the wave takes place.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.31) (9. (9. (9. 9. In the case of Eq. (9.28) (9.9.29) where (i.e. (9.30) Ti{FRp{®I$) = Tr(F.
M . Comparison of these equations with Eqs.2) has to be taken into account.36) (9. (9. i.21) and (2.39) If the dependence on a is not contained in p^1' but in F' (cf. = Tr(R^Rp^). j£ = i[H. CHAPTER 9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.e. the matrix M is antiHermitian: Mf = .37) (9. (9.p(°\ao)]da. (2. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9.e.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.F}. (9.B}:=i[A.35) Thus the matrix R has to be unitary. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. (9. Eq. We set therefore M^:=iH. (9.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.40) can be considered as quasiequations of motion.39) and (9. (9. (9.168 i.46) suggests to define Poisson brackets of matrices by the correspondence {A. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.40) Equations (9. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. In view of Eq.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.B]. .p].
the Schrodinger. 10. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor.Chapter 10 Quantum Theory: The General Formalism 10. (10. and consider timedependent perturbation theory.2 States and Observables Every state of a physical system is represented by a certain ketvector \u) which is an element of the corresponding Hilbert space "K.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Heisenberg and interaction pictures. 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). To insure that the 169 .1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. Thus in particular we establish the uncertainty relation for observables in general.
B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. 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.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space. (10. \pi. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. we assume this now. Here. q).Pi are called fundamental observables. we must have (u\u) = 1.B] = ih. E.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. which clearly does not hold in the case of operators.2. 10. such as for instance of the angular momentum operators Li.Qj] = 0.1 U n c e r t a i n t y relation for observables A. Such a correspondence with classical mechanics is not always possible. [qi. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. classically qiPi = PiQi.* a conspicuous exception being for instance a spin system. Naturally one can also write (10. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator.g. however.4) *Even in those cases where this is possible the operator correspondence may not be unique. Pi • Pi.Pj] = ihSlj. where i = 1. The phase space coordinates qi.Pj] = 0.2) (F(A)) = {u\u) <"lf. . y.2.170 CHAPTER 10.y.q).3 represent the three Cartesian coordinate directions of x. and assume that A = A(p. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. (10.
(10.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.9a).(A)2)1'2 (1 °= a) ((A2 + (A)2 . (A{A))(B{B)){B{B))(A(A)) (10.12) where \u) 6 "K is a ketvector representing the state of the system.{A)2)1'2 {{A2) ..5)) (for A\u). (10. using Eq.11) (A2) = (A2) .7b). AB defined by the mean square deviation AC=((C2)(C)2)1/2.13) AB = AB = (B2)1/2. (10. 111 (10. (10.(A) = 0.2 States and Observables then their uncertainties AA.6) B:=B(B).B\u) G IK as vectors) we obtain (for an application see Example 10.(A)2 = (AA)2.12) the Schwarz inequality (cf. (10. Thus AA = AA = ( i 2 ) V 2 .2A(A)}  {A)2)1'2 (10.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). (10. Then [A.8) and. Eqs. (10.B.. (4. we have A2 = A2 . Applying to the expression (10.2A(A) + (A)2. C = A.9) ((A2) + (A)2 . (3.7a) (B) = 0. Ai = = = ((A2) .10.. so that (A) = (A) . B}= ih.10) .. = \\A\u)\\2\\B\u)\\2 (10.B] = ABBA > ^h.2(A)2 . Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).7b) = = [A. Since A = A— (A).0)1/2 = AA.
(10. R= (AB + BA).18) >R2 + I2.15) as (u\AB\u) = R + iI.17) as had to be shown. AAAB>I=h.3. From this we obtain (uiJBu) 2 = i22 + / 2 . for a complex constant c. (10.16) I = h. and with Eq. 4. A\u)=cB\u). Zi + BA)\ + ^ih.e.13) (AA)2{AB)2 i. (AB + BA) This means (AB + BA) is real.14) Hence (u\AB\u) = /^(AB where.4). (10. (10. The equaltosign applies in the case when (a) it applies in the Schwarz inequality.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*.172 CHAPTER 10. . since A and B are hermitian.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). Zi (10. (10. and hence for condition (b) 0 = = tSee Sec. The condition (a) is satisfied when. (10. We can therefore rewrite Eq. and (b) when (AB + BA) = 0. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. Zi Zi (10. (10. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih.
1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. i. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. i.20) or. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. ALX 10. (ALZ)2(ALX)2 > \h2 = (Ly)2. Separating the product into symmetric and antisymmetric parts. = 0. in view of Eq. = 5R2 + G 2 > 9 2 = hi2(Lk)2. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.) 2 > = x. Example 10. ALy = (u\L±\u) ^ 0.e. 9 = h{u\eiikLk\u). q) = ih.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier.Lj] = (LiLj + LjLi) + ih£ijkLk. For states \u). the following uncertainty h2{Lk)2. (Ly)=0. h2(Lz)2. {u\LiLj\u) = R + iQ. . (ALy)2{ALz)2 > ~h2(Lx)2.3 OneDimensional Systems 173 Thus c = — c*. (Lx±iLy) ^ 0.e. i.13) to operators Li — Li. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. (10. Ly are not "sharp".e. 9ftc = 0.z. Thus in this case Lx.10. such that Lz is "sharp". For these we have in the onedimensional case the relation [p.i relation holds: (ALi) 2 (AL. The observables like angular momentum are functions of p and q. m\lm)). (Lz)2^0. (10. i. implying {Lx)=0.e.7a). (Ly)2=0.y. What does the relation imply for states u) = \lm). for which one component of L is "sharp"? Solution: Applying Eq. + LjLi\u). (A{A))\u)=i(Zc)(B(B))\u).
p] = ih we obtain therefore [q. Eq. Similarly we obtain \p. Q2] = [p.i h ^ ^ . Using [q.e. q]q + # > v] = ~2ihq and \p. Then (see below) we can derive for A the following commutator equations [q. We obtain these equations as follows. A — A(p. i.qn] = ih^(qn).A(q.p) can be expanded as a power series in q and p.21b) (10.23) = ihg(q2) (10.A(q. i.C\. and (c) the spectrum is not degenerate. (10. This is why the operator p requires a representation in the space of eigenvectors of q.p)]=ih^^and \p.21a) and (10. they do not have a common system of eigenvectors.p] = ^ihp ~ Proceeding in this way we obtain [q.B]C + B[A.e. as remarked earlier. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).p}p + p[q.p2] = [q.174 CHAPTER 10. . (10.21a) Equations (10. (3.21b) now follow with the assumption that A(q. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable.34d)) [A.pn]=ih^(pn). Next we have a closer look at the operator q.p)] = .22) d ih—(p2). We have (cf.BC] = [A.q). We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). (b) its spectrum is necessarily continuous.
(10. we obtain [q. 10.21a) A by U. 10. (10.27) (10. i.25) It is shown in Fig.e. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.e. U] = ih^= aU. an operator.1 how the delta function arises in the limit K —> oo. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10.f^ ^ J—oo dke ikx —ikx' (10.1 The delta function for K — oo. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. Replacing in Eq. . For the cases (b) and (c) we consider the operator U(a — pipa/h (10.10.3 OneDimensional Systems 6K(x) 175 Fig.26) it Here p is an observable.24) lim SK(x — x') — S(x — x'). = eipa/h = Uia)'1 = U{a). > We begin with (a). U is unitary.28) i. and a a cnumber.
e. CHAPTER 10. q any other observable F(p. which means that the spectrum of this operator is a continuum.e. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). q) can be constructed representing a dynamical variable. cf. Eq.00).29) This means: U\x) is eigenvector of q with eigenvalue x + a. ^Observe that {iji\A{p.00). or qU(a)\x') = (x' + a)U(a)\x'). Evidently every eigenvalue has only one eigenvector. 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 (—00.3. 10. b a cnumber).30) (10.1 T h e translation operator U(a) We saw above. and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p. (10. (10.176 i.29). the spectrum is not degenerate. (10. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. Thus all these values belong to the spectrum of the operator q.q)U^U\ilj). we can see that also p possesses only a continuous spectrum (from —00 to 00).q)\i>) = (i>\U^UA(p. WU = 1. i. Since this holds for any arbitrary value of a in (—00. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).31) . or qU\x) = U{q + a)\x) = (x + a)U\x). that with U(a) = e~ipa/h = U.
5(x> .x") U{x'\p\x" % pz. U(a)\x') = c\x' + a).x') = = i. (10.e) ~ 5{x' .34) {x"\U\a)U{a)\x') c*(a. \x') = U(x')\0).e.x") .36) i. it follows that 5{x" .35) (10.x" .31) and (10.e.38) h 6(x' . (10.32) we see that \x' + a)ocU(a)\x'). i.37) With the help of the latter expression we can calculate. since q\x') — x'\x'). offdiagonal elelemts of the {x}representation of the momentum operator.3 OneDimensional Systems 177 Moreover.32) Since U is unitary. (10. (x'\p\x").e) (x \p\x ) = .x')5(x" = (x" + a\c*c\x' + a) x').e.x")c(a.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' ./ (10. The operator is therefore called translation operator. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' . (10. i.x" . we have q\x' + a) = (x' + a)\x' + a). (10. (10. the operator U(a) acts to shift the value x' by the amount a.x')\ = l.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). Comparing Eqs. (10.x" .e.a).39) . In the expression of Eq. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). i. for instance. We choose the phase such that U(a)\0) = \a).e. \c{a.10.33) (10.
it follows that 5'(x) is an odd function of x. to) with respect to t we obtain ihjtU(t. a conservative system. In this case the classical Hamilton function does not depend explicitly on t. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.4. i.178 CHAPTER 10. The operator 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. i. H\uE(to))=E\uE(to)).L fdpeipx. (10. (10.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.f dpipeipx. .43) (10.45) (10.° ° d(p) (ip) e"« = S'(x). (io. Differentiating U(t. Quantum Theory: The General Formalism We mention in passing that.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. 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)).x") = (x'\p\x")*.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .to) exists which is such that^ m)) = u(t. (5.e.44) (10.e.27) and (5. .4i) We assume first of all. 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).to)m0)). so that U{t.tQ) = eiH(tt<Mh. This is in conformity with our earlier postulate in Sec.35).t0).to) is called the time development operator. 5.40) 10. 6'{x) = !. See Eqs.t0) = HU(t.e. since 5(x) = .
44)) we have U(t.z / dt'H(t')U(t'.to) (differentiating we obtain immediately Eq. Ufa. t0) = H(t)U{t.t0)]1 Let us set for small values of e: U(t + e. = U(t0. (10.41) we obtain ^(t 2 )> = = J7(t2. The operator H is d efined by this relation.t0)U(t0.t).47) e.to) = 1.4 Equations of Motion 179 We now demonstrate that Eq. (10.t)U(t. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. (10.t)l]U(t. (10. But obviously (cf. Evidently U(t.10.to). i.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. (10. Then with U(t + e.45) has very general validity.48) with U(to.t) = 1.48)).t) = l. (10. (10.48): [/(Mo) = 1 . From Eq.t) = l^eH{t).e.t0) or ih—U{t.tiM*i. Hence U(t.t0)\^(t0)). t0) (10.t) and therefore [U(t.t 0 )^(*0)> U(t2.49) .*l)^(*l.to) can be expressed as an integral as the solution of Eq. t0) = E/(t 2 . Eq.t0) [U(t + e.*l)^(*l)> = I7(*2.
t)\t/>(t)) =U~ and since H is hermitian. (io.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). Quantum Theory: The General Formalism We obtain the differential equation for the development of states of the system in the course of time by differentiation of Eq. (10. (5.e. the predictions of different descriptions are identical. The operator U(t. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.180 CHAPTER 10. it follows that U(t + dt. Another such description is the Heisenberg picture. Comparison with the equation postulated earlier. (10. Since the only measurable quantities are moduli of scalar products.(£ + dt)) = U(t + dt.53) . 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.41).51) Since </. and these remain unchanged under unitary transformations. requires the identification H(t) = H= Hamilton operator.t0)\ii>s{to)) (10. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. the instant at which a measurement is performed on the system. Now let \tp(t)) be the state of the sytem at time t.30).5o) ihjtm)) = Hm)). Then \Mt)) = u(t. the Schrodinger equation ju(t. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.t) = l % ^Hdt) \il>(t)) Hdt (10. i.t0) m0)). Eq.e. >(*)> = i. The eigenvectors of these observables are also constant in time. 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.
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. for instance in cases of nonequilibrium or if part of a system disappears through absorption. As(q. Thus. . (10. {AH.55) and using Eq.p. It does not always have to be the case that As does not contain an explicit time dependence. in order to obtain the Heisenberg picture (description).e. as a rule.e. (10.H]U = = = U]ASHU AHHH — U^HASU .With explicit time dependence.48). we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.p) in the Schrodinger picture dAs/dt = 0. (10.4 Equations of Motion and \^H) = U\t.57) and rf[As.t0)\ips(t)) = \ips(to)) = constant in time. In the Heisenberg picture it is HH = U]HU. 181 (10.HH].(tf HU)(rfASU) (10.e. i. we have to take into account the contribution dAs/dt. i. Differentiating Eq.H]U + iW]^U.56) U^[As. subject to a continuous change.HH]+ihU^U. (10.55) Thus AH is explicitly timedependent. However the associated observables are timedependent. but here.58) HHAH (U*ASU)(U^HU) [AH. we consider only cases without explicit time dependence of As. (10. if ^ = 0. For As(q.HH]. even if As is not timedependent. a scalar product) remains unchanged: AH(t) = U\t. so that a matrix element (i.i).54) Observables transform correspondingly with a similarity transformation. t0). t0)AsU(t. Here H is the Hamilton operator in the Schrodinger picture.10.59) This equation is called Heisenberg equation of motion. = so that ihjtAH or ihjtAH = = [AH.
An observable CH is called a conserved quantity. we used in essence unitary transformations of matrices. provided the Poisson brackets in the latter are replaced by commutators. For an extensive discussion see B. Finally we add a comment concerning conserved quantities. In our treatment of representations. . A.21b). In the treatment of the pictures we used unitary transformations of operators (and vectors). since in the transition from one to the other the commutator remains unchanged. 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. (10. in the case of the fundamental dynamical variables qi.182 CHAPTER 10.21a) and (10. In particular. In many cases the Schrodinger picture is more amenable for explicit calculations. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. Example 10. We end with a word of caution. (10.Pi): and where the expressions on the right follow from Eqs. 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. This holds also in the case of the Schrodinger picture.e.61) Thus conserved quantities commute with the Hamilton operator.HH}=0. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. We also observe that the momentum pi is conserved when \pi.H] = 0. i. Orfanopoulos [224]. Solution: We have for an observable A (A) = (i>H\AH\TpH). if ffCH = 0.13) to derive the uncertainty relation for energy and time.Pi we have in the Heisenberg picture (replacing in the above AH by qi. We see that formally one obtains Hamilton's equations of classical mechanics. [CH.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10.
the "centre of mass" of the statistical distribution of A.H]) = 0.)\2 ( = \{i. dt in. 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. in a different form: The eigenvalues of an observable are called "good" if [A.67) 1 d(A)/dt is oo and AE = 0.64) = 0. since \IPH) is timeindependent. then if a measurement is made at time t'. (A). at ^±AH>h.Such a characteristic time interval can be defined for any dynamical variable. H\tl>) = {H(H))\f). AW = {A(A»V>.HH]) Hence we obtain by ih(A). We construct the following vectors using Eq. If r is the smallest such characteristic interval of time. a position or momentum observation is independent of t.e.68) To put it The eigenvalues a of A (i.62) The Hamiltonian H does not depend on time. (10. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. if we set AH = AE we obtain TAAE (10. > (10. Let \ip) = \"4>H) represent the state of the system.10. this is practically the same as at time t for times t with \t — t'\ < T. 2 d{A)/dt " and.e.13) and (10. we can replace ([AH.63) so that with Eqs.e.7a).66) This means. (10. (10.\HAM\2) \(nA. is displaced by A A in the interval A T = TA.4 Equations of Motion Then. and TA O C (10. (10.e. of A\<t>) = a<£)) are then called "good quantum numbers". i. H] = 0.62) and dAs/dt > \WAH\i.H}\f)\2.p) we then have 0=^M) = U[A. (10. This means that for a physical variable A(q. 183 dV ' \ dt >"•"»» + ( £ ) • (10. i. (10.2 > h.65) and rA AA d{A)/dt' . . i.
/oo (10. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. But the number of radioactive (thus excited) nuclei decreases exponentially.)exp(iEt/h)).72) = V'EoW 7/2h + 0i(EE )/h  oo . The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. (10. This means that in the case of decaying particles.71) From Eq.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . (10. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. the wave function ipEo(x.(x)fEo(E')e«EE'Wh 2irh f dE'^E.70) by From Eqs.Eo(x)eiEot/he'^2h9(t).69) and (10.H ) ( £ . First we obtain from Eq.E1) 2irhfEo{E)4>E(x).e.t) — ipE(x.184 CHAPTER 10.tyEt/hdt = = = / J—oo dt J dEJrl>E. (10. (10. for instance. have around EQ the form ipEofct) = i.69): oo poo p / oo *PEo(x. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. states with sharp energy (AE = 0)..69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. the particle density diminishes in the course of time (t > 0). as is observed. the wave function ^ ( x . On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime". i. i ) = J dE'^El^)eiEltlhfEo{E'). (x)/ E o (E')5(E .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. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime. Quantum Theory: The General Formalism 10. Recall for instance radioactive decay.70) we obtain: oo / oo TPEO (x.70) we can determine the function fEo{E) integration with respect to t. This means.
10. to) be an approximate but unitary solution of this equation. We obtained the result (10. i.11). W)) = U(t.e. every unitary transformation of states and operators defines a possible "picture"). They are called "resonance states". (10.e. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state.10. that** EQ.to) = HU(t.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). Considered as a function of E. (10. i.75) (10. Let U^°\t. and is called "interaction picture" or "Dime picture". where U is the solution of Eq. . of course.e. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. Eqs. ^(*o))> then \xp(t)) for t > t0 follows from Eq. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. (14. This description contains effectively parts of the other two pictures and is particularly useful when Eq. £/ = £/W(Mo)^'(Mo). i.45).e. We saw previously that if the state of a system is known at time to.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). i.105) and (20. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. The state with lifetime r — h/'y is not a discrete state. 7 > 0.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use.74) Thus the solution of this equation is the main problem. (10. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. (C/(0) unitary).g.41).t0). (10.e.76) **For specific applications see e.toM{to)).45) can be solved only approximately (in actual fact. ihU(t. (10. 185 i.
= HU<®U'. Now let H^(t). dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). [/' has only a weak ^dependence (i.e. with Eq.h^U> = at ( io. i. £/(°)£/(°)t = 1. (10.79) ^'~°i.e.e. In addition we have (10. with Hamiltonian H = H^ + H'.78) (10.82) (10. i. H(°Ht) = ih^uW.186 Then CHAPTER 10. Quantum Theory: The General Formalism ih^(uWu') i.e.e. i.t0) 0 = l.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.e. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0.81) But [/(°) is unitary.83) ih^U^ ishermitian. . varies only slowly with t).e. (10. (10.
We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.78) at (10 85) (10.84) U^ where H' is also hermitian. lfe(*)}.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.e.87) = We define correspondingly as interaction picture state iM*)> = t/ (0).85) = where we used Eq.89) where ^s(i)) and As are state and observable in the Schrodinger picture.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^.90) = U^H'\^s(t)).U^HU^U' H'fi'. (10. We also set H^U^H'UW. (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).92) . (10. U^H'U^U' (10. (10. We multiply this relation from the left by and from the right by U^ and obtain umHU(o) = c/wt (^WtjW) + umH>Tj(°) i n u ^ ^ + U^H'U^.85) multiplied by U') on the right of Eq. U<®\Mt)) = \Mt)). (10. (10. and obtain from Eq.78).10. i. (10. H = H{®{t) + H'. (10. (10.80). (10.86) at = U^HU^U'.
(10. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). near zero and \tpi(t)) is almost constant in time. we have i.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. dAs/dt=0) . (10.60a) or (10.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .90) and replacing the left hand side of this equation by the right hand side of Eq.89)^0. the vector varies with time (different from the Heisenberg picture). ^ (0) l^(*)>. a perturbation so that H ~ H(°)). A^t) = U^AsUi0). comments after Eq. In order to obtain the equation of motion of an interaction picture observable Ai.89). Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. Since H'j is assumed to be small (i. i.e. i. (10. i.60b) with iff0) instead of H. we differentiate with respect to time the operator defined by Eq.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). H'j = UWH'UW. (10.e.e. ot i.e. cf.94) Then we have (normally with dAs/dt dt (io=8o) — 0. (10. (with ihjtAI(t) = [AI.188 CHAPTER 10.94) _Hf)Al 1 + lhdAl+AlHf).55)) at + imm<Ms_u(o) at dt _umH(.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.e. so that with Eq. (10.H\0)}. (10.e. (10.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation.
7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. ^ \<Pn){Vn\ = 1n (10. i.102) We assume that before the perturbation H' is switched on at time to.e. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em.93).94) Hi = uW(t.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.e.97) The equation to solve is Eq.i0) = e<Ho(tto)/ft. We obtain the operator U^ from Eq.7 TimeDependent Perturbation Theory 189 10.99) (1098) Instead of the original Schrodinger equation we now solve Eq.100) (10. ih±\Mt))=Hi{t)\Mt)). H' = H'(t). (10. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10.10. H0\ipn) = En\(fn).t0). (10. (10. {ipn\<Pm) = Snm. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')). (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. (10. We have the Hamiltonian H = H0 + H'. where according to (10.98). i.80) as E/(°)(t.t0)H'uW(t.87) and (10. We assume again that the spectrum and the eigenvectors of Ho are known. Since . i.e. (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 .103) The actual state \ipg) of the system is solution of ih\^s) = H\^s).107) Jto With this we obtain from Eq. (10. n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon of the perturbation H'. (10.102) and truncate the series after the first perturbation contribution.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. (10. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^).108) .190 CHAPTER 10. Quantum Theory: The General Formalism HQ is independent of time. (10.106) We insert this expression into Eq.^ V ™ ) . Hence we set ^> = £°«(*)l4 0 ) >n. (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„. i. We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the timedependent perturbation depend on time.104 ) Here we insert Eq. (10.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).e.103) for m — n and use Eq. the time dependence of \ips } can be separated as for stationary states. we can write l 4 0 ) ) m = elHot/hWm) =e .
Em)t' (<Pn\H'(t')\pm) = Wnm(t). where the momenta of the aparticles form a continuum. or if by emission of a photon with continuously variable momentum an atom passes from one state into another. A further example is provided by adecay.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^.109) 10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. We assume that the perturbation is switched on at some time to.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En .En (10.Em)t} \{tfn\H'\ipv h 2h(En . (10.110) With this we obtain from Eq.111) Different from Eq. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions. if in the scattering of a particle off some target its momentum p changes to p'. but that otherwise the perturbation is timeindependent. » — < (10.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . Thus we set H'(t) = H'0(t). (10. This is the case.109) the transition probability Wmn(t) = jp Jo . t o = 0. < KO?t .10. (10.112) for a^0. for instance. (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum.
Otherwise.— = o (En 2h En 2h5{En .io.2 The delta function for £ — oo. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .Em)t} r i 7 TTT. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. so t h a t Wnm would be zero.E7m)t} = (10.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. t—>oo Hence also (shifting t to the other side) sin2{^(ffn. 10.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. we would have En — Em ^ 0. if the energies belonged to the discrete spectrum.Em).113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . > 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.192 CHAPTER 10. 10. and no other state would be nearby. .2. though sufficiently small. Fig. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. (10. Hence lim St(a) = 5(a).
(10.3: Application of test functions to formula (10.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation .113). sin 2 (tx) ax i—( * According to F. so that initial and final states belong to the continuum.113) Use the method of test functions to verify in the sense of distribution theory the result (10. We evaluate the integral 0(0) = 1.117) in many applications. Example 10.e. If we are concerned with pure scattering. Pauli [227]. Let p(En)dEn be the number of states in the interval dEn. 266.117) provides the outgoing particle current (see Example 10. then Eq. (10. _ 2irh AE > > Se.8 Transitions into the Continuum 193 as equal. for instance from a potential.r Z x 1 ^ ) = 4>{Q). Then.W . pp.118) where 8e is the separation of states around En which in the case of the continuum goes to zero.117) were derived by Pauli in 1928. (10. F Jo . 142. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.10. 75. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .113) this is* hm . if p(En) varies only weakly with En. f See E.116) The transition rate V is the transition probability per unit time. Fermi [92]. (10.4).115) and (10. In the case of the functional (10.* In view of the usefulness of Eq. p. i. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). Fermi gave this formula the name "golden rule" . in the article of W. . Schwabl [246]. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10./ t>oc TZ J_00 dx .
. . * ^ = . 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). p. L \3k2dkdUn It follows that T.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . = 0 .? 2TT / L \3k2dkdUn 1 /V(x). p = 1 we obtain I = ./ cfc a n p 2 +fc 2 . . and hence . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . § We have r Jo \p\x dx sin ax sin bx  V. . .4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. The ingoing particle flux is therefore v / L 3 . . Jin § m = See I.<p). . x . E x a m p l e 10. 1 . Ryzhik. . ..947. with n .( ^ V V • < ih 2 _ i k n . Quantum Theory: The General Formalism with the help of Tables of Integrals. S. formula 3. 491. Gradshteyn and I. dEk = h2kdk//j. 2 T / ^ M.tan p 2 + a 2 . Then ^ . M..( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). iknx . Then p(Ek)dEk Since Ek — h2k2/2[i.. 2 . 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. with v the velocity of the particles. ^o n .194 CHAPTER 10.k.e " 1 * 1 = 2 lim — .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . = ( — ) dk= 2ir (—  2irJ k2dkdfl.„I. K = k 0 . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . Hence the result is 0(0) as expected. V +(ab)2 a _x 2p6 7 m —^ '.a 2 ' i that for a = b = t . E = h2k2/2fi. . 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. and is obtained as ih. 77^ + ..
10.121) . by which we mean without leaving the Schrodinger picture. specifically in the case of the Coulomb potential.69). we present another derivation using the usual method of timedependent perturbation theory. (10. 148.122b) S e e also L.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).x)^(E'.120) In the case of Eq.10. (21. pp.122a) (10.119) only the time dependence of HQ can be separated from the state \ip) in the form of the exponential factor of a stationary state. For a possible application of this result.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx. 193. We expand the wave function V( x . t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.t).t)^°\E^t)dE (10. .t) = ^On(t)^(x. Schiff [243]. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known. (10. (10.9 General TimeDependent Perturbation The differential cross section is defined as da •• Theory 195 outgoing particle current x r2dfl ingoing particle current da / /i V (hko)/(lJL3) J dxV(x)eil where for purely elastic scattering k = ko.^°>(E.t)+ / n a{E. see the discussion after Eq. Thus we set ^(x.^ We start from the Hamiltonian H = HQ + H'. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>)./continuum = <San(£)</4°)(x.x) = S(EE'). (10. 1 rfx^°)*(x)^)(x) = 8mn.
Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.125) where an(t) — dan(t)/dt. The second term on the left of this equation and the first term on the right cancel out.t)dx.E').. . Next we use the orthogonality relations .122b). since f rJ>(°>(E.x.iJ5 »*/Vn(x).127) we obtain from Eq. i.196 CHAPTER 10. f 6kn in case of discrete tp's.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. V>(x.x). (10.t) = eiEtlhy(E. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).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.123) We now insert these expressions into Eq.x.120).„.122a) and (10.127) i n c a g e Qf c o n t i n u o u g £. (10. dnptMrnW = { ^ _ En) (10. t) = San(t)<pn(x)eiE^h. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.129) . (10. (10.„„. (10.x..t)rJ><®(E'. We multiply this equation from the left by </?£(x) and integrate over all space.i) = e. .e. (10. into dip (10. = e^EE'^h5(E Thus with Eq.. Observe that these conditions do not contradict Eqs.. (10.E') = 5{E ..128) . / *. .. ^0)(E.126) + H')<pn^)eiE^h}. .
132) These equations can be integrated successively. Thus first we have ak0) = const. i. (10. and we set ak°}=5km or ak0) = 8(EkEm). i. to perturbation theory in the lowest order.e. (10.130) H'kn by \H'kn and at the end we allow A to tend to 1. In order to solve Eq. We restrict ourselves here to the two lowest order equations. (10. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10.130) we develop a perturbation theory. since it is equivalent to the latter.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. (10. SH'kn(t)e^~E^ha^(t). Thus A serves as a perturbation parameter which permits us to equate coefficients of the same power of A on both sides of an equation.130) along with the parameter A in front of H'kn. let's say am / 0. (10. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.e. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.10.134) . determine the state of the system at time t — 0.133) The numbers ak = const. 0. they are fixed by the initial conditions. SH'kn(ty^E^l^\t) (10.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . In the first place we replace in Eq.131) and insert this into Eq. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . (10. Comparing coefficients of the same powers of A on both sides.130) This result should be compared with the Schrodinger equation (10.120). We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.
an expression of the following type AS?) = " 2 ^ 2 / e j ( k .135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum.137) . so that ihakl\t = oo) = 0.132).135) and (10. The amplitude / B o r n is a Fourier transform in spatial coordinates. (10.Emy in agreement with Eq.111). i.e. ) ' X V(x')dx'.e. i. ipm) into the state k. Integrating the second of the equations (10.198 CHAPTER 10. In this case we obtain from Eq. (10.ml2sin2{^frn (En . What is the physical significance of Eq. Equation (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. (10. H'^y^"^^/hd£. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9. (10.k .135)? The quantity ak\t) is the amplitude which determines the probability that as a result of the perturbation ^H'km the system makes a transition from the initial state m (i.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. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). (10.136) we see that both "amplitudes" are Fourier transforms and look similar.136) Comparing Eqs.e. (10. ip) in the context of timeindependent perturbation theory. ifm{kl) = f J —oo (10.135) We put the additional constant of integration equal to zero.
Ti. The Hamilton operator is then given by w= jt + ji_^M. meaning the inclusion of the creation and annihilation of field quanta). r=  r .m.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e.i be respectively the momentum. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory.y. Let r : = (x.Chapter 11 The Coulomb Interaction 11. 11. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics. 199 (1L2) . the position coordinate and the mass of particle i with i — 1. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1.1 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. Z<i = Z) y(r) = _ M ^ . such as the electromagnetic field (the quantization in these contexts is often called "second quantization". ( 1L1 ) We let pi.2 Separation of Variables. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.2. 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.
3) and centre of mass momentum P and the relative velocity p/mo by setting r» . so that as expected we obtain the expression for p in Eq.200 CHAPTER 11. y. (11. r = rir2 mi + m 2 (11. p are such that as operators in the position space representation or {R.2/ mi+m2 The mass mo is usually described as the reduced mass. where m 0 = mim 2 .5) The momenta P . mi + m 2 iTi'2 r2 = R ^ —r. Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' .4). (11.3) for the individual position coordinates we obtain ri=R+ •^ —r. r} representation they have the differential operator forms Setting R = (X.4) P = p i + p 2 . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. Y. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . Solving Eqs. .U7&. 2Pip2 m\m2 pip2 mi + m 2 . (11._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. mi + m 2 va\ (11. so that as expected according to Eq. Z) and r = (x. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. /'Pi p = m0 P2 A v. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. P = pi + p 2 . (H4) \mi m.
(11.Pj).Pj} = ih6ij.e. (11.2 It follows that p2 2 P2 n2 n2 n2 (H.+ ^(r) 2(mi + m 2 ) 2m 0 (H.8) and so W =777 v + 7 T . Angular so that Momentum 201 ^ ^ . Hence Hamilton's equations apply and we obtain. Pj over to operators ?i. pj.pj.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. (i.pj.z).Ri. 2(mi + m<i) 2mo 2m\ 2m. Ri.12) (11.2 Separation of Variables. r) is a canonical transformation.= ^. P on the one hand and those in r. for which we postulate the commutator relations* [ri. (1114) m0 We observe that the equations of motion in R. p = W. with M := mi + vn^R = ^ P.r2) to (R. (11.+ ^2.y.11. Pj (which in the following we write again ri. (11. we mean n = rx s x.+ P. . 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).13) * and = M ' 1V1 P = 0. r = —.j = x.pj]=iMij. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.10) we set H = Hcm + H. (11.il) ""Op' i. p on the other are completely separated.Ri. The canonical quantization must always be performed in Cartesian coordinates.16) Although we write ri.10) One can convince oneself that the transformation from (ri. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii.15) In accordance with Eq. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. etc. [Ri.
r).17) (11. i. V(r) = V(r).21b) + E.23) y — r sin 9 sin tp.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R.22) If.21a) L*+v{r)] ip(r) = Ei/j(r) (11. it is reasonable to go to spherical coordinates r.20) r2MAR_ $(R) = ER*(R). = rcos9.Q 2M' + V(r).e.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . ip: x z = r sin 9 cos ip. (11. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11.202 CHAPTER 11. (11. as in the case of the Coulomb potential. 0 < 9 < ir. r ) = J E t o t a i*(R. Thus the twoparticle problem is practically reduced to an effective oneparticle problem. The Coulomb Interaction nc H = 2m.r) = *(R)V(r). 9. (11.24) . 2mo (11. —n < tp < ir. (11. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r). In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . 0 < r < oo. where Hcm$(R) = and Hip(r) = with ^total — ER (11.
(11. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. with tijk — + 1 . if i.2.)In Example 11.26) However. is the operator of the relative angular momentum.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.29) .3. (11. (11.k a clockwise permutation of 1. (11. = (p2 + ^ V (r + 0).31) .1 it is shown that pr is hermitian provided lim n/>(r) = 0.33) i. 2 1 9 / 1 9 .28) .12 — I2.32) (11.j.j. . We have I2 = (r x p) • (r x p).5r = . j .i ' . k an anticlockwise permutation of 1.2. (11. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r._0di})\ x We now demonstrate that 1. where (r x p)j = ^eijkrjpk.11.3. otherwise. = 0.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. i. i—>0 (1L25) (11.e.2. = —1. k = 1.pr]=ih.3. n ( .2 Separation of Variables. i( i. _ h2 \ .27) We can now write (see verification below) p V = h2Ar^ where 2. that the above expression follows from l:=rxp=ift(rxV). (l + .
ih6jkrjpk (11. * ' S Since r =xz we have (cf.39) + ih).37) .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.38) + y* + z .j.ih5jk)pk (11.35) j.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.ihr • p r2p2(rp)2 < + ih(rp). The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.(r • p ) 2 + 3ih(r • p) .41) o = r pr. Eq. (11.40) (11.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p . (11.k Using Eq.k ~ rjpkrkPj) (11.ihr • p ) . This means that l2 = so that 9 9 1 (11. (11.43) r2(p2p2). (11.44) .ih) — rpr(rpr With Eq.ih)pr and x 2 2 9 9 2 2 Q O (11.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. (11. r— = r • — or or (11.27) we obtain r{prr)pr = r(rpr .42) (11. (11.
11.26). prip) = = I <j>{r.46) (11. <p) d °° .e lim rip(r) = 0.2TT]. 8. ip 6 X>Pr C 7i.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. <p)r2dfls = . 6. <p)* . or J .<P) = Etl>(r. /'(r.6.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3). reR3. S o l u t i o n : Set d£2s = sin. H 1 2mo pi + K) +V(r).2 Separation of Variables.28) and (11.6d0dip.[ i J With partial integration with respect to r this becomes (<l>.ip) = (<£.e. <p))r2drdns % J r or r<t>(r. 6. 9 € [0.<p G [0.45) oe Thus in Eq.47) Example 11.1 Separation of variables We deduce from the expressions of H and l 2 that [H.— (riP(r.<p)drdna " r r=0 J \ll§r~(r4'(r. r := r : lim rip{r) = o i ..e. (11. <p)*Pri>(r. 9. i.[ 4>(r. (pr4>. h f l d —(r<l>*(r. Pri/Oi V0. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.d_ r dr the operator pr is hermitian.7r].Pril>) J / r= dQs rcj>(r.2.30) I2 is to be identified with the square of the angular momentum operator. (11.<p))drdns.e.<p))ril>(r.d. pr = .! /').Q ' 2mor2 with the condition (11.l2} = 0.11. (11.e. ¥ p)r 2 drdn s = (p r 0.<p) (11. Then {<j>.<p)*?(ri.48) . 0.(r.6l. <p)*rip(r} 0.e'iP^} . Angular Momentum 205 Comparison of Eqs. r—>0 il>(r..e. i.
zpx . (11.lz]=ihlx. 11. These operators here play a role analogous to that of the quasiparticle operators . 2 . i. However. [lz. (11. 3 .l2} and correspondingly [lx. xpz] = y\pz.l2] = 0.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. Example 11. z]px + py[z. (11.51) (11. i.q2 = V.206 CHAPTER 11. The Coulomb Interaction i.53) Either of these operators is the hermitian conjugate of the other. y. so that [lz. we have [h. [ly.xpz] = [ypz.pz]x = ih(xpy . g i = x.2: Verify that [h.lz (or l . 2 2 2 = 0. ?3 = z).ypx) = ihlz. (11.e. [lyA = 0. z we have altogether [lx.e. Since 1 = r x p .ly]=ihlz. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. We therefore first search for a complete system of eigenfunctions of l 2 .50) Thus the components of 1 do not commute pairwise.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. /z> *xJ = l <i\}ylx + tx'j/Jj [h.lx] = ihly. ly] = [yPz ~ zpy.52) This means that l . H and l 2 possess a common system of basis eigenvectors. zpx] + [zpy. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0.49) By cyclic permutation of x.lx or l .qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .e. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. for any two of the components there is no common complete system of eigenfunctions. they are incompatible variables. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.
e. (11.11.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq.49) we obtain the following relations: [iz.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. (11. (11. (11.l)hl\l). i + ] = [l2. [ig.61) But lo is by definition the largest eigenvalue of lz.Z_] = [ l 2 .Y\l) = (lQr)h(lY\l). l\l) is eigenvector of lz with eigenvalue (IQ — l)h.h~lz(1156) (11.59) = 9) h(l0 r)\lr). With the help of Eq.lh)\l) = {l0 . (11.+ ll.i+] = m+. . (11.3 Representation of Rotation Group 207 A. (n.52): 0 = [ l 2 . (11. y . and with Eq. (11.i] = m.r) := (Z_)rZ) so that lz\l .51). Hence we must have Z+I0=0.).r) (11 = {l02)h{l)2\l) (11. (11. (11.57) i.54) l 2 = \{l+l~ + 11+) + ll = l+l. Therefore (lo + l)h cannot be an eigenvalue of lz.54): lzl\l) = {llz . We write or define: / . (11.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i).54) On the other hand we obtain from Eqs. Similarly we have lg{l)2\l) and more generally lz(l.58) {l0 : max. [i+M = 2%iz. assuming a finite dimensional representation space. A^ in the case of the linear harmonic oscillator.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. In the present context the operators are called shift operators for reasons which will be seen below. since lz is hermitian.60) Similarly for l+ with the help of Eq. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.
l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 . (cf. (11.\lr) terminates at some number (let us say) r = n (n a positive integer).n) (1 = 60) {l)n+1\l) = 0.Z_] = 0. (11. \l). (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.h)\ln) {(lQnf(lQ. i.l2\l) = l0(l0 + l)h2l\l).e. l\l . (11.55)) we have 12Z_Z> = l.\ll).67) i (11 60) = But according to Eqs.e.63) + l ) ^ 2 .From Eq..60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.68) .66) It then follows from the second of relations (11.n)}h2\ln). (11..65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .e.y\i) l)h2\lr). Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_. IQ(IQ 1 0 (11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.208 Thus for l 2 we obtain: V\l) ' = 11.60) and (11.64) i.56) CHAPTER 11.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11. [12.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). i. (11. The Coulomb Interaction (l+l..+ 11+) + l2z\l) = i+i.n). (11. Eq. i.54) (n = 57) (11.e.. (11.
Messiah [195].e..m). We write these* for integral values of IQ: \l. In the following we consider primarily the case of integral values of IQ. R.l_. As in the case of the harmonic oscillator we introduce normalized basis vectors. (IQ — l)h.11.l.. *D.m) with — lo<m<lo2 (11.m) = = mh. y (lo .m) = = = = y/lo(lo + 1) — m(m +l)\l. The dimension of the representation space is therefore n + 1 = 21Q + 1.3 Representation of Rotation Group so that with Eq. The operators l+..g. Satchler [38]. —loh.m)...m .  U o .m) l\l.63) and (11.70) they satisfy the eigenvalue equations (with m = l0Jo l.71) can be established in analogy to the method used in the case of the harmonic oscillator.m / + l)h . The phases are chosen such that§ l+\l..m) — mh\l.l)h \/(lo + m + l)(l0 . (11..l0): l2Z. 17..70) According to Eq. The normalization factor in Eq.. \l — n) into one other. The nonvanishing matrix elements of 1 = (l+.. . where lz\l. lQ = .m\lz\l.69) Prom this it follows that IQ must be either half integral or integral.n) 2 .m) = mh\l. § See e.m)\l.lz transform the vectors \l).65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. lz) are then: (l.. M. (11. p. 24.m) = IQ(IO + l)h2\l. (11. II.m±l\l±\l. we skip this here.m).1). (11.l)h. According to Eqs. (11.m y/lo(lo + 1) — (m — l)m\l.n){l0 .. i. \l — n) are eigenvectors of lz with eigenvalues loh.1). m + l)h. Vol.m) (l.{lo .. lo = l0n. Brink and G. A.m + 1)(Z0 + m)\l. lz\l.2.n .... y/(l0±m + l)(l0^m)h.Z 0 ).m) : Mo).n)} = (l0 . We therefore write the 2/o + 1 orthonormal vectors: \l. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.. The basis vectors \l). p.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo .
as this is sometimes described. l_\l. 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. and m = —IQ.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.l0)=0. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). (11. CHAPTER 11.yp.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. (11. ip.. (11.210 Then.76) d2 (11.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. ip) = e(9)$((p).2. —lo + 1 ..With the ansatz Y(9. We now perform the analogous procedure for the angular momentum or.1. for the simple "rotor".<p). $ oc eim*. with i2Y£{e. where IQ = 0. . /o. . . as above.. mhY£(9.e.<p) = i0{io + lzY£(0.73) 1 d d sm0 06 \Sm9dd) + (11. 9.75) the variables contained in the expression (11. d2 sin2 9 dp2 _' (11.. In these coordinates we obtain: h l± so that h2 = xpv . 11. The Coulomb Interaction Z+Mo> = 0. i. We .77) $((p) = m2<b(y).74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9). ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . as remarked earlier.<p) = i)n2Y^{e^). .
t»'\Y7n((l'. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. m > 0. This is the case if one demands that l 2 and lz possess a common complete.n>). and that their phases are such that Yt0(0.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.(5cos 03cos0).76) and (11. (11. (11.ypx = tfrK(11.V>)Yir'(8. is given by (2Z + l ) ( Z .(p) are called spherical harmonics. (11. so that im(ip±2ir) The functions Ylm(8.* ? ^ . (11.i».<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. Uniqueness implies immediately that m has to be integral.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.80) y? Furthermore Y°l = V^fcosO). orthonormalized system of eigenfunctions.r o ) ! YT{e.82) .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. We mention in passing that in spherical coordinates d lz = xpy . / 3 / 5 J — casO.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.<p) Jo = 6mm'5w. (11.79) .0) is real and positive.^ = *(n . These functions are defined in such a way that they are normalized to unity on the unit sphere.11.\ = W .
85) (11.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 . oo (11.90) .e. (11.x2)y" .86) y = 0..d.p. (11. the designation s.2(m + l ) x i f + [1(1 + 1) .87) we obtain the equation (1 . respectively.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) . For the solution we use the ansatz of a power series. i=0 (re lowest power > 0).2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.2xP[ + 1(1 + l)Pi = 0.g. 22 (11.89) Inserting this expansion into Eq. we would now have to search for those values of I for which the solutions are square integrable. i.x2)Pf . It is instructive to pursue the appropriate arguments. i.e. is also in use. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . (11. Consider the equation for Pi(x).e.m(m + l ) ] i f = 0..h..1.88) we obtain for every value of x ^2 ai(K + 0( K + * . (1 .1(1 + l)]xK+i = 0. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. (11. in the replacement ip —> if + 2K. f. i.x2)P[' .m = 0.2. (11. of Ytm. Instead of the values I = 0. exp(imip).x2)y" ... Setting y = (lx2)m>2Pr(x) m . If we did not know yet that I = IQ = an integer > 0..212 CHAPTER 11.88) y = ^aiXi+K.l)xK+i~2 i .2xy' + 1(1 + \)y = 0. i.e.
94) ^For further details see H.91) from which we deduce that K = 0. However. Margenau and G. (11. Eq.46)) p.78) to be square integrable. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .E Xl(r) =0 (11. . M. Schmidt). We have therefore (multiplying Eq. l 2 and lz are the solutions of the Schrodinger equation of the form (11. when the series terminates after a finite number of terms.*" 11.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. . p. Eq. For \x\ < 1 we can obtain from this the coefficients of a convergent series.92) This is a recurrence relation.26)) yi(r = 0) = 0.11.96) + V(r) . (11. This is precisely the case. ai(re + !)« = 0. first line of Eq. (11. (11.94) from the left by r) h2 d2 h2 2 + H! + \)IL. For the solution in accordance with (11. (11. (11.5 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. We see that this happens precisely when I is a positive or negative integer or zero.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.Qr2 with (cf.95) Vl = rxiThe hermiticity condition for pr then requires that (cf.2 + 2mo dr 2mor V{T)E Viir) = 0 (11. but also those for a different series for a. l(l + l)h2 _2mo + 2m.1.93) where Xl(r) is the solution of the radial equation (cf. these are of little interest to us here. Murphy [190].1(1 + 1)].1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. 93. > 1. they have to be polynomials (recall the orthogonalization procedure of E.
This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. The Coulomb Interaction (11. Obviously the irregular solution has to be rejected. The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. (11. oo) of a onedimensional space. s = 1 + 1. The equation also has an irregular solution.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. i. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r.1. He + . Equating the coefficient of 1/r2 to zero. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. (11.1)+1(1+ 1) = 0.102) .100) with (11. We now consider the hydrogen atom with V(r) — —e2/r (i. nor the condition of normalizability for l ^ 0.e.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. and substitute this into Eq.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ).99) e ip(r) = Eip(r). for instance. In the first place we investigate the behaviour of yi in the neighbourhood of the origin. L i + + . since it satisfies neither the condition yi(0) = 0. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.e. an analogous consideration applies to hydrogenlike atoms like. Z\ = Z2 = 1 in Eq. and others.96).214 with CHAPTER 11. r m0 = meMp me + Mv (11. we obtain the socalled "indicial equation!'' s(s .1)). (11. which behaves like 1/r1 in approaching r = 0.
p. for E < 0 they > decrease exponentially. (11. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . (11.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.6.a * = 0. i. see e.1 Discrete Spectrum of the Coulomb Potential T h e eigenvalues Here we are interested in the solution which is regular at the origin (i. We can see from Eq.v) vi(x) = 0. behaves there like xl+1) extended to an exponentially decreasing branch at infinity.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. Oberhettinger [181].g. Inserting the ansatz into Eq.88. i.11. .x ) $ ' . T This equation is known as Whittaker's equation.+if>x/2 vi(x). hypergeometric (11.e.6 11. (11.e. (11. (11.105) we obtain the equation for V[(x).105) 11.108) This equation has the form of a confluent x$" + (6 .106) W i t h this ansatz we separate from the solution the behaviour around the origin as well as t h a t at infinity. (11. *The factor 47reo appears in SI units.105) t h a t for x —• oo there is a solution behaving like exp(—x/2).e.107) equation.103) °y moe2 ( = 0.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. It remains to determine the function vi(x) for a complete determination of the solution.104) W i t h these substitutions we can rewrite Eq. Magnus and F. We set therefore Vi X J. W. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. (11.
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.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function.(0_i)!(6 + n l)!n!' (11. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases.106).! ) ! = imlFI' c a l led inversion or reflection formula.e.z(zl)\ = z\.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). However. (z)!(z . meromorphic function. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z.1/2)!. called dupZication formula. .112) The gamma function is a nonvanishing analytic. ^ ( 2 ^ ) ! = 2 2z z!(z . (2) r ( z ) r ( l . (11. (5)r(n + l ) = n ! . particularly pure mathematicians. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. more precisely.216 CHAPTER 11. (11. 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. (11. it is convenient to use only the factorial notation. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . with many of these around in some calculation. i. (4) r ( i ) = V5F.J. The function T(z) is defined by the integral /co T(z) := / eHzldt. and hence the *Some writers. the factorial for an arbitrary argument. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . and for both cases.z) = ^f^.
e. 2 .101) to (11. in our case: a = 0. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).. .V))T(v . which are also described as its nodes. For every value En. .1) + n = n(n . One defines as the principal quantum number the number n. the orbital or azimuthal quantum number I can assume the values 0. for z. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression.I) r(vl p)sm{n(i/ I . i. (11.e.1) + n = n2. and so to the number of different values of m. n' = 0 . must be a polynomial.2. i.» • » • » ..11.. Eqs.v) T(l + lv) _ ~ K sin{7rQ/ .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. This is achieved.104)) 1 n =v = K. every n. . (11.e. 1 .114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. if the series terminates for certain values of u.113) This is a quantization condition.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. Thus the infinite series must break off somewhere.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ .1..110) as follows: T(l + 1 + p . = n = Z + l + n'. With (cf. d = 1. i. The degeneracy of the levels En is therefore of degree* n—1 1=0 n—1 V(2Z + l) = 2 V Z + n = 2^(n . 1=0 (11. which is thus seen to be a consequence of demanding the square integrability of the wave function. (11. The quantum number n' is called radial quantum number. .6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. .
'. 2 . from n. nd. / = 1. (b) The spin of the electron was not taken into account.p.218 CHAPTER 11. stand for I = where n denotes the principal quantum number and s.. 0.'..1 = 1 (called diffuse series. respectively. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.. . I = 0. from an initial state n > 2. I = 0 to n = 2.. from an initial state n > 2.l = 2 to n = 2. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.1 Hydrogen atom energy levels. 11. hence s). The following spectroscopic description of states is customary: nl —> ns.4. .1 = 0 (called principal series. (c) The nucleus was treated as pointlike and not as something with structure. np.1. .. One describes as Balmer series the final state n = 2.. ' O n e describes as Lyman series the final state with n = 1.§ One also writes n2l+1l.1 = 0 (called sharp series. 11..1 = 1 to n = 2.. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. . ne.d.\ I = 0 or 1. 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. and hence the fine structure of the energy levels is excluded.1. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. nf . and from n. hence d). hence p). .
414(3). .6. (11.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). Ryzhik [122].z) = (r + k)\ Lkr(z). Messiah [195]. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. p. • fc. formula 7.118a) where Lk. . W. Oberhettinger [181]. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr).p)\(2l + 1 + p)\ The spectrum (11. Appendix B. k + l. With the help of Eq. however. .6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. Unfortunately the definition mostly used in mathematical literature differs. S. see. 11. • ( . We distinguish between different definitions in use. the polynomial contained in y\ of Eq. (11.r f . Each can be recognized by reference to the generating function.116) Laguerre p o l y n o m i a l s : Various definitions in use! The polynomial obtained by terminating the hypergeometric series (11. We mention here three different definitions used in the literature.1. For clarity and later use we denote the function used there by its normal form L^. several different definitions are in use. l . 2 . I.= t . Gradshteyn and I. Unfortunately. p. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. in fact. Apart from factors.114b) yields the binding energy of the states. (u.115) we can reexpress this function as Lr{z)  r\k\ p=0 \* p\ (rp)!(fe+p)! (U 118b) ^This definition of *LJ? is used by A.2l { + 2.115) is effectively the associated Laguerre polynomial normally written Lk(z). . • (11. a + 2:*) = t v .110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.x). (11. M. Vol. so that one always has to check the definition. for instance. . 84. and I.11. 3 .115) ' =0 (ri . "1'f„+11" P] „. 844.117) The function *Lk(z) is a polynomial of degree r.2 (11.r = 0 .2. Magnus and F.
and Ll(z) = l. L1(z) = l . Sneddon [255]. 163. **I.1. L2(z) = l2z + z2. ^ F o r this definition and the following equations we refer to I.„M„ z"*L. Example 11. L\(z) = 2z. Vol.d z—^z + {k + l~z)—+r Lkr{z) = 0. There at some points n + 1 is misprinted n + 1. N. .„\*Kk.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.. Appendix B. N.** The associated Laguerre polynomial here.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. r r Z r=0 v tr k k (11. It is said that this is the definition preferred in applied mathematics.220 CHAPTER 11. pp.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.3): e2t/(lt) (1 . The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. 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. which we write ^Lk. Ll(z) = 33z + ±z2.164. Sneddon [255]. p.z .3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. is defined by^ dk (—'\\k(r^\2 (11.122) "See A.2. (11. I.. .119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . 162 .121) In Example 11. Messiah [195].
122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). (11. Then.1 . (11.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. N d rk 0. and obtain with this the normalization condition (11.z . (11. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11. and partial integration.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp..1 8 + 18z 3z2.124b) Comparing Eqs.124a) (i*) pi The orthonormality condition is (cf. '"''Comparison with Eq. Solution: We consider the following equation of the form of Eq.120). Example 11.124c) Example 11.119) derive the generating function of associated Laguerre polynomials (11. ^L2(z) = 2 .107) implies k = 11 + 1 and r = n + I.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq. (11.. i L1(z) = l . with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp. The limits will be determined later.11..4 + 2z..121). ^L\(z) = .120)) it)k& zt/{lt) ~*P p=k fc (11.8): (11. (11.119): ty"(t) + (a + 1 .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 . Jfpd>2s{p))eptdp. and *L\(z) = .118a) and (11.4z + . *L\(z) = . (11. .
with requirement of a finite solution. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a./ ( l — <?) —— = G(t).q) and P~ t ( . s bK> (lq)°+i (11. (3) around p = — 1 if a + b is a negative integer. We demand (to be considered later for integration contours) the following condition. which removes [..( a + l ) p s ( p ) + 6s(p) dp 0. (11. Then dp = dq/(lq)2.] from the above equation: C(p) := p(p + l)s(p)ept = 0. p and one has (as polynomial of degree 6. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp.li I u {p(p+l)s(p)} : p(p + l)s(p) dp a + 1 b p+ 1 p(p + 1) a + b+l p+ 1 b p Integrating the equation we obtain s(p) and hence y(t): a+b+l^b ln[p(p + l)s(p)] = In ip+ir+o+'p (p+l)a+b f (p+l)a+b The condition C(p) = 0 now implies for possible contours of integration in the plane of complex p: C(p) = p(p + l)s(p)e'pt („ _i_ na+b+i = ^ ' e~pt = 0. It follows that 1 d i I _i. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . Case (2) implies the exponentially increasing solution ~ e + t . j p = (P 0 "V?.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). Hence one defines now as associated Laguerre polynomial L£ and as the generating function g(p) of these (Cauchy's residue theorem requires the coefficient of 1/p): L b(t) •= — <£ 27T. the only possibility is the last.p ) « + i q c + i ( i . p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo.p ) o + i 7 t = 0 .a = 21 + 2. Hence the expression in square brackets vanishes.p + 1 = 1/(1 . since for n > b : (b — n)\ —• oo. agreeing with Eq. (2) from —1 to oo." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. case (3) similarly. and (4) around p = 0 if b > 0 and integral.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. this is the rejected irregular solution. Thus in the case of Eq.222 CHAPTER 11.. .) 2 J J p 6 + i ( l . 2TT_ Jq=0qb+l(lq)a+l 1 . t > 0. 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. (11.119) with a and b as there. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} .^ .
11.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.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. I. 11.111). so that I = (c + a)! 1 f .6 The Discrete Spectrum of the Coulomb Potential 223 We evaluate the integral on the right with the help of the integral representation of the gamma function.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.(x). f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as .125a) and the factor a~ 3 ' 2 with a = Bohr radius. Clearly only the second form is sensible. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = .6. Jo (11. is introduced for dimensional reasons. The wave function of the state (n.112). .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. where Fnl(x) = yi(x) (11. and 0 if cj=b. (11.?). (11.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. Eq. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! . m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.—i.
3fc + 2). 11. (11. (11. from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix ()r2dr 2 ^i + /„.131) .! ] + (21 + l ) 2 . thus on the average the electron is the farther away from the proton or nucleus the larger n is. a (r)is = ~a.1 + 1) (11.224 CHAPTER 11..3(2Z + 1) + 2} 2 {2(n + 0 .2Z .(l) n+l.7. The Coulomb Interaction One finds in particular (see also Example 11.v (6n2  6nk + A:2 + 6n .2.8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ . For the ground state (n = 1. + .128) These expressions permit us to obtain the mean or expectation values (1/r) and (r). This behaviour is indicated in Fig.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 . = 0) we have ) r /u =.6(n + Z)[(2Z + ! ) .(2Z + 1) + 1} = ^[Zn21(1 + 1)].n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .130) We see that the mean value of r is the larger the larger n is. and for the explicit derivation of the second case below Example 11. 2 (11.
In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton .(r) = x W 2 n + 1 =.j . This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. Problem 5. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 . or (r ) ~ n a 2 2 ^ [ 3 n 2 .2 1 A r = V (^2) .20. = l a nl n N'nll rFnl.6 The Discrete Spectrum of the Coulomb Potential r. 11. In the case in which I assumes its maximal value n — 1.132) (11.( n . Sneddon [255].l ) n ] =  [ 2 n 2 + n] an[ n + (11. v ^ V2n + 1 (11.J n 1s 2s 3s *r Fig. \f{r ) ~ n a.134) *These are taken from I.11.2 Hydrogen atom wave functions. / ! J — . we have and tp becomes particularly simple. with n +1 replaced by n and 21 + 1 by k.e. 173/174..133) for large r.2 I 3/2 Kl r 225 \2 1 R„. S.^ We obtain therefore for the quadratic deviation (r) . i. pp.
6.135) This expression becomes very small for large values of n.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. and separating this in cylindrical coordinates r. in which periods are obtained from the BohrSommerfeldWilson quantization condition. .3. we have pg = const. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. i. but rather like classically in a plane (i. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.4 the relativistic equation in cylindrical coordinates. p2 = p2 + — pg • eV.e. This means. In order to obtain an impression of relativistic corrections in a simple context. i = 0. However.226 CHAPTER 11. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory".— > where ampler is the length of the semimajor axis of the elliptic orbit. we consider in Example 11. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. pJ = —— and j> Pidqi = nth. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2).r. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. the states Ytl are predominantly concentrated in the xyplane). = ngh. \/v vT 7 v^ See also Example 14. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. for these values of I it is not distributed with uniform probability over the spherical surface. The Coulomb Interaction or TT = ^ = (11. Since 6 is a cyclic coordinate. (c) The nucleus was treated as pointlike and not as something with structure. Example 11. the electron remains practically localized within a spherical surface of radius (r). „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler .e. Thus. 6.
which are related to Cartesian coordinates in the following way. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'.(p . 11.V . Whenever there is degeneracy. 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. it is possible to separate the Schrodinger equation in other coordinates. r\ = r ( l — cos#) = r — z. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. p. 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..11.136a) (11. . Schiff [243]. where r = ^/x1 + y2 + z2. The deeper reason for this are symmetry properties of the problem.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». z=^rj). ip are the spherical coordinates.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. where the constants are identified by comparison.T). 2 2 Therefore the energy is no longer a function of n = nr + ng.6.9. y — 0: V ^ c o s tp. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.3. where ip = 0. as also indicated in Fig.* See L. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . y=y/%qsanp. 86. (11.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. <p = <p. 11. which correspond to a new choice of polar axis.136b) £ = r ( l + cos#) = r + z. ( v c + i ) .2. (p).
22) becomes — as we show in Example 11.3 Parabolas £ = const. for uniqueness of the wave function.136b) the Schrodinger equation of relative motion (11.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. there is really only one basic equation to solve here. with zaxis as axis of rotation.228 CHAPTER 11. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . = 0. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. 1 . 2 . We rewrite the first of these equations with the help of the relation M (w ^  m + e) . .137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. . and r\ = const. a separation constant. so t h a t the variables can be separated. . In terms of the coordinates (11. 11. (11. The Coulomb Interaction Fig.
.11.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). n 2 = n" .105)) Hence we write I = (m — l ) / 2 .6. E x a m p l e 11. 1 . . . We leave the calculation of the degree of degeneracy to Example 11. . . m0Z1Z2e2 ^ 2 = n (say). " .114b). i. (11. . 1 . ds2 = dx2 + dy2 + dz2. as well as the Laplacian A = V and the Schrodinger equation.± ( m + 1) = 0 . 2 . In the spherical polar case we had (cf. 2 . we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". Adding now / .125b) (dividing by p 1 / 2 ) £ oc e . T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . from Cartesian to parabolic coordinates.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric.106) and (11. .o. Eq.6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. .e. To obtain a closer analogy with the separation in the spherical polar case. p = /?£ and cr = /^r?. . (11. . ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. . T h e n by comparison with the spherical polar separation the answer is as in Eqs.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 .
e. we have to find the number of values of the quantum numbers n\. and so dz = (d£ — drj)/2. Oberhettinger [181]. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .g. are elements of length.l)a + (n .1 in this expansion is seen to be n. (11. + l 4 {^^)drl f i + v \ . for m / .* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n .g<p 9 \ . " .e. Here we have (see preceding text) 711+77. (n _ !) n = n2 §See e. 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. Similarly we proceed with dx and dy.230 CHAPTER 11. see Eq. \— < / — cos ipdr] — \/£r] sin ipdtp. .2)d/2.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. and attach a factor to to each term introduced in the wave function. N2 where g^dS. i. . The Coulomb Interaction Solution: We begin with z of Eq. i.114c).e. n 2 (1UJ)2 The coefficient of a . (11.. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. / — cos ipdl. Thus dz2 = . Solution: For the wave functions ip = £(£)J\f(r])&(ip). > .[df + dr]2 .136a).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 . Thus. In the case m ^ O w e have n\ + 77.2) and summing over m. i. as t h e two signs of t h e exponential indicate. We use the method of selector variables. 712 and m which give n. </. 0 there are two..2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . 2 + ^dV . . etc. {g^Y. Eq. Observing that the highest possible value of m is n — 1 (i. for instance. Beware! For m = 0 there is only one mstate. Magnus and F.2 . W.l ) ( n . = e±im^e("+^'2{par'2L^{p)L^ (a).1) ~ .2d? dr)}. n the double sum ]T]2 is TI — m.e. 2 2V? V^ One now performs the square of this and obtains similarly dy2.2 = n — 1. We then sum all possibilities and select the coefficient of ojn~1 in the result. when m = 0 = 77.2)d} = (n. all cross terms cancel in the sum of 7 the squares.E E n l „ . We consider first the case m = 0. 2 _ = (9id02 + (gvdvy + /„ .tx2 . z = (£ — rf)/2. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr.136a). Thus we want to obtain the coefficient of wn~1 in £2 . dx = d{ y £77 cos ip) = . last chapter. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). (11.
with the connection » (11.128). i. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).6). s 0/ . (11.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I..drid<fi oc (p + o)dpda. 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 8.11. = j (2ni+m+l). ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. Ikr+k. with Z\ —> Ze. Z2 —> e.r. However.•* n (l) = / Jo It follows that .™(p)] 2 By contour integration (see Examples 11. {p — a) — = dxe~*xkvLkn (z)] 2 .—r~n = 3(mn 2 ). where F is the electric field.128) if one makes there the substitutions n —> n\ (or 712) + m. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. with RayleighSchrodinger perturbation theory).{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—. i. 8. . In fact. we would have to use degenerate perturbation theory and obtain the same result. . eF{z). 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.6 The Discrete Spectrum of the Coulomb Potential Example 11. . these expressions agree with those of Eq.124c). . fc — m. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .127) and setting n = r + k.a). Thus e. we can relate the expressions Iq to those of Eq. .g. With the help of the wave functions we derived above.8 and 11. if we expressed the perturbation in spherical polar coordinates. and determine the average value of v for a given state. with z = r cos 9. Sec.1 (cf.) = ^eF(p .e. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. I. L ni\ ny.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. with Eq.e. (11.e. (11.
232 CHAPTER 11.x ( n + m)! ( n . tt„:_ (1 . (11. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .g )(l.124d) of the alternative integral representation obtained in Example 11.n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .p ) ( l .3 by one power of t.l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. The Coulomb Interaction E x a m p l e 11.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . c positive integers. _ „ .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.b.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.111) /*oo /*oo „ — sam+l / e~attm+1dt= (m + 1)! ds = am+2 I ' Jo For use in the following recall that Jo «m+2 {i + X)n = ~ ^2(ir . dt = ds/a.p ) ( l . the integral arising here becomes with Eq.. In Example 11.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. (11. s = at.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.a. E x a m p l e 11.1 ) ! (n .^~T^——^ ) = coefficient of (pn . the integral / is the coefficient of pnqn in the expression _ / .3).0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .128).( m + 2 ) is equal to the coefficient of (qn — qn_1) in (1 — pq)~(m+2\ The coefficient of qn in the expansion of (1 — pg)~ M is pn(n + fi — l)!/rt!(/i — 1)!.3: /"OO I := Jo et[L^{t)]2tm+1dt. and is a special case of Example 11. .1)! n\ This verifies the expression I™t in Example 11.pK — — —L .9 we consider integrals with an arbitrary power of t.p ) r + 2 [(lp)(lg)]"l+1(lpg). 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 .9 )« Solution: Using the generating function for both associated Laguerre polynomials.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n .9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. (n + r .
7£. Using the contour integral representation (11.w(r) = and / a ) = r\(P ~ r . we have 2r = 6 + c = even.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . ° ^ + a ( l .f ]. c + b — 2r = l (i.r ) ! a ' .124d) of an associated Laguerre polynomial.q)(l . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. i. i. — (cr)!r!(6r)! Kv.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.a f + c)\(vr)\(l + ayrP „ • ^0^r(cr)\r\(a •• Ha) a\ f dp ac(c + a)\ ^ =%*??* c!a!(l + a ) ^ c + l 9 ( p ) = g J c+*"«a> where with /3 = a + c + l + b (observe that for a = 0 a term with c + b — 2r = 0 remains) „_ = 2 ^ 2 ^ K «. c + b odd).p ) ] . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.(i) = i ! ( . so that = (q+fc±£) . c + 6 — 2r = 0 (i.e.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.a _ 1 = [1 + a ( l . we have to have y + (c + b — 2r) = 1. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l .™( r > (1 + a)P r £^0y^w Jc+6_2r = . (lp)(lq) + a(l q)(l .111) of the gamma or factorial function. c + b even) and/or j / = 0. (a) j / = 1.^ ( l . i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. .L.e.?(t)t a e.11.a t [Lg(i).e.l ) * x coefficient of a* in the expansion of 7 ( a ) .p) .e.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. .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 . where (a+**£)! c j n Jo = .
is the work of J. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. In view of the slow falloff of the potential at infinity. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). . t We have the Schrodinger equation of relative motion of particles with charges Z\e. ^A very readable source to consult.7. + (!±<)«. however parallel to the Coulomb case. b\ V cited in Example 11.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. . the standard principles of nonrelativistic scattering theory. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.234 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 . Taylor [267]. These results are seen to agree with the expressions 1^. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. 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. also with regard to earlier literature.)!' f^)(.1%^ 11.^)(. cannot really — and this means in a rigorous sense — be applied in this case.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. R. 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.
z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a.z).M) = l + . (11. (11. (11.140) The simple Coulomb potential permits particularly simple solutions.87 under Kummer functions. Magnus and F.140) and using (11.e.144) *(a. pp.7 The Continuous Spectrum of the Coulomb Potential Then Eq. Oberhettinger [181].138) becomes 235 ^A + k2 .z). i. (11. Our question is now: How does ipr behave for r — oo.z)). In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr).z) + W2(a. that we encountered previously.^ W ) = 0. According to Eqs.. for instance. i..109) we can therefore write the solution ipr = Aeikz$(i>y. z = rcos<9. (11. 86 .. A = const. b. .145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). . with the asymptotic expansions (11.141) Inserting this expression into Eq. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.11.e. (11. (11. v = ik(r .b.z) = Wi(a.146) See.. a(a +1) z2 ] (11.+ ^  + .b.108) and (11. where ^. for r — oo) • ipr = Aeikzf(r .143) and v = ikrj. „ az 1. . W.142) This equation is of the form of a hypergeometric equation. what is the > behaviour of 3>(a. (11.136b).b.
kz+l^ktrz)) C . A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.a) (z) na)"~ r(la) T(ba) n\ ' (11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n. J\") ci(kr~/\n2kr) (11.l. > A = Aeikz[W1(i1.ik(rz))} =^ + ^d.(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. (11.150) ^(rz)]1"^ 1 + 0 T(«7) i.151) with Eq.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.148) .z)T(l .153) The asymptotic solution represents t h e stationary scattering state (E = const. with 2.152) determines the quantity f(9) as /(*) = 7 r(i + i7) exp k(lcos6)r(lij) 2sin2(f) ?7 In sin 2 7 2 exp 2 i a r g r ( l + ij) — ij In sin ( ^ 2&sin2(f) .152) Since z = r c o s # .z) + la)T(n + b. (11.^l*7 l + O (11.149) (11. Without t h e logarithmic phase factors. i.i 7 ) (11. (11. a comparison of Eq.b.l. = r cos 0.154) . We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.e.e. The Coulomb Interaction W2(a.7 ln*(rz)) k(r . if tbr were „ikr ibr ex el"z + f{6) r (11.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].236 n\Zzab CHAPTER ^T(n '£ fo 11.
155) Since the gradient is proportional to the configuration space representation of the momentum. also called phase anomaly. the effective range of the Coulomb potential is infinite. (11.11. 11. the current density is * h .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. As a consequence of the masslessness of the photon.4 Variation of the observed differential cross section with 0. The quantity /(#) 2 would then be a measure for the scattering in the direction 9.Q hk mo v0. we see that this expression is the exact analogue of a current density in classical considerations. 11. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. For the incoming wave ipi = exp(ikz).n . that we encounter here is a characteristic of the Coulomb potential.7.. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. (11. We define as current density the quantity h J := 2imo [^(v^)v(v^n. 2im. The problem of the logarithmic phase. In other words.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity.156) . do dQ K 12 K Fig. z — — oo.
We see that a depends only on the modulus of the potential (i. i.e.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. We obtain the total scattering cross section by integration: „2 — /£<" = £/777*V sin 4 « 1 »«°) We observe that this expression diverges in the forward direction. for 0 —> 0. In reality this divergence does not occur. (11. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). 'screened ZlZ<2e2 cr/r0 ° . This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . (11. In Eq.159) as the Rutherford formula which can also be derived purely classically. We recognize the result (11. since in actual fact the Coulomb potential occurs only in a screened form. thereby screening itself off from the surroundings. Correspondingly the other part of (11.238 CHAPTER 11. is called scattering amplitude.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq. which yields the scattering cross section. in the current this would in any case contribute only something of order 1/r.152) as 1.153): 4fc2 s i r  The expression /(#).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. the square 7 2 ). That this formula retains its validity here in quantum mechanics is something of a coincidence. We make a few more important observations.e. 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.
11.8 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. whose maximum moves with the particle velocity. which in the case of free motion (zero potential) are simply plane waves. i. 11. of the equation r fi.e. / dk 7^340(k)4(xxo. i.t). t) = U(t. (11.t). of continuum wave functions.4. (11163) (11.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. be given by ^ f dk AL. 2mo V + F(x) k (11.yJ ^ j A k o W 727)3/2 e 0. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .i i f (''°>/Vk(x. The particle velocity v is v = — . m0 m0 and V. t 0 )^k(x. (11. The wave packet is a superposition of waves ip(x.11.164) Ek~Eko + {kk0)v0h. i.166) Let the incoming free wave packet at time t > t$.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0.2 fr2 .*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * .161) where Ak0(k) ~ A(k — ko) and ^ k (x. and if H is timeindependent one has the stationary wave function ^(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. t 0 ) . (k).^ ^ % ( x ) . and which provides the uncertainties arising in quantum mechanics.t) = e .165) ^ ( x ) = £ fe Vk(x).x° ' r j .e.e. t 0 ). 2 (11162) where Vk(x) = ^ ( x . the wave packet in the absence of a scattering potential. t0) = e. v0 = .
i) dk (2TT (11. this solution > has the following asymptotic behaviour in which f(6. With Eqs.e. the stationary scattering wave.ko) k0 (11. The Coulomb Interaction In Example 11. i.171) The phase of fk(0. (11.e. k = kez. (11.ik . (11.t) = J ^ A k o ( k ) e .167) we have ^k0(xx0. ip) could be carried along but we ignore it. i.166).154)) that for x — oo.173) .165) ko • (k .172) h h k0 {rv0(tt0)} (11.169) and (11.ip) is the scattering amplitude with respect to the incoming plane wave. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.163) we obtain </>ko(x .^ ( * . ^k(xxo) ^k(x) ~ = e.k 0 ) 2 ~ Jk20 + 2k0 • (k .169) The wave packet in the presence of the potential is obtained from the free form (11.161) and (11.t o ) / V k ( x .x 0 ).<p) + Ol Akr l (11. Then with k = ^(ko + k .xo.x °Vk(x).x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.10 we show that the solution V'k(x) of Eq.k0) ~ k0 + and the expansion of E^ in Eq.240 CHAPTER 11. (2TT)3/2 eik* + —fv(P.iB *(**°)/ R °'tj + 1 pikr e * . Eq. From Eqs. (11.170) A ko (k) e .167) by replacing the plane wave exp(zkx) by this scattered wave. (11.
*0ko (x . t) . E x a m p l e 11. Solution: We define the Green's function G ( r .11. 11.y)eiko'Xoe'[fcor£.fco(tto)/ft]^0o)(x/xo. and determine the Green's function G(r.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r.z Fig. A particular solution of the Schrodinger equation is — J G(r.(«)/ ipko(x . i. (11. U(T) = ^ V ( r ) .8 Scattering of a Wave Packet so that with Eqs. (11. 11.x 0 .to).r'). Adding a suitably normalized solution of the free equation. Here ^ (x' — xo. T h e scattering amplitude is t h e same as t h a t given by Eq.167). (11. *. t) f.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11. however.^y /k0^. (A + fc2)t/.154).to) has t h e form of t h e wave packet in radial direction as indicated in Fig. '.r')V(rW). r ' ) by the equation (A + fc2)G(r.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.x 0 . .5.r ' ) .5 Scattering of a wave packet with scattering centre O.r')l/(r>fc(r'). r ' ) = 47r<5(r .171).e. the solution can be written iMr) : (2TT)3/2 ^dr'G(r. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.r')U(r')ip{r')dr' /4T7 (as we saw earlier) .174) r /k0(^. with the energy of the maximum momentum of t h e wave packet.
one obtains another Green's function given by G_(r)= r eikr.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist.=k+ie/2k 1 d f. 11. displaced slightly away from the real axis as indicated in Fig. Thus we consider the contour integral taken around the contour C+./2 dk'd(—cos 6)dip iki. Imk . 11. dk' ik'*r 2TT2 fc'2 . The Coulomb Interaction We find a set of Green's functions G(r.Rek' kfc/2K Fig.6 The contour C+ in the upper half of the k'p\ane. These are the socalled outgoing and ingoing Green's functions. Hence we write G(r)G+(r). which is such that the contribution along the infinite semicircle vanishes. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k).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 . 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. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . e > 0 small.242 CHAPTER 11.6.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. .fc ) oo dfc' 1 d nik r 7rr dr / .
g. where Mo is a parameter in Chapter 16. e. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. 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. z) + W2(a. whose continuation to infinity is again given by (cf. If we proceed from the radial Schrodinger equation. z) = W^a.2l + 2. under the stated conditions 2/07 = Z\Zie2 = —Mo. Then £^<fc + (2l + 2 £ = ~2ikr. z).176) " 0—A ~ (I + 1 + iy)<t>i = 0. (11. . b.11. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. Then yW (11. b.140) (previously we solved this only for the bound state problem) § . 2ikr)] [Wi{l + 1+11.z2 27fc 1(1 + 1)' yi = 0. b. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. Before we introduce the general form of the socalled partial wave expansion. Eq.e.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij.0 + W2(l + 1 + iy. i. (11. Eq. (11.e. (11. (11.£).145)) F(a.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. Thus we proceed as above. 21 + 2.9 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.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). i.2Z + 2.9 Scattering Phase and Partial Waves 243 11.21 + 2.178) (11.
n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .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 . . T(2l + 2) (2fcr)*T« (Alliy ikr .«7) „ .180) ^ ' (11.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .' . (2kryr{i" V 7 . +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .ii)' (11. Then e2i5< = Reie Re~ J2i9 9 = Si.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.183) r(Z + 1 + 17) := Reie.244 CHAPTER 11. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.181) so that eldlT{l + 1 .l e*"" + T(l + 1 + H ) T(l + 1 .146) and (11. The Coulomb Interaction Using Eqs.j syyll iy il( _ . (11.ry) = e"^r(Z + 1 + ry).
Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . That the energy E appears in Eq.139) n' = 0.e.n ' .. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. considering I as a function of E extended into the complex plane. with Eq. (11. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.11._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). . considering I —• an(E). . (11. g = e«* = rSi + 1+ *V*ri (11. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.114a) reveals). i..e. a positive integer.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). i.2. For instance the poles of S are given by Z + l + z7 = . i.1.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). where the continuum of the spectrum starts.e. (11.188) in the form y/E. i.e. kr) the regular solution with the following asymptotic behaviour: r 0 (11.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase.187) contains almost the entire physical information with regard to the Coulomb potential.. In the present quantum mechanical considerations I is. One defines as regular Coulomb wave Fi(j. 245 (11. However. as we saw.
we obtain trajectories which are known as Regge trajectories. 11. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V.7 A Regge trajectory of the Coulomb potential. Singh [252]. In these cases.7. we obtain the partial wave expansion of the scattering amplitude as follows. (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. and thus in general.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. (11.191) . For the Coulomb potential a typical such trajectory is indicated in Fig. The Coulomb Interaction and plotting these for different values of n.246 CHAPTER 11. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt. 11.
k)=me2iS^k\ in which r\i^\ indicates absorption. (11.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.196) "See e. Oberhettinger [181]. Legendre functions) and its asymptotic expression for r — oo and z — r cos 9 is given by" > eikz ^rJ2(21 + 1 ){e i i ' r e.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. . W. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg . The potential disturbs this phase relationship (by delay or absorption or redirection).195) with the scattering matrix element S(l. (11. 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„ ' .r ^ V ^ Wostf). 21 .i f c r .~ J > Z + l)Lil*e~ikr . 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 ' . The mathematical expression can be found in Tables of Special Functions (there cf. (11.g.^2l + l^S^k) lik (11. Magnus and F.11.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—.192) Thus the incoming and outgoing spherical waves have a definite relative phase.eikr\P^cos9).22. pp.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. r e. The incoming plane wave can be reexpressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase.
This page is intentionally left blank .
Similarly difficult is. states of a finite lifetime. These newer methods will be presented in detail in later chapters. It is very instructive.Chapter 12 Q u a n t u m Mechanical Tunneling 12.1 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.g. can be found in any traditional text on quantum mechanics. that the explicit derivation of such quantities in nontrivial contexts. which a particle in classical mechanics would never be able to intrude.e. in texts on quantum mechanics is rare. which in one form or another. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. It is not surprising. in general. therefore. the computation of finite lifetimes of a quantum mechanical state. i. as — for instance — for trigonometric or double well potentials. that quantum mechanically a particle may have a small probability of being in a spatial domain. Such a vital quantum mechanical phenomenon is socalled tunneling. one of the main and now wellknown methods to perform such calculations.e. For various general aspects and applications of tunneling we refer to other literature. Razavy [236]. i. In fact. namely the instanton (or more generally pseudoparticle) method. was developed only in the last two to three decades. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances.* *See e. however. to explore first much simpler models in order to acquire an impression of the type of results to be expected. 249 . M.
t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.e. According to Eq.4) (12.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. (12.250 CHAPTER 12.1) as the probability described by ijj(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ. M^ Then dt + Vj 1=0. t) was developed by Born in 1926..t) = HiP(x.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. so that p = T/>2 is the probability density for a spatial measurement at time t.t).t) (12.t)\2 =1 (12.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. ihr(x. i. . r (12. Quantum Mechanical Tunneling 12.t) that the one particle concerned is somewhere in space at time t.
In much the same way that we define there reflection and transmission coefficients.6) . From this follows that both p and j have to be finite and continuous everywhere. If there is no absorption. Gosdzinsky and R. 29 . with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. t For the singular delta function potential a modification of these demands is necessary. 12. where „ mpVo V(x) =+V0S(x). as we mentioned above. 1 0 ~x o a= f See L. Tarrach [119]. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous.33. Schiff [243]. we can do this also here. Nonetheless.2a6(x)]il> = 0. This case plays a special role. although in general small. If ^>(x) and V^>(x) are known at every point x.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. which is the case we consider here. 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. ~hr> k = n^ (12 7)  . the equation can be integrated at every point. The Schrodinger equation for this case is* ip" + [A. (12. higher dimensional delta potentials do not have properties like that in one dimension. Delta potentials in more than one dimension do not permit bound states and scattering. even for discontinuous potentials. However.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x.12.2 . The wave phenomenon which permits transmission and reflection is familiar from electrodynamics where the laws of reflection and refraction of optics are derived from continuity conditions at the interface between dielectric media applied to the fields E and B of Maxwell's equations. Such a region is for instance the domain where V is larger than the total energy E. the problem is analogous to that of the scattering process from a potential in a onedimensional case. as shown by P. This effect is known as "tunnelincp. pp. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). Thus it is possible that the probability for the particle to be in such a domain is nonzero. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. 2m E 2 V0 > 0. 'Maybe the reader dislikes being confronted again with the onedimensional case. 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.
8) (12.e.: (i2iQ) Ce for x > 0. (12. i. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). (12.= 2a^(0).12a) (12. a 2 — a 2 a — ik a2 + k2 ' . e Q (12.e. [<//]0+ . Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.11b) 2aC. 2 (12. In the region x < 0 we then have both types of waves: the incoming or incident wave to the right as well as the wave reflected from the potential.' : It follows that ik .ikC = i. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .252 CHAPTER 12.(ik . i. (12. The continuity conditions at x = 0 are therefore for V : and according to Eq.1) .9) for ij.[^']o. i. a — ik £Qr < { eikx a eikx •t" .12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—.e. except in domains of the potential V.13) ~A R= for x > °(12.ikB) = 2aC or 2a(l + B). (12.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes. . _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .11a) ikC .9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.ik(C .2 a J 5{x)4>(x)dx = 0.e 1 + B = C.
2 a C .2 = . the quantity T.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. = .ip'(0) = . Omnes and M. ( 1 2 . K = a. Proissart [223] or other books on scattering theory like R. z<0. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. and C follows from the normalization: OO A.a l s l. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. possesses simple poles at a ± ik = 0. .e «x = J \ CeKX for for x > 0. 2m 0 (12. when normalized to 1. oo (12. i. R.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.(K)C = . See e.i 8 ) . at k2 = —a2. Newton [219].19) Thus the delta function potential supports exactly one bound state. 'Note that the physical pole has negative imaginary part.K C . We write the solution ^ so that ^(0+) = ^ ( 0 .e.K 2 < 0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.2 o ^ ( 0 ) . i. or h2 E = —a2. f Ce~KX c. (12.g. . as coefficient of the outgoing wave.17) da#(s) 2 = 1.12.e. Thus ^ = Ce.) and V>'(0+) .
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. x G [a.1.22) + De Feikx + Ge~ikx *We follow here largely F. or in the lectures of E. 12. \x\ < a. p. 12. p. (12.20) for for \x\ > a.4 Scattering from a Potential Well We can immediately transfer the above considerations in a similar way to scattering from a potential wen* as indicated in Fig. 55. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.1 The square well potential. 58. Schwabl [246]. . We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. Fermi [92]. oo]. —a]. x€[a. The latter contains also specific applications in nuclear physics. V(x) V=E a v=y Fig.254 CHAPTER 12. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.a]. We want to consider the case of stationary states for E > 0. Quantum Mechanical Tunneling 12. (12.
27) v ' Similarly Eq.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.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.f )e" t\ K\0—tKa—ika . ( V C n vD ).26) K^gifoa—i/ca fc.12.e.4 Scattering from a Potential Well 255 The connection relations obtained by equating ip(x) and its derivative from left and right at x = — a and a yield the equations Aeika + Beika = CeiKa + Df. (12. M(a) .ikBeika = inCeiKa + DiKe~iKa.24) yields F ^ G. .25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12. Fikeika . (12. (12. (12. 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . ).ikGe~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 \ . fc Af(a) = . ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. I (12.iKa ? ikAe~ika and .
i. (12.(1 + e 2 ) sin2 2na 1 .—2ika [cos 2/ta + \e sin 2«a]e~ (12. xe[a.e.a ] .33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. we obtain A = F( cos 2na — .a . \R\2 = T(E)^sm2Ka (12.31) we have A AT(a) = )MFo B ^r)sin2na . = < Ce~lKX + DeiKX .e sin 2na IJlika e B = r)F sm2na.32) [cos2«. Quantum Mechanical Tunneling Hence with the complex quantities IK (12. (12.oo].34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .256 CHAPTER 12. x € [a.^esin2«a]e 2lfca \r\ sin 2/ta Thus if we now set G = 0 in order to have only an outgoing wave proportional to exp(ikx) to the right of the well. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.35) with with with x G [oo.37) .a]. i.34) into Eq. (12.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.4T(£)e ifci. (12.e. .
. 83.e. 2\k K.12. and k real.12. Dwight [81]. We see in particular from Eq.e. (12.4 Scattering from a Potential Well On the other hand according to Eq.e.38) as expected. cot(fca) = i— or tan(fta) = K K . (12.39) 2 i. an equation of the type f~1f = g19.41) i. (12.43) Thus we set k — —i ' H . However. (12. they have solutions. 2m0\E\ h? ' (12.44) . for VQ < E < 0.31): 1\T(E)\2 257 1 . (12. (12.40) we obtain cot(na) — tan(/«j) — i K IK . Li ( — H— I sin 2na = 0.e.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. which is satisfied if either / = g or / = — g'1.35) and using Eq. in the domain of A imaginary and K real. B. i. (12. (12. nor when both parameters are purely imaginary. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)].34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. p. formula 406. i.42) These equations have no solution for K.
i. .+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . The remaining wave part becomes exponentially > decreasing. i. (12. Quantum Mechanical Tunneling .l ) n . 2Ka = nn.g. Schwabl [246].258 and obtain the equations CHAPTER 12. A = 0. (12.49) Taylor expansion of the denominator in powers of (E — ER) yields .e. This can be understood.fl = ( . The amplitude of the transmitted wave assumes its maximum value where (cf. Also. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. (12. . (12. pp.47) These states in the continuum are called resonances. t&n(2Ka)ER = 0. . (12. (12. F. t2 2 (12.46). 2m0\E\ and Kcot(Ka) = y—^—L.' We observe that T(E) has poles at those values of E which supply binding energies.20). . ± 2 . around the values given by Eq.65.c o .48) We expand T{E) around these energy values. = 2/rn (EER).42) which then leads to bound states with associated even and odd wave functions.e.34). Eq. T(E)e2ika = — r . We now return to the scattering states. (12. 64 .e. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — .l .35)) for E > 0 sin2Ka = 0. 2m0\E\ Ktan(«a) = — y——^— .45) These are the equations that one obtains from the connection relations for the case of Eq. ± l . n = 0 .46) This condition implies the eigenvalues E^ER = n2^^ V0>0. „ . (12. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. Then according to Eq. (12. i.50) See e. from Eq. and is itself infinite there. At their positions cos(2/«z)£. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.
e. We restrict ourselves to the period — la < x < 2a. This result is physically very plausible.12. s \ V0 /• ~\ 'X . 12.52) Here ER is the energy of the resonance and 2/F the lifetime of the state. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.5 Degenerate Potentials and Tunneling In the following we consider a potential which is very similar to a periodic potential and can serve to illustrate a number of aspects of the process of tunneling in quantum mechanics. 2a<x<2a.' ' 5a \V / Fig. V(x) x=2a x= =2a . T(E) has poles at E= ERi: (12. . V(x) = 9(x + a)VQ6(ax).' ''' 5a 3a \ a ~~~. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. These are precisely the resonance states.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. 12. a 3 a. .2 The PenneyKronig potential.2. 12. = 0. A potential of this type is known as a PenneyKronig potential.^ l / .51) i. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i.
. = EVip(x).. the differential operator is even in x.. HV^(x) = H. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r.e.55) The Hamilton operator of the problem. and hence for = Eip(x).2.2 < a . i. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\. (12.53) *w«w^( « j)'*^{i i ^ ..1. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.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. . with V as parity operator. i..e. .Vo) n. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo. The particle of mass TUQ whose quantum mechanics is described by the Schrodinger equation is free to move in the domains under consideration to the left and to the right of the central barrier.260 CHAPTER 12. (12. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). and we choose (12.e There is the additional boundary condition (for Vo = oo) ^ ( . E^En = ^ ^ t n = 0. In view of the periodicity we have ^(x) = ^(x + 4a). We set .a ) = 0 = ^(a).2 = W h 2 _„2 = 2mo(E .
V'(O) = const. of course. we shall not exploit here.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . (12. (12.2a]. that these are such that for VQ — oo they approach those of the previous case.Be Ka (12.il>]?0. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ .5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.a]. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K.63) = KBe~ F = . At x = ± a we now have to connect ip and ip' of neighbouring domains. (12. The central region of the solution is the classically forbidden domain.a]. We > obtain even and odd functions ip± by setting ^ = ±1. (12. 12. however. — a].2).58) D sin k(3a — x) with xE[a.59) Asink(x + 3a) with x & [~2a. V+(0) = const. xe[a. We demand these for even and odd wave functions ip±(x) which we can clearly construct from ip(x) (as examples see the dotted lines in Fig.2a]. (12..60) The boundary conditions that ip± have to satisfy. ±Asink(3a — x) with x€[a. are ^_(0) = 0. which. We expect. = Ka ±Asin(k2a). This wave function is no longer restricted to the interior of a box. This means that i>±(x) = <  = ±1. — a]..12. We therefore have different boundary conditions. so that (W meaning Wronskian) W[tl>+. BeKX±Be~KX with xe[a.2a. VV(°) = °. ^kAcos{k2a).61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].
F ^odd (12. In general one now rewrites the equations in matrix form.. = AA. where Akoce'2™. i.+ • • • 1! cos^(n7r) 2ak . However.e. in the present case it is easier to continue with the above equations. and is — as we see — less than that belonging to the lower sign case. For large values of Vb.2 a .262 CHAPTER 12.67) The value of k belonging to the upper sign. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.3 Level degeneracy removed by tunneling. i. (with reinsertion of nir/2a for k) 2ak ~ nir .Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. Quantum Mechanical Tunneling Fig.even. 12.nir)2}} = 1 ± 2e~2Ka. + 0 ( e . i. (12.mr)2}. (12. is the one belonging to the even solution.nir + 0[{2ak . say A. for VQ — oo.e. (12. k = fc0dd.65) and we expand tan(2afc) about nir. (12. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo.e.e.2 e .66) Inserting these expansions into Eq.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .— (1 ± 2e~2A + O (—).nir) + OU2ak . the odd solution with.64) we obtain ~[(2ak k . Taking the ratio of the equations in both cases. i. we set tanhx = l . say.68) . It is plausible therefore to use appropriate expansions. 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.
V ^ 0 are given by the equations ^o + " J . (12.70) xo = (a + b) 2' .4. .1: A car's quantum mechanical probability to pass a bump A very slowly moving car of mass mo (kinetic energy almost zero) encounters on the road between x = a and x = b. and is a reformulated version of a problem set in lectures of Fermi. 0. Example 12. Solution: The wave functions i/>o>Vv in regions V = 0.^ 0 = °' Hence in dominant order •00 — exp .12. . 12.3. Compute this probability. 12. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. exp .. '. . 1pV 2ro 0 4>o '.e.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.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. Classically the car is unable to overcome the bump. w(x0) •• = = mogy(x). 2mn.. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. 57. exp r rb rb / •J a 2mo (V(x) . ^v + ^[E. b > a. but quantum mechanically there is a finite probability for this to succeed.v = 0. y(x) = sin and y(xo) = 1. . p. Fermi [92].2 dxy/V(x) (12. i.4 A car meeting a bump in the road.** " W ~^7~ Fig.V{x)\i. as indicated in Fig. 2m0E „ 2mnE . 12.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states.
**.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. (12. . p. We find™ (with ( . 369 and 8.17)). n I . ^ ) = 2 ^ . **For instance E.l .1.1 / 2 ) ! = v^r) TT/2 /.264 we have ro / dxyfV(x) Ja CHAPTER 12. which give /•TT/2 dx sin'M . S. 950. formulas 3. Jahnke and F.807 m s . h = 1.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. we obtain ( .4 x 10 3 9 ]. p. Ryzhik [122].y) is the beta function (see Eq. 14. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. (15. one evaluates an approximate probability of exp[6. Gradshteyn and I. We can now evaluate the probability. p. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w.1.055 X 1 0 _ 3 4 J s .621. and assuming a length of b — a = 100 m of the base of the bump.9191 ~ 1.2 .e.384. : 2 ^ B ( ^ .71) The integral can be looked up in Tables of Integrals. Emde [143].2.72) Assuming a mass of 1 ton of the car (i. 1000 kg) and using the following values of the natural constants: g = 9. M. dx sin ' x •• Looking up the value of the factorial in Tables.
in Chapter 15. the last case allows only a discrete spectrum. 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. * A highly abridged treatment of this problem is given.* In Chapter 14. for instance. Then considering the probability distribution determined by the squared modulus of the wave function. This is the problem of Galilei in quantum mechanics. Siissmann [266]. whereas the first case permits only a continuous spectrum. Finally.Chapter 13 Linear Potentials 13. we consider the linear potential in three space dimensions that forbids the particle to escape. p. 265 . 13. 144. we consider the corresponding case of a particle above the flat surface of the Earth. each of which propagates with the classical momentum and energy of a Galileian particle.1 Introductory Remarks We distinguish here between three different types of linear potentials. 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. Thus.3. in the book of G. 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. 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. Example 14. This is the quantum mechanical version of the problem of Galilei in one space dimension.
= l V 27T J_oo 1 dkeikxijj{k. (13.1 S u p e r p o s i t i o n of de Broglie waves We consider a particle of mass mo falling freely in the timeindependent homogeneous field of the gravitational force F(x.1).t) = mogx + V0. V(x) = —mogx. g > 0.t).= l V 27T Joo dxelkxip(x.t). Therefore we use the Fourier transforms il){x. The equation of motion follows for instance from the derivative of the constant energy.266 CHAPTER 13.t) = . The present case is an exception and a good example of its applicability.2. and yields rriQX = mog. i. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.t) = c\x\ + V0. 13. (13. We see " that the case g < 0 can be related to the reflection x — —x. (13.t) = m0g with potential V(x. c> 0. Recall the classical treatment of the freely falling particle. (13.t).2) which is the case with a discrete spectrum. x > 0. A different > "linear potential' is the socalled "confinement potentiaF V(x. We consider here the case of Eq.t) = . (13.1) where g is the acceleration due to gravity and we can put Vo = 0.3) In general the momentum representation is of little importance.e. (13.4b) . from d_ m x2 0 dx + V(x) 0.4a) f°° ${k.
(13.t) m0gx /2TT m0gxip(x.8) With this result and the substitution (13. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt.< (13.t).6) the latter equation becomes d ~( m0gt ^(^M<=+=?. (13. Equation (13. (13. i.6) the argument k is then replaced by this. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.6) and wish to solve it.e.t) h2k2 d ~J(k.7) In the wave function on the left hand side of Eq. • fd ~(' m0gt %h .6).3) yields us now by inserting for ip(x.t) dk nikx k=oo '2K (im0g)il)(k.t) the expression (13. Applying the total derivative d/dt to this function.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.t) dketkxiP(k.d m +imo9 .6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.e.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.t).5) m+tm9Wcr{ktt)=2^^ktt)t (13.7) applied to every quantity in Eq. We rewrite this equation in the following form th (13. d \ ~f1 dkr[ k + m0gt ^r't (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.13. (13. multiplication by a test function and subsequent integration).9) . we obtain the differential operator on the left of Eq. (13.t)im0g^k. To this end we consider the following substitution in Eq.
12) .t).t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13. Our next step is to reexpress the solution (13. Thus we have first (the second term arising from the upper integration limit.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.6) to the solution (13. (13. i.6). Solution: We apply the two operator expressions on the left of Eq.11).6) is then $(k.268 CHAPTER 13.i) .6).1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. the following result h 2moi tp[k\ ^—. (13. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.t In Example 13. t ) . E x a m p l e 13.11) solves the partial differential equation (13.10) Note the amplitude function ipo(k) in front. On the other hand the second operator expression implies dtj>o imog^—ip(k. Thus addition of both indeed leaves the expression on the right hand side of Eq.gh [ d t ' j f c + ^ ( t ' .1 we verify t h a t this solution indeed solves Eq. We define the wave number kt as kt. (13.6). we have to reverse the substitutions and replace in (13.t ) ' U ( f c . (13. (13.11) in a neater form.10) k by m0gt back : k —> k h The solution of Eq. We observe t h a t t h e solution is actually a function of kt — k — mogt/h. We have thus obtained the solution of the transformed equation.e.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.11) *[k?f.= k rmt.6). To obtain the solution of the original equation (13.
11).t) defined by Eq.17) = J_/ C X ) h2 [It(k) .15) we obtain / = .t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . 27T ^ . . we have ip(x.t) into this integral and recalling the property (13.2„2+2 mfaH 3^3+3 .13) rn0_t+2 + mlg2t3 .t) = . Inserting ip(k.o ut = 2mo —kf 2mo A. (13.+^f (13.I0(k)].2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.13.18) h =k mogt p h S' h .4a).} fA ' _ _tkt 2 k+ t) 2 k . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.19a) . we obtain rp(x.gfci + 2 g2m0 2/i 2 ' (13. m0g (13. (13.o o 2 —  exp iktx h mZgH' .14) (13.16) Next we evaluate the Fourier transform of ip(k.— [It(k) ..11) can be written j'A ^. _ k t  h2 3 Inserting here the explicit expression for fct. . we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.
Linear Potentials f dt'.23) .18) into a • different form in order to be able to perform the integration with respect to k.o o Jo This result can be interpreted as follows.22) We can rearrange these terms in the following way with the ^dependent contributions contained in a quadratic form: ht 2moi ht 2moi k mpx ht mogt 2h mogt\ 2h J + IX .ut> = Jo t2 g2m0 i 3 k t . 7 u>t'dt' (13. fm0gt\ \ 2h ) (13.20) (13.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x. (13.v = kh/mo.t)\2 for t — oo (distant future or past).19b) h3 3 we can rewrite ip(x. Such a decomposition is possible only for homogeneous force fields. (13. ipo(k) is the probability amplitude with wave vector k at time t = 0 and varies in the course of time with the timedependent wave vector kt = k — mogt/h.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .270 and CHAPTER 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. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13. To this end we convert Eq.20) expresses the wave function tp(x. 13.21) In other words.t) as 1 I00 ip(x.19a) exactly as for a falling particle.t) as a decomposition into de Broglie waves.The associated kinetic energy of the particle is mv+ = huJf.2. the relation (13. (13. These wave vectors vary in accordance with Eq.t) =y= dkip0(kt)) exp i< fctx V27T J .
QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.QX 2 ^ 0 7 (m.t) = 1 1 oo 271 27T /.26) with the formula 1/2 (13.28) (for a verification replace a by a + ifi.24) We now set fit / 2m 0 V" 0 V m0a. 2h h 24ift ' mogt^ r m./m 0 a. fi > 0. being now independent of £. Then ip(x. may be put in front of the integral. (13. (13.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.t) 1 exp r .t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13.18) thus becomes ip(x. /m0gt\ \ 2ft J ° ) I m09H3 ft J 6ih m0g2t3 _ m0x2" 2Aih 2hti m gt IT / dk^oik t) exp oo ht k2mo« ~fti 2ft~] (13.13. We evaluate the Fresnel integral in Eq. m0xcl ht m0gt 2h gt O v: °° 2hti (13. It follows that ip(x. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f.29) . rnogt m0g2t3 tX^r~.e. (13. integrate and then take the limit fi —> 0).25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ .2 The Freely Falling Particle: Quantization The integral (13.26) d£e ie £ kt^ m0' 0./oo dkip0(kt) exp l x ixI exp oo . TOO#A ft and.
31) and obtain from Eq. . These are incoming and outgoing waves.t) = eiEtlhct>{x) (13. kt and ut 13.1) has only a continuous spectrum.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i. (13.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. For stationary states we set xl){x.t) ~ W —exp i V irlt m ktxd J cot. V(x)=m 0 gx V=E Fig.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . 1/2  ±^(2mlg/h^x^ / 2i (13.dA 4>o(h (13.e.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. For x exponentially increasing and decreasing solutions are possible. this expression can be brought into the following form: if *\ 0 ip{x. Linear Potentials Replacing here x by the large time value x c i.33) oo for E <C \mogx\. xc\).272 CHAPTER 13. 13. (13.1 A turning point at x = —XQ.
The Fourier transform is given by the relations (13. i.t). Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. but also d 00 dt Eijj(k. (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. i. t). which is not possible for real particles.t) = dt (13. at which the classical particle bounces back. is reflected. The point — XQ. so that ih dtp{k. i. 13.36) m0g \ 6m0 with c = const. 1 f01 dketlcx^{k. i. exponentially decreasing.e. probability.4a) and (13. E)dk. we .t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m.34a) Then d &k 1 f°° ih—il>(x.e.35) It follows that /W tm0g \ 2m0 i.3 The TimeIndependent Schrodinger Equation 273 Classically the particle coming in from the right with energy E cannot penetrate into the potential barrier since (where V > E) with conservation of energy E — V = p2/2m. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation.1.13.6) that also for 4>{k. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. Fig.00].e.t). t) = Eil)[x. cf. at which E — V = 0.0 0 dkelkxE^(k. i.34b) It therefore follows from Eq. Akx (13. However quantum mechanically this is possible with a small. t) = — = / V27T J . and hence is continuous and extends from —00 to +00.e. H2k3 4>{k) — cexp Ek (13.4b).e. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. J \ „ (13.e. its momentum p would be imaginary there.
Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). / <>~ikx VZ7T J (13. (13..L [ dkeikxMk).38) we have for <j)E{k): OO /"OO / oo (13.39) dk4>*E{k)4>E. (13.43) y=2^x+2^E.40)) into (13.M *K Joo Setting dt cos ( ts n Jo st).a : ' ) . </>E(x) = .(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.37) (eikxdk.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.36) into (13. (1340) i.36) (with c replaced by (13. In a similar way we can verify the validity of the completeness relation. V 2lT J Mk) = . i. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').e. (13. / dE(j)E(x)(t>E(x') = ^ s .£') = <*(#£')• 1 27rm0g' = /o .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 .41) J—oo Next we insert (13.e. apart from a constant phase which we choose to be zero. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).38). h? h2 ..39) we obtain dkcc*exp m0g It follows t h a t cc = &(# . 13.L dxeikx(f>E(x).
J * . and we shall see t h a t one branch has periodic behaviour.46) This follows since with Eq.45) One can now derive normalization and completeness integrals for the Airy functions.13. (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. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s.^x + ^ E (13.44) is then expressed in terms of the Airy function: . For a further solution Bi of Airy functions we refer in Chapter 14 to Tables. whereas the other is of exponential type. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential.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.^ . / s M ./ 2m 0 E (13. (13. . the function 4>E{X) i and setting s y = — = 1 .. 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= ^ .
51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.43) contains the phase (with x — z): > *(*) := sz . (13.. Thus we must have sin30<O.49) (13.e.r 3 sin 30 ).. (13. .51b) .TT < 6 < 0 for r > oo. i. o (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . 13.4 The Saddle Point or Stationary Phase Method The integrand of Eq. (13.r 3 cos 30 J + i ( sr sin 9 . or n <6 <TT o for r > oo.f dzJ*W 2TT JC with z = reie.\zz.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig. (13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 .276 CHAPTER 13.2 The contour C for s < 0 with ends in the angular domains (a) and (b).e.TT < 30 < 0. (13. . i. Linear Potentials 13. We want the integral to exist and so desire that lim e**^ = 0.
( \ / ^ ) 3 2z. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.51b).e. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.55) into (13.13. (13. (13. (13.S )V2 (13. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. 13.51a) and (13. and the main contribution to the integral comes from the a region around the saddle point.53) Since we shall have the contour C in the lower half plane.45) exists for the contour C.55) s < 0. —i ${zs) = szs .58) .z3s and V^s + . we choose zs = iy/^s. Such a piece can be found in the neighbourhood of the socalled saddle point zs. oo] as indicated in Fig.4 The Method of Stationary Phase 277 The integral (13. Inserting (13.e. provided its ends at infinity satisfy these conditions.56) 2n(s)^Jc But I daefr"?*"^.54) where a is the new variable with —oo < a < oo. and we set with a real: z := zs + a (S)V4' (13. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. i. (13.54) i ¥~s)3/2' (13. so that exp(±2i9s) — 1 and — s — rl.2.52) = exp(±i7r) Let s be real and s < 0. i. (13. Then 6S = ±7r/2. Then the ends of the contour lie in the domains specified by Eqs. at which the phase becomes stationary. 0 = &(z8) = s~zt s = za „2„2i6>. where the derivative of the phase function vanishes.57) $"(z s ) = 2% s.48).
64) tSee I.60) The reason for describing the point zs as a saddle point is that the integral in Eq.32) with the solution (13. V 2h2 (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. (13.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). V^s / 4 V3 (13. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction.44) we can convert Eq. (13. (13. M. With the help of the substitutions (13. > so that Ai(s) 2 v ^ ( . 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). cos . ._ 1 . For (—s) — oo the integral converges towards T/TC. 2 s 3/2 .. This is an important aspect in the WKB method to be considered in Chapter 14.s ) V 4 exp \(s) 3/2 (13.44) and (13.32) therefore becomes dz2 + zd) = 0. 8. Sec. According to Eqs.2/32u = 0 (13.59) In Example 13.491. Gradshteyn and I. (13. S.2 we show that for s — oo: > Ai(s) 1 . Ryzhik [122].61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.56) decreases exponentially around a — 0 for real values of a.278 CHAPTER 13.42) into an equation with a more appealing form. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.45) 4>E(X) ~ Ai(z). (13.
changing to the variable p. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. we obtain f° Joo dp=—[a+ i\a\]da.51b) are satisfied._ . . Then.z ± ) 2 * " ( z ± ) . with *"(z) = 2z. * " ( z ± ) = =F2Vi. It follows that for s —> oo: Ai(s) ~ . We have therefore r /*0 poo zs = ±y/s = z± / Jc dze^M = / Joo dze^^ + JO dze^z\ In these integrals we set respectively P with *(z) = #(z±) + i ( z . the conditions (13.60) 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 14.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).13. 27T Ic Jc 3 This time s is real and positive.^ e ^ / V2 4 [°° Jo dae'2^.4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13.51a) and (13. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .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. Eq. Choosing the contour C to lie along the real axis. (13.59) and (13.s 3 / 2  . Example 13. Thus there are two symmetrically placed real saddle points on the real axis. P Introducing a new variable a by setting p = ai\a\. cos  ..
(11.280 CHAPTER 13. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en.*. 13. i. ^/2KJW1. since f'(zo) = 0 and (z — ZQ)2 = —x2. Linear Potentials E x a m p l e 13. n!= / e~zzndz= / e~z+nlnzdz.e.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. Thus here / ( z ) = — z + n l n z . Jo Jo \ .dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point).Be 2 Fig.111). Solution: Briefly. / " ( z o ) = —n/z2 = —1/n. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. and hence integrating as described above through the saddle point indicated in Fig. Then. 13. .3 The saddle point at z = n + ix. ~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq. z o = n .3.
Dingle [70]. will there be that of a coupling parameter. O. § See e. i. been developed previously by others in different contexts. however. Brillouin [37]. It goes without saying that such comparisons are very instructive.e. 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. N. Froman and P. ^ The central problem of the method is the topic of "connection formulas".295. and the role played by h in our considerations here. A critical overview of the history of the method. 292 .* Descriptions of the method can be found in most books on quantum mechanics and in some monographs.317. named after its first promulgators in quantum mechanics: Wentzel. 281 . for misunderstandings pp.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. Froman [99]. Jeffreys [144]. of misunderstandings and other aspects may be found in the monograph of Dingle. where *G.g.§ 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.Chapter 14 Classical Limit and W K B Method 14. H. Kramers [153] and L. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. B.* The method had. Kramers and Brillouin. *R. For the history see pp. ' T h e most prominent earlier expounder is H. Wentzel [282].e. A. The WKB result is therefore frequently described as the semiclassical approximation. 316 . the linkage of solutions above a turning point to those below.
* We illustrate the use of this condition by application to several examples. Classical Limit and WKB Method E — V changes sign).V(r)]^(r) = 0. 14. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.1) (14. A(r) and S(r) being real functions of r. the derivation in Example 18. V>(r) = A(r)eiSWn. (14. In this equation we set. the function S being called phase function. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T). For stationary states ^/. for which the observation of position and momentum is independent of the time at which the observation takes place. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E .g. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e.3) . the energy has a definite value E with V(r.t)\2 = \iP(r)\2.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh.t) = eiEt/h7P(r) and \V(r.7.282 CHAPTER 14.
V(Y) + and 2VA • VS + AAS = 0. (14. We set in this equation.13) .4) (14.Q A 2i 8A 1 (14. (14.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r. These equations are equivalent to the Schrodinger equation.14.10) m 0 ^ . (14.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.12) VA A (14. (14. we obtain the following two equations: E .t)}2.* VAVS+AAS + V(r)A.6) 2mo dt now with A also a function of t.10) with a continuity equation by defining as probability density p:=*** = [A(r. (14.1) and separating real and imaginary parts.t)eiS{r't)/h. V(r. h2 ih— = H$ A + V(r) * . We can identify Eq.+ VAVS + AAS = 0. h h (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.
Since we obtained this equation from Eq.10). mo <™> (14.J = 0. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq. Since J = — A2VS = PVS.20) With this we can rewrite Eq.284 CHAPTER 14. we have mo—(A2) + V(A2)VS + A2AS = 0. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real.14) A2VS. m0 But now (14.21) . (14. (14.9) for the phase function S.15) and (14. + h2 AA V = — ^ .01. (14. (14. (14.16) this becomes m 0 1 rr + m o V . .17) With Eqs.10) we obtain = — [V(A2) • VS + A2AS}. TTIQ (14. (14. (14.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. the implications of this equation are also those of the equation of continuity.n.+ AVA • VS + A2AS = 0. ie. (14.. ^ + V • J = 0.16) BA 1 moA—. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. (14.9) as 9S 1 mo^ 2 + T. Next we investigate the significance of Eq.18) ot ot This is the equation of continuity that we encountered already earlier.
v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. In the first • place we require the ansatz (14. The particles have no interaction with each other. d — = (14. and hence''' ^(m 0 v) + m 0 (v. (14.24) d 1 dr d •— = d Lv _ .23) (14. with the help of Eq.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. We deduced this equation solely from the phase S in the limit h — 0.22) (14. where. which for very small values of h yield the dominant contributions.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. 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.20). . This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.14.V ) v + V 7 = 0. i. the wave function \I/ describes effectively a fluid of particles of mass mo. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). The equation of continuity then describes the stationary flow of a fluid. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0. This expression with h in the denominator shows. V • J = 0. However. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. so that V ( v • v) = 2(v • V ) v + 2v x rotv. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.e. This means. (14.1/h. This result is very general. they are asymptotic expansions. dt 2 Constructing the gradient of this equation. a singularity different from that of a pole).e. that the quantum theory implies an essential singularity at H = 0 (i. Such series are typical for expansions in the neighbourhood of an essential singularity. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. we obtain ^ V 5 +  v ( v v ) + V V = 0. r o t v = ( l / m o ) r o t g r a d S = 0.2) for the probability amplitude.
. B. We consider the onedimensional timeindependent Schrodinger equation y" + ^. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V). (14. There are also other more specific descriptions. (14.e. the method is widely familar under the abbreviations of the names of Wentzel. such as "LiouvilleGreen method' and "phase integral method'.30) ax *For details see R. i = eiW(x^h. Classical Limit and WKB Method 14.26) We observe already here that an expansion in rising powers of k.286 CHAPTER 14. The expansion in powers of h2. i. 6. Dingle [70]. z (14.1 The W K B Method T h e a p p r o x i m a t e W K B solutions Without exaggeration one can say that (apart from numerical methods) the method we now describe is the only one that permits us to solve a differential equation of the second order with coefficients which are nontrivial functions of the independent variable over a considerable interval of the variable.2. is an asymptotic expansion. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large. §We follow here partially A.4) and (14.3 14.3. We set.§ y = A(x)eiS^'h where W(x) = S(x) + .5)) and 2 ^ f dx ax + All=0.$ 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.In A(x).[EV(x)}y = 0.27) (14. Kramers and Brillouin. 318. similar to our earlier procedure with A and S real functions. p. As remarked at the beginning. Sec. (14.28) The calculations we performed at the beginning imply the following equations (compare with Eqs. Messiah [195]. or h2/2mo(E — V).
(14.28). Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .33) 3 (S")2 .35) .32) is therefore (S'0)2 + 2H2S'QS[ . (14. + (14.14. (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.3 The WKB Method 287 Equation (14. 3 / 5 . 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.V) and (14.10)).34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .31) \a. (14. Since the expression (14.A = const.e.2mQ(E . .36) Equation (14. Li This expression can be substituted into Eq.35) can be integrated.2rn0(E V) = h2 lS'A 2S'0 (14.\ In S'.32) s" = sz + h2s'( + •••. Eq. (14.29) becomes (note that this is still exact!) (S')2 .30) corresponds to the equation of continuity (i.4 (S>)2 1 S'" 2 S> (14.
± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2).and So = ±h / — — A J k(x (14. Chapter XIII.' For the expansion (14. Eq.33) to make sense as an asymptotic series.41) We now ask: What are the conditions of validity of Eqs. it is reasonable to set (cf.40) rx dx v "7 W)\.39) and (14. (14. and instead is asymptotic which we shall not establish in detail here.288 CHAPTER 14.19).39) where a and /3 are constants. (14. (1. p = h/X) k{x) :Then S'n +.38) h y/2mo(EV)' (14. Dingle [70]. .31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . (14.37) For Eq. y/2mo(V(x) .e.27) we obtain now with Eq. i. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). B. The solution is seen to be periodic provided k(x) is real. E > V(x). Classical Limit and WKB Method has the form of a momentum. we expect that in any case So » ft2Si. (14.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) .E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.41)? We note first that the expansion in powers of h2 does not converge. 1 S e e R. (14. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.
we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A. E > V(x).42) [(^).V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14. (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.36). Eq. Since k(xy 1/52\S'0 (14.14.3/2 s&T + 3 {So Il« we have (cf. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .1 / 2 ]" [§0SS).45a) .
at which E — V{x) changes sign? We note first that in this case near x = 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. h = 1) . *2 A(z) = V / 2m 0 (£ . (14.e. 14. (14. 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).V(x) = V(a) .3.a)V'{a). We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.48) behave in the neighbourhood of such a point called "turning point'. V(x)) (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. Roughly speaking one therefore requires E to be large in both cases.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . i.V{x) ~ Or . the WKB approximation is in general invalid (see below).290 CHAPTER 14. How do the solutions of Eq. In the neighbourhood of E = V(x).46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. and hence (with UIQ = 1/2. around the classical turning point. Classical Limit and WKB Method we can rewrite the inequality (14. (14.
V=E Fig.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq.1. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . (14.e. We assume we have a turning point as indicated in Fig. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.1 there are WKB solutions on either side of the turning point and some distance away from it.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. We shall not derive these conditions here in detail but rather make them plausible in Sec. i. 14. 14.48). which connect one domain with the other. and on the other side.1. 14. The transition is provided by matching relations.1 The regions around the turning point at x = a.3. 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. in the domain E < V the WKB solutions are exponentially increasing or decreasing. 14. 14. As indicated in Fig.14. in the domain E > V the WKB solutions are oscillatory.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. (14. On one side. since this requires different approximations there.3 after stating them first here.
has the same asymptotic behaviour in the domain i « a a s Ay\. (14.50) x S> a. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a .1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14.H— J =^ v/exp + m (14. Note also the extra factor 2 in yi.292 CHAPTER 14. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow.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. (14. as in Fig. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. a interchanged. one could simply replace everywhere /•a / pa • • • b y / Jx J x .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 / —. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. however with the limits of integration x. This means the asymptotic part suffices only in t h e exponentially decreasing case. 14. in order to make the formulas independent of whether the potential is rising or falling. 14.52) For a potential rising from left to right. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. Thus.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a .
One should remember that the matching relations above are those for the dominant terms in the WKB expansion.54) .14.Airy overlap* Fig. whose exact solutions are the dominant WKB approximations. Therefore we consider now this original equation in the domain around x = a. i. 14. (14. the equation becomes approximately y" = (xa)xiy. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. the matching relations have to be altered accordingly.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.e.51) — always with the decreasing exponential on the left — is valid in both directions. the equation V approximately linear around turning point V=E Airy solution WKB .3 The WKB Method 293 In that case Eq. 2 1 X (14.53) Since at x ~ a. If higher order contributions are to be taken into account. 1 ~ (x .3.48). as we saw. is singular at a turning point x = a. (14.46). (14. 14.a).2 The overlap regions around the turning point at x = o. y" + Toy = o.
.57a) and (14. Dingle [70]. formulas 10. In particular we encountered a solution written Ai. (14. i I 772 \3 + > for ia>0. In the following we require the asymptotic behaviour of the solutions Ai(—z). (14. or M.^ 4 I 7T x — a <C 0. Cf. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant.57b) UV2 3 11 by J dx .59). This is given by the following expressions.1 / 4 by \^.294 CHAPTER 14. With the substitution z = (x . p. A. (13. A second solution is written Bi. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. (13. B.56) We have discussed some aspects of solutions of this equation in Chapter 13. > for a.Bi(—z) for \z\ — oo.5960.a)X\/3.291.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0.55) the equation becomes the Airy differential equation d2y dz2 zy. (14. — a ^> 0. Stegun [1].a)3/2 ^ /2 =^ 3/2 o Q. J (X (i4 58)  Replacing in Eqs. Abramowitz and I. p.4. (14. as > may be seen by referring back to Eqs. R. 448. we have r x * f x ' /2(x a)i/2d*=^(x .57b) To the same degree of approximation as the equation y" = (x — a)xiy. (14.60).
In this way different asymptotic branches of one and the same solution are matched across a turning point. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = . In the direction to the left as indicated in Fig.52). 14.3 The anharmonic potential well.l v/2m0(EV(x))dx 7r = (2n + l). These solutions. As we approach a limit of this interval in going away from the turning point.59a) .14. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig.(zf2 by J we obtain the matching relationsX (14. e.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution. In this sense we have now verified these relations.l. We summarize what we have achieved.3 The WKB Method and 295 •. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). 14. (14. The latter domain is a small region where the solutions from either direction overlap. and hence must be proportional (in view of the uniqueness of the solution there). n = 0. are only valid in a small interval around the turning point at x = a as indicated in Fig.51) and (14. 14.g. Example 14.1 to the quartic oscillator. we enter the domain of validity of a WKB solution. by Ai(z)..2. As an illustration of the use of the WKB solutions we apply these in Example 14. however.2.
The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. i. According to Eq. extends only over a length of very few wavelengths and is therefore valid for n small. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator.l. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator.296 CHAPTER 14. j/i. ^ A = (2n + l ) . (14. Here we are interested in the discrete states (the only ones here). . yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. 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. We argued earlier that the WKB method is suitable in the case of large values of E. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.y o dx TT T provided 6 _ 4 /. 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.60) Using the matching relation (14.3.59b) Solution: We consider an anharmonic potential well as depicted in Fig. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . Classical Limit and WKB Method where x = a. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. Sometimes this distinction becomes imprecise. n = 0 . The condition for this is ya{x)=yb{x) for xe(a. as in the case of the harmonic oscillator itself. ..61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). 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 ( . 2 n = 0.(x) = c ' v Acos I / — I for a < x < b. 1 . (14. regions 1 and 3 in Fig.e. .^ .51). 2 .2.3. One can also express this by saying that the linear approximation of the potential around the turning points must extend over a length of several wavelengths and is therefore valid for large values of n. (14. (14. Hence we search for solutions which are exponentially decreasing in the far regions. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.44) this implies large values of n.b). 14. . (14.62) The condition for this to be exactly satisfied is — as we show now — that the Bohr—Sommerfeld— Wilson quantization condition holds and c' = ( — l ) n c . Then around these points the potential would be well approximated by a linear potential (as we saw above). b are the two turning points. /n = s u ^ TJcos^ya TJ+cos^a T J . Let E = V(x) at x = a and x = b. . 14..
and knows of no published form or script. equivalently. Bohr. In this socalled "old quantum theortf (i.64) (14. pdq = n + 1 — . 2 . . ri2 I Jqi :pdq n + 1 — . .4 BohrSommerfeldWilson Condition 297 14. Dingle. with n — 0.14.4 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. . Sommerfeld and W.x number of turning points TTH.1. Wilson by supplementing classical mechanics by Planck's discretization. The corrected form* is. This relation is sometimes wrong. before Heisenberg's discovery of the canonical algebra and the formulation of the Schrodinger equation) this condition was always given as §pdq = (n + l)h.1. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. 14. .63a) or. n = 0. 2.e. (14. B. A. (c) V Fig.63b) *The author learned this in lectures (1956) of R.x number of turning points h. (14. .4 Three cases with different pairs of zeros.
where n —.298 CHAPTER 14. where /(g) is positive. together with q — — oo.4(b). as illustrated in Fig.e. 14. In this > case the wave function for g > go is . the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. i. We have the case of trigonometric solutions. i. Since sine and cosine have zeros spaced at intervals of TT.66) J<1\ Thus in this case the socalled old quantum theory is correct. Jqi ^ fpdq J =2 pdq = (n + l)h. i.4(a). (14. this is an exponential zero. / • n an integer > 0. 0(g) ex . In quantum mechanics we have dq2 + Ki/j(q) = 0.65) We use the knowledge of these solutions to find the eigenvalues of the equation.e. the case of tp(q) — 0 at q = gi. if /(g) is negative.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. i. this means dqyf—f = (n + l)7r. this is a trigonometrical zero.e. Case (a): The case of two trigonometrical zeros. Before we continue we recall from Eq. where /(g) is negative. There are three cases. 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. (14.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0.67) Case (b): Next we consider. (14. 14. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h.
Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT.e.2: The harmonic oscillator Use the relation (14. (14.63b). As illustrations we consider Examples 14. the sines must be proportional. Hence we have pdq= i n +  . Therefore their arguments differ only by (n + l)ir.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). Solution: As in the case of the quartic potential. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. i. (14. i. Thus in this case the socalled old quantum theory is wrong.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . and • • for the wave function to be exponentially decreasing to the right of q'0. ri\ / J do y/f(q)dq (n + 1) 1" (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.69) pdq = (n + l ) " 4 h.70) We can therefore summarize the results in the form of Eq. 14. (14.63a) or (14. Since both cases refer to the same region qo < q < q'0.68) (14.2 and 14.e. this is the case of two turning points.4(c).3. For the wave function to be exponentially decreasing to the left of qo and using Eq. E x a m p l e 14.63b) to obtain the eigenenergies of the quantized harmonic oscillator. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r.14. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . or Pdq (n + l ) " 2 h.
2 3 J x (n + 3 / 4 ) 2 / 3 .5 A particle above a flat Earth.807 m s . Calculate its quantized energy. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. :: V=m0gq Fig. where g is the acceleration due to gravity. whereas a turning point does. permit this although with rapidly diminishing probability. Thus there is one turning point at E = mogq. Example 14. one obtains E ~ 2. 14.3: 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. g = 9. so that E = hu{n + 1/2). 4 / x CHAPTER 14.2 . as illustrated in Fig. 14.28 x 1 0 .63 x 10~ 3 4 J s .300 With potential V(q) = 2ix2mov2q2. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . . in principle.V(q))dq The result is E/u. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo.5. and h = 6. 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+ • (14.71) For m 0 = l x 1 0 _ 3 k g .
72) \\2dxp=Qdx^{Eh Hence.14.4: 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 £ = . 1 . 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). with p = modx/dt. Chapter 16) one obtains the quantization relation (for Mo = const.l.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 ( — = . ra 2/ = 0. and u> is the oscillator frequency in either well.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 .5 Further Examples 301 14. In the case of screened Coulomb potentials (cf. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .73) Jx! rp2a dx I^<E 2 v) = ^ n — = .( 1 — = — = it — = h\ n H — 2TT OJ 2K J p dE V 2 In the case of the harmonic oscillator we have hi n E (14.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB. E x a m p l e 14. (14.2. (14.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.5 Further Examples E x a m p l e 14.f V ft h JXl h 2TT (14. d dE V)={n+)n.7.75) which verifies the formula immediately. 9 .= — <*— = h(n+\. Solution: The proof is contained in Example 18.
302 and T CHAPTER 14. which determine the imaginary part of the energy. (14. The q term in this equation can be removed by transforming the equation to a cubic in y.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf.y2)(y . Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.a . Eq. where .6 The cubic potential. 14. Example 14. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.~ \ (14. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. Fig. i. q = y+ a .y3) = 0. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). (11.6. where 1 2 • 1 2 y = Q.e. 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 .<?2i<?3 for E > 0 given by E — V = 0.79) .77) Determine (a) the turning points.133)). 14. (14.78) (a) We determine first the three turning points at qi = <?i.
80 a n d 361.9 2 [ .g i < 1. B y r d a n d M . F .g i ) ( g 2 .0^6O°.1 / 2 . F r i e d m a n [40]. g 2 ~y{3}>0. 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. .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . D w i g h t [81].k 2 2tan# \/3 + tan! .e.a 2 [ l .9 3 — = = 2 " 3 " [ . 6 = 0 i m p l i e s E m a x of E q .2/3 i n t o q = y + a 2 / 3 . (14.< 1. we h a v e qx = . (l_fc' 2 a + fe'4)l/4. B .80) I n s e r t i n g t h e r o o t s yi. we set 0 = 60° — e.= s i n t 3 v 3 (14. V cose' + sin 8 (14.T h o m s o n [196]. V3 + t a n 0 2 .93). _ \ / 3 — tan(9 2 V92 . 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 = 1.14. f o r m u l a s 236. p . 93.78)) c o s 30 •• 4P3 ( ' \ aeb2} . 92 = a2[l .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 . 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 . 92. 0<6O°. 14. q3 = a2[\ . 78: sin 120° = v ^ A c o s 120° = .93 2a2 [ . tan60°=%/3.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . 37. § T o h e l p we cite H. ' P .6. 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 . i.' : ± 6 0 ° w h e n E = 0.2 c o s 0 ] .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.120°)] = . (14.04. 91 < 93 < 9 < 92(14.81) m a 91 .q)(q .{ V 3 e } > 0. D . 212. gi ~ y {V3e} < 0.08. ui=sn 1 \l— — = l.3/2. p .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. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . H e r e cos(<9 ± 120°) = . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . (14.\tan0^ v/3. M i l n e . (14. p .c o s ( 0 + 12O°) + c o s ( 0 .9 l ) f f / du s n 2 u en2 d n 2 u . M .cos 0 + cos(6> + 120°)] = a2 92 . 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.c o s O + cos(6» .84) 92 .85) T h e angle 9 = 60° c o r r e s p o n d s t o 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 .gi = 1.79) a n d t h e n s e t t i n g m o = 1. (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 c o s ( 0 .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. p . e > 0. a n d t h e a n g l e 0 = 0° t o E = £ m a x . q3 ~  .ui=K(k).2 c o s ( 0 + 1 2 0 ° ) ] .
E(u = 0) = 0.304 CHAPTER 14.00. E(u) the incomplete elliptic integral of the second kind. F. Friedman [40]. 0 (14. G(fc) =86aS (ITS£ li[fe.) . 72.^ + ^ ) (l + fc'2)2 . V?2 . D. 91 < q < q3 < 92 bJc.91 . Next we evaluate the required integral across the well from 91 to 53 at energy E.fc'2 + fe'4)"1/4. Thus Iwen=lgF(ip. V 93 . (24. V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . (14.89) This expression will again be obtained later with the help of configurations called "bounces" (cf. (14. 2 2 . A:') = # ( * ' ) • (14. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. The WKB Method elliptic integral of the first kind. p. the integrand of the integral across the well is effectively the inverse of this momentum.86) 15 tan. g= 2 . formula 233. as in (14. E(u = K(k)) = E(k). 1+72 = and J o s 0 + ^ = (1 .32)). Whereas the integrand of the above barrier integral is effectively a momentum.87) Inserting G(fc) together with the expressions for the prefactors into (14. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. Byrd and M.k). with K(k = 1) = oo./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. (14.S .~ y ^ ) a< [ COS 5 + .qz) (14.?)(?3 .91) a2[cos0^] y/3tanfl ? L . We can evaluate this integral again with the use of Tables of Integrals.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — .85).84) we obtain Carrier = 2a 2 sin6> 2 b( .^ V3 / V V3 y a / c o s e J.= —^ = k a2[cos0+^] v ^ + tanfl V=". . E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .2(fc2 . Eq.92) ''P.sm t %/3 £\.91 F(TT/2. sinV = q3 qi ~ = 1. .
95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . so that even in the case of the ground state the zero point energy will contribute to the prefactor. pp. Thus one obtains from (14.63b). (14. S. (14. and hence at E = 0. Ryzhik [122]. M.96) Expanding the coefficients of the elliptic integrals in Eq.88) in rising powers of fc' . formulas 8.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.3 and 8.114. 906. .l. 2a 5 b K(k) + 2 E(k).94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq.93) where w is the harmonic oscillator frequency in the well. we obtain 15 . Gradshteyn and I. .97) 4 32 4 32 Here we insert the corresponding expansions of the complete elliptic integrals'^ (note the argument of K and E is fc): K(k) E(k) fc'2 = = < $ ) fc'2 1+ f l n\k'l f H + 1 4 5 3fc' 1T Hfc^J ~2_ 1^ \k'J 12 (14. (14. to represent a physical decay rate. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. we obtain > .! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .4 2n + 1 fc' = 4 — ^ — . (14.89) can be expanded to exhibit a simple pole at z = 0 allowing evaluation with Cauchy's residue theorem: 2 dz 2Ez2 a2b2 %2dz Ez2 2nib\ o?E o?b2 2KE ab tt I.113. 905. 32 (14. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. (14.2 IE 4 fc' ln 8 (" «V iU^fc' 4 .85) (14.** However.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. With q = 1/z the integral (14.80). we obtain E a6b2k'4(l + k'2). The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. a°b n = 0. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). With algebra — which it is impossible to reproduce here in detail — one can derive expansions in ascending powers of fc' (which is small) of all relevant quantities..3.2.14.
7 . . V (14.J + 0(VoJb).> tf = = 2/wfiii = ±tv8a°b——P.fc'2 + fc'4)1/4 * ± t ^ ..306 CHAPTER 14. this has not yet been done. (18. also Eq.100) 'barrier = ~0°b In [ ^ — j .101) (14.2 7 b a r r i e r ] Z Z7T where (cf. It follows that (for one complete orbit back to the original turning point) .„. (0) i hw En = E\> ...( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) ./^~^l^ab ^aub . 15 a . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . (14. _ exp[2/barrier] >.j .ba / 25a5b\n+2 ±l_l eT5 a f >. n ~ V27T . _ T T • (14.e. . 8 Bfc . 7 ~ —. (20.178)) call the Furry factor set equal to one. so that we obtain 4 . We observe that with \j2je as 1.102) . 2n + 1 — / 26a5b \ .— (14. i. the result agrees with the ground state path integral result (24.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.e.e x p [ .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. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.Wigner formula (10. i.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .96) for the imaginary part of the energy as in the Breit.. The WKB Method For k' —• 0 the first and the third terms dominate.73).
C. Gottfried. T. See particularly E. after the realization that quarks as their constituents might not exist as free particles and. the Coulomb potential. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. D. [231].* 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. Kinoshita. There are numerous. which permits only bound states. B. 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. the classification of the various states of nucleons and mesons and other particles. along with inclusion of angular momentum and spin effects and their interactions. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e.M. indications from the nonabelian generalization of quantized Maxwell theory. L.g. Thacker. that this is indeed the case. Quigg and J. at least indirect. Rosner [230]. Yan [82]. describing free fall under gravity. led to a spectrum which contains only scattering states. This potential became widely popular in the spectroscopy of elementary particles. [233] and H. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. Eichten. K.1 Introductory Remarks The particular linear potential we considered in Chapter 13. The study of this potential. 307 . C. L.e. Lane and T. Quigg and J. in fact. [232].Chapter 15 Power Potentials 15. i. Rosner [269]. now known as quantum chromodynamics. K.
we return to the Schrodinger equation in three dimensions. These particles therefore cannot be separated by any finite amount of energy. (15. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15.1) i.2) tt(r) = Ylm{9.2 The force The Power Potential F = . We leave consideration of the logarithmic potential to some remarks and Example 15. f dr{u(r)}'2 1.6) = (I + l)u(r)/r = R ~ rl. R(r) = ^ . For a central potential V(r) we write (15. and u'(0) > [u(r)/r]0.308 CHAPTER 15. </>)R(r). mi + m2 and r is the relative coordinate. Power Potentials general power potentials.£ ] * ( r ) = 0. ~ (I + l)rl (15. (15.1.4) Instead of considering the linear potential. In order to have a clear starting point. A > 0.e. m\m<i M= : . and the normalization [ dr\y(r)\2 = 1.V 7 = const.u ~ r i + 1 (/ + l)R(r). This force maintains this value irrespective of how far apart the particles are separated.e. (15.4>)ru{r) = Ylm(9. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). 15. h2 2n V 2 * ( r ) + [V(r) . i.m2. where \x is the reduced mass of the two particles of masses mi. v > 1.r acting between two particles is constant and directed towards the origin (of the relative coordinate).3) with the boundary conditions''' u(0) = 0.5) u'(0) u[r) r=0 = R(0). and so u'(r) .
."MA.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.63b) for the onedimensional Schrodinger equation..15.1.10) Here rc is the turning point and n — 1. odd integer ^=(nll)vh.2.7) k(x) = h/y/2m0(EV(x)).9) Threedimensional case: We transcribe the above considerations of the onedimensional case into that of the threedimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!). (15. N = 0. (15. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality. or f l H " ^ ) ! .8b) This condition implies that we have only one pair of turning points. .4) in the threedimensional case.63b): 2 and V(0) = 0 (15. In order to to obtain the . We now want to find a quantization condition for the threedimensional problem.V(r)) = (2nl) +1 N.8a) ^ JO = (2Ar + l ) £ .2. We therefore proceed as follows. (15. (14.8a) we can write the quantization condition (14.. is to be interpreted as the principal quantum number in three dimensions. d5.9) into the threedimensional case in the following form: / Jo C dry/2n(E . so that with Eq. and the turning points are at a — — b := — xc. . (15. Then k(x) = k(x).. This means that we transcribe Eq. (15. (15. We also assume that V'(x) > 0. 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. x > 0. which has been reduced to an effective onedimensional problem in the polar coordinate r. 3 .63a).
1W or Xv \E I It follows that (I) / ^""^(l*)1/2. c (nj\irh. s.14) Prom Eqs.'(1*)1/2= J1F\lt)y1dt. _ —vr" *dr. (1516) (15.y) defined by (xl)\(yl)\ y ^ (x + y .\rvc = 0 .6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E . Now. Power Potentials eigenvalues E —• En for the radial potential (15. (15.12) and (15. (15.310 CHAPTER 15. E A fv.13) dt . (15. .15) I ' = ^ 1 (( 1) B{x.17) The integral on the right is the integral representation of the beta function B(x.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/.13) we obtain _ X_ „ _ so that A£_ (15.11) rc = (15. dr = .
* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. 15. ^ '"Compare with Eq. 15. (15.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . One can convince oneself now. i r ( i ) = r ( l + i ) = ( i ) ! .1 Approach to square well with v —> oo. and n — — ]7r/i. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.15.1)!.55). $T(z) = {z.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. T(z + 1) = z\.1. for instance simply graphically as indicated in Fig.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. (12. . The case v = 4 is that of the pure anharmonic oscillator also discussed by M. Weinstein [281].
The potential V(x) = X\x\ (15.61) h2 = I. we obtain E„.2) implies A s fj.1 + \)hx> = {An • l)£fiw.> r ( 2 ) A V 2 in \)*\W \it^T& (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. 15. discussion at the beginning of this section). In the case v = 1.312 CHAPTER 15.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones. the harmonic oscillator. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 . = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. Physically it does not make sense for the probability amplitude to have a discontinuity there.oj2/2. we obtain En = (n . 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. (6. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. Power Potentials In the case of v = 2.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. (15.2.41) for the .g — —2.mo = \.oj = ^/2X/JX. eigenvalues (JV + h)tkj — (2n .24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). For the following we set in Eq. the case of the linear potential.21) (4nl). > and with Eq. (13.25) "Comparison with Eq.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. x > 0. (6.
i.x).59) applies. i. 15.x)\x=0 = 0. i.28) In particular at x — 0.15. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. Eq.29) . Ai(s) 1 2v^(s)V4 exp x —> +oo.27) i.e. (14.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig.e.e.e.2 Behaviour of an odd wave function at the origin. and for s —> oo the trigonometric behaviour of Eq. this wave function has the required exponentially decreasing behaviour at infinity.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .56) and (14. (15. dz2 According to Eqs.e. (13. E > V or s > 0.60) applies.26) For z = s := E — x —> —oo. \(s?» (15. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . In the domain —a < x < + a . s = E. (15. i. Ai(s) "v^ G ' cos s 3/2 _ t (15. (13.
It is interesting to note that in the dominant approximation both methods agree. *This Table is reprinted from C.03914 11.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large.30) ! ^ / 2 . of course. The expression (15. with permission from Elsevier.314 CHAPTER 15. (2/3)E3/2 or 3 / 1 2/3 E — En — . = im — (7r/4). (15.82878 1 2 3 4 5 6 7 8 9 10 2.93602 12.33811 4.. [231].04017 11..3. (15. the eigenvalues En are determined by the zeros of the Airy function. i.. Table 15. i. Rosner.1 demonstrates the quality of the WKB approximation by comparison with the exact values.94249 9. p.e. (15.2. 201.^ = !(2nl).93528 12.e.32025 4. Table 2.e. Co. copyright of NorthHolland Publ. n = l. Quantum Mechanics with Applications to Quarkonium.82814 i. to obtain the zeros of the Airy function numerically. Power Potentials Table 15.29) implies that cos (  ^ .78671 7.28) is really only valid for s large and s — oo.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .08181 5. but for large > values of E Eq.52056 6.02265 10. L.51716 6.En from Ai(£„) = 0 2. It is possible.94413 9.= ) = «.31) (an odd function has an odd number of zeros!).00767 11..00852 11. . so that n — 1/4 ~ n.02137 10.08795 5.78445 7. Quigg and J.
e. The decay widths T. In general these quarkantiquark pairs. ee.** r ( # ^ e e ) o c *(0) 2 . *Note that this makes N a W K B normalization constant. D. Weisskopf [239]. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. of course. We shall encounter WKB normalization constants at numerous points in later chapters. (14. have only a finite lifetime and decay into other particles such as electronpositron pairs. cos2()^— / 1 fr° 2 > w .34a) **The most frequently quoted reference for this result is R. e.g. the charmanticharm meson ^ . can be shown to be proportional to the modulussquared of the particle wave function at the origin.3 The ThreeDimensional Wave Function 315 15.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. which are inversely proportional to the lifetimes. The resulting masses agree in most cases very well with those extracted from experimental observations. i.• •) by a mean value. See also J. i. e.g.15.50)) 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. . in Chapters 24 and 26.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. We saw previously (see for instance Eq. r <53) 1 3  dr cos 2 () (15. F.e. In order to be comparable with experimental data these investigations. P.e. We indicate briefly here — without entering into further details — how these quantities can be calculated. van Royen and V. Jackson [140].
40) VWY .e.37) (15.38) N : sin V ^ Jr dr' 7T (15.rcu w X{r)dr = Kh . (15. From Eq.\ \ * n = l.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 . (15. (15.. i.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.316 In accordance with t h e relation CHAPTER 15.1 )n . Power Potentials this averaging implies a numerical factor like 1/2.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .39) in / . & (15. dE a—— / dn J0 so t h a t with Eq.2) and (15.l N n .35) N2 .36) \i dEn irh2 dn According to Eqs..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. so t h a t approximately l N2 / Jo k(r)dr .34b) VMEvy Variation or differentiation of Eq. (15.2. A = (15.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. (15..
41) I*(0)2=_L^)^.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. Kaushal and H.23) for the energy levels of the linear potential.38) implies 317 (15. E. G. J. we have ^'(0) = ( . (15.15. Itt^pocn2*"1^"^. S.l ) n . . K. (15.1 ^ ( 2 ^ „ ) 1 / 4 . for various potentials: (a) For a confinement potential of power v. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. MiillerKirsten [147].3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. 3 . and hence with Eq. the expression (15. MiillerKirsten [28]. W. W.44) (15. S. Bose and H.37) we obtain: 1 (2M^)1/2 47r ft. J. MiillerKirsten [29].* Using the result (15. An enormous literature exists on the subject in view of its relevance to the spectroscopy of mesons and baryons made up of quark constituents and the necessity of checks with results derived from experiments. Hence Eq. W. The case of arbitrary positive power potentials as above has been considered by R. J. . . (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. for instance. K.2. Bose.g. Hite and H. 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). . 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. § S.
S.— oo < z < oo. K. Eq = i q .j . .133). 7 With the substitutions r = exp(z — c).. E' = E + gln(r0).318 (c) and for the Coulomb potential^ CHAPTER 15. . (11." 1 11 L .I l n f ^ p .L2 h4 = 4/3exp[(2a — /3)/f3]. D. *=^f.g ^£+(<*Pl»riy = 0. J. M.e. and there Eq. Solution: Details can be found in the literature.46) and the classical Kepler period of Eq.1. = 1(1 + 1). (15. in agreement with Eq.46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14. Lifshitz [157]. MiillerKirsten [29]. n = 0. ip = exp[(z — c)/2]. q = 2n + l. Power Potentials *(0) 2 oc . = (I + 1/2) 2 .2.^ +•••.5. {S = ^ > 0. Landau and E. 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: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). rewrite the equation as ^ z + [ . U(z) =  dz zf3e2a/?e2*.^ ( 3 ^ + l)55p^«(3^l) + ..c = —a//3. Bose and H. Example 15..75). (14. (15.L 2 + U{z)]<f>. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . i. W.. .
r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. This effect sounds unphysical. it is.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . It is clear that for ro — oo the > potential becomes of Coulomb type. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. Historically this potential arose with the realization 319 pVr . In reality. The exponential insures a rapid falloff of the potential at large distances and for this reason the parameter ro can be looked at as a measure of the range of the potential.Chapter 16 Screened Coulomb Potentials 16. that no matter how far the particle is scattered away from the source of the Coulomb potential. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. This implies. it will always notice the latter's presence.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. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. and in fact. A very similar potential is the Yukawa potential V{r) = g2 — . so that like charges are repelled and unlike charges are attracted by each other.
e. in fact the parameter [i represents the mass of such a spinless meson. . Yukawa potentials and superpositions of such potentials play an important role in nuclear physics.3) where for the real potentials we consider here.ii k .1) We encountered Green's functions earlier. Thus in the following we consider a generalized Yukawa potential which can be expanded as a power series in r and is written oo V{r)= }Ml+1(r)\ (16. (16.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). the mathematical expression with an exponential is not so easy to handle analytically. 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 .* 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. (16. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons.320 CHAPTER 16. and that. all coefficients Mj are real and independent of the energy E = k2. however. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. Since. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.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. i. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. from the relation Pi + P 2 + M2c2 = 0. one frequently resorts in calculations to the expansion of the potential in rising powers of r. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq.2 ) or (p = hk) 1 /" . The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. Heisenberg [133].
Kinoshita [21]. Cheng and T.r). S S.k.e. . S. i. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. however. A. (16. 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.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. Bethe and T.e.§ 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. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. They are usually written I = an(E).mo being the reduced mass of the system. Mandelstam [185]. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. 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. See also the other references below. i. to be of the order of 1/k. if>oo. Wu [47]. J. £+**PVV t/.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.e.r)=0. T. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. Basically this argument amounts to an argument similar to that used in the case of the Coulomb potential where the integer n arose from the requirement that the wave function be normalizable. f H. (16. Omnes and M.4) where E = k2 and h = c = 1 = 2m.1 in the above. Frautschi [97] and E. l+n+l = ^P.5) where K — ik and A n is an expansion in descending powers of K. (16.1.{l. . C. Potentials expandable as in Eq. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. Lovelace and D. i.16. (16. only for the cases of n = 0.3) in the radial Schrodinger equation for the partial wave ip(l. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. Squires [258]. Masson [180]. bound state eigenenergies and the 5matrix.Q. "These were first considered by C.e. . .k.3) have been considered by various authors. Froissart [223]. 2 . physical) values of angular momentum as functions of energy E.
Miiller and K. Longoni and T. Muller [204]. tf H . as may be seen from Eqs. J. (16. A. J. Squires [258]. Muller [201] and [202].t Considering such superpositions of Yukawa potentials with /•oo / a(/j. .19).ndp. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. W.15) and (16. In the following we follow mainly the last two of these references. W. starting with the power r _ 1 of the Coulomb potential. Proceeding now as in the case of the Coulomb potential we change the variable of Eq.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. Regge [31] and E. Section 11.)/j.z) Then x is a solution of the equation VaX= (16. J. J. Under these conditions the S'matrix is meromorphic (i. o rV(r) I dpp\V{peie)\ regular at r = 0. 16. in the present case at fc2 = 0 or k2 — Mi = 0.z) = ez2/2zl+1X(l. Bottino. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. *See in particular A. H. M. (16. 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. < oo for all 0 < vr/2. < const. Screened Coulomb Potentials as a function of the energy. Schilcher [203].9) the cut starts at E = k2 = 0.k. W.e.k.322 CHAPTER 16. one can assume an expansion of the potential V(r) in ascending powers of r. for all n. 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.** The energy is later obtained by reversion of the resulting series. 'In the case of the Coulomb potential (cf.4) to z = —IKv and set TP(l.
a) —b — 2a. The function <&(a.* Recurrence relations for coefficients of perturbation expansions for Yukawa. and all Sm(a. j)Smi(a.j) = (a + j l. a + 1) = a — b+ 1.2 Regge Trajectories where d2 Va = z^ and (cf.11b) ^ The associated boundary conditions are: So(a. Sharma and H.i 0) = 0.awhere (a.9) we obtain m l)$(a .a + j)Smi(a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. anharmonic and cosine potentials have been derived in L. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).b.0) = 1. J. (16.5)) a = + 323 .b. By a repeated application of the recurrence relation (16. so that to leading order we have VaX{0) = 0.z) = Ha). {0) X = Ha. W.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. all other So(a.a+ l)$(a + 1) + (a.7) is seen to be of order 1/K.a — l) — a — l.a)${a) + {a.j)$(a + j).j) may be computed from a recurrence relation which follows from the coefficients (16. (a.a + j)Smi(a. .j) for \j\ > m are zero. (16.j l) + (a + j. Eq. which means that the hypergeometric series breaks off after a finite number of terms.5) is equivalent to a = —n.9) (16. (16. MiillerKirsten [249].16. (16. We also observe that the ansatz (16.j +(a + j + l. where $(o.11a) The coefficients Sm(a. K. (a.1).10) zm$(a)= ^ j=m Sm(a. as in the case of the Coulomb problem.j) (16.10): Sm(a.a + + l).
' — . i.a + l ] 2 [ .13) one obtains the quantity An(K) and hence with Eq. 1 . the latter can be written .7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.a\4.aj 2 (2K)A _ [a. 0 < \j\ < i.12a) where [a.<&(a + n) on the right hand side of Eq.a + j]i+l = MiSi(a.324 CHAPTER 16.o]i = M 0 . and hence Va$(a + n) = ra$(a + n).13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. .a]l^a) + ^2(2K)i+1 Y. Any term /j. [a. when n = 0.[a>a + J]i+i*(a + 3). (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. of course. [a.7). Va+n = Va . [ a . This follows from the fact that D a $(a) = 0. + (16. a .aj 2 \ [a.j). We first cite the final result .l. 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). This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. oo _ . (16.1] 2 . #(a + ra) fjL$(a + n) V.n. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq.5) the Regge trajectories I = ln(K). Evaluating the first few terms of the expansion (16.[ a + l.An{K). (16. i R »] = 2K[a. (16. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . (16. for the coefficients M^> of the expansion Details can be found in the references cited above.e.
15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . of course.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n .5). is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. Burke and C. Obviously the expansion — which.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n .16. Ahmadzadeh.^ p [ 3 M 4 ( n .1. G. (16. 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. Inserting this expansion into Eq. However.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . i. .^ 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. With numerical methods one can achieve more. Thus the expected appearance is that shown in Fig. we obtain the expansion for the Regge poles. 16. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.e.2. including an exploration of the domain of small energies E.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .1) + 3M3M02 + M22n(n + 1) (16. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.1 by evaluating the first two terms. Tate [6]. .( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n . (16.§ This is an important observation which indicates the equivalence of the methods. P.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8).
just as a decay width T represents the width of the resonance in energy. so the imaginary part ai of I — an represents its width in angular momentum.—" + In = — .1). 16. Similarly the conjugate variable to angular momentum is angle. 16.a)So(a.^ . Frautschi [97]. The conjugate variable to energy is time.326 CHAPTER 16. ±1) = 0. through which the particle orbits during the course of the resonance.0). 0). and the angle A#. p. a] 3 = M 2 S 2 (a.12b) we obtain [a. satisfies the relation ajAO ~ h.11b) we obtain Si(a. [a. ai is small and A6 is large." Re I 1 0 1 2 Fig. Hence Si (o.0) = (a. (16. Solution: We evaluate the first three terms of expansion (16. 114.2 a = In See S. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. .a) = b . a] 2 = M i Si (a.1 Typical Regge trajectory for a strongly attractive Yukawa potential. For a resonance with a long lifetime.13). . where the imaginary part of I is very small.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. However. 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. Example 16. one can expect a resonance with the lifetime determined by this imaginary part. C. (16. From Eq. 0) and Prom Eq. (a. For a bound state a j = 0 and the orbit becomes permanent. In fact.O) = Other terms vanish since So (a.
0)+ 0 + 0 = (a1). 14. Sec.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .b)(a .6 + l)a + A. (16.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . 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 . (n+l)U+^pJ .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .0) + 0 + 0 = ( a . 1) Hence 52(a. 1). 0) + a S i ( a . 1) (a . A" An An A" = = = (a.11b). We obtain the quantities S i ( a .6 + 1 ) . .0) Si(a.1) Si(a.1 ) + (6 . (a. One sets z = 1/r so that.n ( n + l Inserting these expressions into expansion (16.a)5o(a.00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.0) = (a . a)Si (a. a)Si (a. i ) again from Eq. we have to evaluate (cf.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 .2 Regge Trajectories 327 Analogously we evaluate S2(a.0) + 0 + 0 = ( 6 . (a.1. Thus (with terms which are 0): Si(a.0) (a . . Solution: Ignoring higher order terms in r.a + l)S0(a. . using the Cauchy formula (6.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.16.2o)Si(a.1) + (a . 0) + (a + 1.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity".al)S0(a.13) we obtain 0 (Mo .6)Si(a.1 ) + (a.2 a ) . a)Si (a.
l. The turning points in this case are at r = 0 and r = oo. 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 .15). we can use Eq. Example 16.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ .2.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . Here we do not perform this procedure. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. Singh [252].e. *J.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. n = 0. by expanding the right hand side in rising powers of v and evaluating the individual integrals. tSee Chapter 11 and V.328 Thus here dr K' CHAPTER 16. . Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. Solution: For details of the solution we refer to papers of Boukema. (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. ±2HK in agreement with our expressions above.t First. i. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . I.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Screened Coulomb Potentials Mo . however. Boukema [34] and [35]. (16..* 16.3: 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.5) together with the expansion (16.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.
0[(KZ + Mx)'6} (16..l. the behaviour of the scattering phase in the domain of high energies. n = 0.18) of the Smatrix now to explore further aspects.16.* 16.19) +2M 3 M 0 (3n 2 + 3 n . This has been done§ and one obtains . 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 .e. J. J.14). § See H.17) and is therefore real for real values of the potential coefficients Mj. Now reversing the expansion (16. Miiller [204].2. Paralleling the case of the Coulomb potential in Chapter 11.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.17).4 The Energy Expansion It may have been noticed that in the above considerations we could have combined the constant Mi with the energy into a combination K2 + M\.l (two additional terms are given in the literature). W. such as. Miiller [211]. W.g.. .l)n(n + l)(n + 2) + 2M2M02 . 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. the following *See e. H. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\. ..19) we obtain the energy. One can use the explicit expression (16. for instance. We observe also that the S'matrix is unitary as a consequence of the result (16. which are precisely the expressions yielding the Regge trajectories of above. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. i.
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) ^ — .^ 16. Sommerfeld [256]. in the following expression with an "See in particular G.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . the transformation was introduced in the form given below by G. N. A.330 CHAPTER 16. Frautschi [97].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 .1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . Screened Coulomb Potentials expansion. p. Warburton [279].1) + 2M2M. J. Iafrate and L. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. MiiilerKirsten and N.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.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. p.11) + 2M22M03(9n2 + 9n .l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . i. B. Large coupling expansions derived in H. 282. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). .10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n .l)n(n + l)(n + 2) . J. Zauderer [286]. 107. Watson [280] in 1918 and later resurrected by A.< 3 M 4 M 0 ( n . VahediFaridi [212] contain as special case the expansions of the first pair of authors. E. Mendelsohn [135] and E. 'According to S. C.* We recall from Chapter 11 the definition of the scattering amplitude F(9.2)(n .l)n 2 (n + l) 2 (n + 2) 1 + • • • (16.e. W.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n .
(11.187)) (16.iJ 11 —[S(l.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. ~:' sin nl (16.k) — e 2i5[(k) ~" p—iirl (16..k)~l] = ~ k" PiSiW sin 5i(k) and S(l.''' The idea is now to consider each term of the expansion (16. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig. z=o in which (cf. c = 1. For this purpose we consider the following integral taken along the contour C shown in Fig. Actually 5j lrr/2 as in Eqs.24) Si(k) being the phase shift. (cos 9).k)\2.26.1.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0..k) = ^T(2l + !)/(*. \x\ <C l .187).k) = 2zfc L " v "'' v . The scattering amplitude possesses the partial wave expansion F(0.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.kmcose).5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1..184) to (11. 16. (11. n = 0.23) f(l.2.25) Setting Note the minus sign in the argument of the Legendre function.k) 0ikr (16. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).fe)P ..22) = \F(9. 16. I = n + x.2 The integration contour C. Eq. .
k) Zi Jc TT{L .fc)P„(cos0). For the momenta indicated we set (with metric +. 16. Consider the reaction of a particle a{p) with fourmomentum p^ colliding with a particle b(q) having fourmomentum q^ and producing a particle c(p') and a particle d(q'). in which case c = a and d = b.n) oo {l)1 Pt{cosO) v ' (16. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. 16.3 Reaction channels and their respective Mandelstam variables. be elastic scattering of a off b. 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. t = (qq')2. a + c —> b + d.3.27) a(p) © b(q) Fig. for instance. we obtain m® = ± I dl^±^f(l.—. u={pq')\ (16. as indicated in Fig. a + d — • c + b.28) . In order to see the relevance of Regge trajectories in a reaction of particles. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. n=0 which is the usual partial wave expansion. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. d(q') (16.26) P((+COS0) = 2 ( 2 n + l)/(n.3. 16.—) s = (p + q)2. Thus the direct reaction may.—.332 CHAPTER 16.
.t). the variables t and u would describe momentum transfers in the crossed channels.2 ^ ( 1 + cos6 s ).^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. (16.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). e. N.16. (16. F. the quantities carrying these quantum numbers are the Regge trajectories.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. L. M. the dynamics is described by the exchange of quantum numbers.u) (if any of the external particles has nonzero spin.t. S. the momentum k there would correspond to qs) s = ~4(ql + ml).5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.et) (16.31) i l Taking all particles to have the same mass mo. except for poles and cuts which characterize the reactions in the three channels. 16. t. In a Regge theory on the other hand. t = 2q£(i . § See G.g. if s describes the square of the total energy in the s channel.u are known as Mandelstam variables.g.c o s 0 s ) .u are not independent. which is an analytic function of the variables s. u. T See e. one then has A(s. s = 2qf(lcos0t)(16.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). A(s.32) t = 4(q? + ml).t) = y^Fs(s. Symbolically we then have the situation shown in Fig.0s).t) = VtFt(t. Blankenbecler.9s)\2.t. B. Chew [48].t. u = . One now makes two hypotheses. there will be several such amplitudes which together describe all three processes). Treiman [25]. . Mandelstam [184] and R. A(s. For a Lorentzinvariant normalization of the initial and final states. Goldberger. one can write the amplitude as depending only on two.g. etc. the only kinematics we require in the following is given by the relations (e.4. if we had been considering the s reaction originally.§ Since the three variables s.The variables s. Thus. Khuri and S.t.(<fe)^(cos 6t). (a) Mandelstam's hypothesis: All three processes (described by the s. N. Ft = £ ( 2 J + l)F.
5 The Lehmann ellipse.5 for Yukawa potentials V{r) u: lm cose dr a(/j. 16. in the domain of schannel physical values of the kinematical of the ichannel). Lehmann [162].e.)e~ •fj. 16. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. For s — oo : cos 9t = 1 + s/2q^ — oo. But > > wave expansion X)(2Z + l)F.4 Yukawa versus Regge theory. However. 16. . Screened Coulomb Potentials Regge oc(t) Fig. the Legendre polynomial Pj(cos 9) does not possess a cut. the partial wave expansion is not valid in the physical region of the (i.334 Yukawa CHAPTER 16. So what can one do? 1 H. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. But for integral values of I.
5 The SommerfeldWatson Transform 335 T h e answer is.26) but for a different choice of the contour.6 The integration contour C".33) . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. to use the SommerfeldWatson transform. Moreover.t). „ i.1 / 2 . sin nl [16. Thus consider now the same integral but taken along the closed contour C shown in Fig. B Iml 1 \ i All \ r\ 0 / closed . 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. all poles of Ql > 0. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. the contour representation (16.16.e. i. contour C / / D ^ ^ .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. and there is only a finite number N + 1 in the domain of 9W = . N F(6.^ B 1 A Fig.6.e. 16.
i.e. this behaviour has been confirmed in high energy hadronic reactions. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. (16. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel.35) This result is valid for s — oo and t negative (cf. we have (16. Predazzi and T. W. Apart from more refined details. of course. the amplitude can be represented as a sum over ^channel Regge poles. Bose and H. for plots and discussion see R. 42. K.t)&B(t)sa°V.e. It follows that under these conditions. . MiillerKirsten [29]. Regge [229]. Eq. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. 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. J. In particular. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. Limic [177] and E. Omnes and M. 16. as we discussed briefly in the above. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. Foissart [223]. if ao(t) is the Regge pole with largest real part.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. Regge trajectories have. i. for high energies in • > the schannel). p. at high schannel energies. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N.32)) and finite. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo.336 CHAPTER 16.e.
Singularities appear also in the form of cuts in the plane of complex angular momentum.g. like a meson or a baryon. 20. Related aspects are discussed in Sec.6 Concluding Remarks 337 important parameter in string theory. MiillerKirsten [209].** Whereas the former are related to absorptive properties of a reaction. The way to do this is similar to calculations in Chapters 18 and 20. and is composed of quarks. For an overview see e. . a composite particle being one with structure. the latter are related to the distinction between "' elementary and "composite particles". The above treatment of screened Coulomb potentials is incomplete.1. called "Regge cuts". Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum.16. and also as socalled "fixed poles" . J. H. W. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated.
This page is intentionally left blank .
See F. the most immediate candidate is a trigonometric form like that of the the cosine function.0 < k2 < 1. motivated mainly by the regularities of the crystal structure of matter. Arscott [11]. however. since many analogous relations hold. 339 . are not much harder to handle than trigonometric functions. called elliptic modulus. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. The reason for these difficulties is. The Schrodinger equation with this potential.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. K being the complete elliptic integral of the first kind.Chapter 17 Periodic Potentials 17. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation.* These elliptic equations involve Jacobian elliptic functions like sn(a. M.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. depending on the parameter k. example 14. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. In essence these periodic Jacobian elliptic functions. 19 and p.) which are functions that interpolate between the n.g. 25.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. p. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. 2K or AK. and are themselves periodic with period e. 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.
J. those of integral order.2h2 cos 2z]y = 0. which does not possess any polynomial solutions. In fact. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. 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. the three Jacobian elliptic functions s n z .1. (17. For large coupling. Both of these domains are important for a host of other considerations. B. need not be afraid of them. We might mention already here the much less familiar but more general Lame equation. The first of these conditions is already tMathieu equation: R. Dingle and H. 8. each with its own characteristic eigenvalue. 17. Like the trigonometric functions sine. cosine and tangent. Periodic Potentials vide periodic potentials like trigonometric functions.340 CHAPTER 17. . these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. This is a significant difference compared with the Mathieu equation.. A main objective of this chapter is the calculation of these level splittings following the original calculations of Dingle and Muller^ with the perturbation method described in Sec. The Mathieu equation.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K.e. i. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0.7. which is conveniently written ^ + [\2K2sn2z]y = 0. which one looks up in books like that of MilneThomson [196] or Tables when required. Muller [73]. or Sw&ve Schrodinger equation with elliptic potential. 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. the Lame equation possesses 2 n + l polynomial solutions. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc.1. See also the discussion in Sec. The most important formulas for our purposes here are collected in Appendix A. Miiller. (17. [205]. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .1) where h2 is a parameter and — IT < z < IT. W.. c n z and d n z are handled with analogous formulas.2 .4. W. and to relate these to the weak coupling bands or regions of stability. 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. J. Lame equation: H. ''The reader who encounters these Jacobian elliptic functions here for the first time.
i. y" + \y = 0(h2). as in Fig. Although the mathematics literature on the subject uses the physical concepts of stability and instability. a bounded function.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. i.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. i. since the solution there is of the form cos vAz. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.e.e. In classical mechanics stability is treated for instance in connection with planetary motion. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. the case of h2 small. Vachaspati [242]. Adding the negative term — 2h2 cos 2z to A. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.e. It is this additional parameter function appearing in the solution of the Mathieu equation which attributes the equation its reputation as being particularly hard to handle. the semiaxis A > 0.17. i. Looking at the Mathieu equation for h? —> 0. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. these concepts are rarely explained as such there. 17. . Thus the solution of the appropriate Newton equation is essentially a periodic function like cos 8 (in the case of planetary motion this yields the polar equation of an ellipse). Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. h2 = 0 belongs to the domain of stability. Salem and T.2. we see that in a plot of h2 versus A. 17.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. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r).1. we compare the For interesting related discussions see M.^ 17.
2/(0)' yfr) . d(z + TT) cos 2z We see that there are solutions which are proportional. i.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. a = const. sin V Az or linear combinations. y(2vr) 2 a = —7—^ = 1.e. y+{z) . Setting z = 0. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz.342 CHAPTER 17.B and a. y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2). TT. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). ° ^ [1 + 0(h2)}. we obtain y(vr) y(2?r) a = — — a = —^^r. (17. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants.2h2 cos 2z]y(z) = 0. d2y(z + 7r + [A . We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. Periodic Potentials equations ^ p + [A . — • .2/i2 cos 2(2 + vr)]y(z + vr) = 0.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). y~(z) = 6sin vXz.e. (17. i.
y{z) = aty+(z) + 0y(z).W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3.17. This means.oWX] .3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. with constants a and /3. and by'+(?r) = —ay/Xy{n).y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above.y+(7r)j/_(7r)} = 0. we choose these with the following set of boundary conditions: y+(0) = a.a2b2\ a± ~ 2a6\/A . 7 y\z) = ay'+(z) + (3y'_{z). Using Eq. i.y+(7r)y_(7r) = 0 or abVXa2 . (17. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. 2/'(7r) = a ^ W + ^ . \y+>y\ = y+{ir)y'{ir) . 2/_(0) = 0.aa][y'_{n) . We can write therefore.e.4). y'(z + n) = a[ay'+(z) + py'_(z)}. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. (17. yV(0) = 0.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. y'_(Q) = bV\ with Wronskian W[y+. Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions. [y+(v) . This expression is equal to the coefficient of a2 in the quadratic equation for a. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .{ay'_{ir) + bV\y+(ir)}a Now from Eq. the determinant of the matrix must vanish.
we have <7± — CHAPTER 17. ** An important consequence of Eq. Eq. this implies y+(Tr)/a = ± 1 .5) a The parameter i/ is known as Floquet exponent.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. <7__cr_ = 1.7. <T__ + <T_ = 2y + (vr) . .1. _ / ± v 7 ^ V . T h e result for A is the expansion A = h4 2 . This calculation is demonstrated in Example 17.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2).6) We can derive various terms of this expansion perturbatively with the method of Sec. (17.e. 8..0 This series may be reversed t o yield the Floquet exponent. . (17.= — . This is the easiest application of t h a t method since only simple trigonometric expressions are involved.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58.344 Setting / = 2by/\y+(Tr).8) "Observe that for integral values of v (in lowest order of h2). 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. we have for the sum of the roots a+ + <7_ = 2 cos TTV or" C O S T T I ^ ^ ^ C O S V A T T . cr+ = 1/<T_. 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. (17.455A2 + 1291A . Periodic Potentials g= 2aby/\.22)) the boundary conditions of periodic solutions.e. i. **This point is not immediately clear from mathematics literature. Setting cr+ = e . (17. i.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ . where the constants are usually taken as unity from the beginning. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . Prom this condition we determine later (cf. (5i/ 2 + 7)/i 8 vz + — .
Schiifke [193]. (17. (17. (17.„„ x n.11) or smvz or e±ivz. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.35A + 15A2)/l8 /.e.2 how the expansions for small values of h2 may be obtained perturbatively. (17. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. It follows that in these cases i? = n2[l + 0(h2)]. (17. cos TT\/A 32A(A1) 2 .17.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 .9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.10) A = n2.3).jl2.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .35A + 8 . For these integral cases of v we demonstrate in Example 17. These cases therefore have to be dealt with separately. Example 17.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. 8. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. as will be done later.12) S e e J.1) 15A2 . — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. n an integer. .1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. VA = ±n. 124. p. Solution: We write the eigenvalue equation A = v2.5) and (17. = J—^ + o(hr). we must have a2 — 1 and hence e 2iv/A7r = 1} i. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). Meixner and F. W. (17.
(17. The recurrence relation can also be obtained by substituting the right hand side of Eq. (17. v)yu + {v. Proceeding in this manner we obtain the solutions ce. '.v±2) = l.v + 2)yu+2].. = . so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . Next we treat terms yv+a in j / 1 ' in a similar way.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. sev(z.h2). Meixner and F. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. The right hand side of Eq.2) _ __ _ (y. me±v(z.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. .e. Schafke [193].15) Hence a term fMy„+a on the right hand side of Eq.14) = = = cos(y+ 2)z + cos{y— 2)z. or A = 0(/i 4 ). v) = 0. With our perturbation formalism — as demonstrated in cases considered in Sees. . (17. as shown.2.346 CHAPTER 17. i. h2). siniy + 2)z + sin(i/ — 2)z. se and me as* oo i y = „«>) + yW + i. h2(u. v . i. of course.2)j/„_ 2 + {v. where except. see Example 17. when a = 0 or 2v + a = 0.(2) + . 17. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.11). by J. i.e. (17.16) directly into the original equation in the form of Eq.2) which applies when v is nonintegral.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.14) therefore leads to the following nextto leading order contribution (v.e. (17. (17. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.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.3 and 19.16) *These smallh expansions are convergent.e. (17.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v.3. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . W. since j / 0 ' = yv leaves unaccounted R\. and (v. up to 0(h2) we have A = 0. . v .!/) = 2A We now observe that DvVv = 0. for instance. e ± i ^ + 2 ' 2 + e±i("2)z.h2). ("• ^ + 2) (o) 2(2^ . (i/.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.„ + £ h2i £ P2i(2j)yv+j (17.
. Following the first few arguments of Example 17. + y~i) (17. p. For v = 1 we then have yi = cosz = y~\.1 with A = v2 — 2h2A but with v = 1. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . however.2—r——— (i/ + 2. . i.. 2 ( 2 i / . thus determining the quantity A.e. 121.14).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. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2.15) (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. .R{ .v)\ .7)..17) Inserting 1 for the stepcoefficients of Eq. Meixner and F. . The higher order terms naturally require a little more algebra. (17.. W.g. (17. (17.17. * Note.u) + h4 (". Solution: We start as in Example 17. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. v) ' ( 2 v + 2) (17.. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. Example 17.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. J. the solutions are not yet normalized."2). See e. Schafke [193].18) Again applying Eq. 0 = h2(u. one has to deal with each integral case separately as in the following example. fo^ + 2) . We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. In the case of integral values of i>.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]..2 ) v(y2. „ .
. m. and v is to be found in ascending powers of h2.24c) below) . 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. J. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. see below!) one obtains the additional terms given by Meixner and Schafke [193]. Periodic Potentials h4 r ••• . Jo dz cos z h2 cos 3z  iC2T It J* cos z H 64 cos 2 3z  64 It follows that 128 and the normalized solution is therefore h2 Vc .348 From this we obtain 2h2A = h2 CHAPTER 17.n integers. Eq. i.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. h4 (17. 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]. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. The reverse situation is later of importance. 120. the case of A an integer.e. p. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . We consider this case again in an example.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). cos z h cos Zz 8 h2 h4 ( cos 3z cos 5z — cos iz A 1 8 4 l 8 48 h4 (cos 3z cos 5z 1 4 I 16 48 (17. (17. W. We introduce a normalization constant c and rename the normalized solution y then yc. 123) (there the solution is called cei(z. Meixner and F. With normalization (not to 1. In the above we considered the case of v an integer and calculated the eigenvalue A. Schafke [193]. h2).* The associated solution is ym+ym.
3 1 x 4 + 8) 274233 (.. J.• 1 + 5n h 2 s 2932 Setting in cosm. ^S. o 1• • • . We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .4)Av A2v A ~ 32A(A . .. . Gubser and A.4 ) + .2 Cosine Potential: Weak Coupling Solutions Example 17. (17.17.1) (A .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A .= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . S.('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 . on the left hand side of Eq.Q . Liang and Yunbo Zhang [189]. Hashimoto [124].l ) 2 (A . (11A2 . fe47T2(A4) 4%/A(A . we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . (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 .9b) and considering the approach A — 4 gives > (A4)2 + . J . (17. Manvelyan.3 5 A + 8)TT2(A4) _ 3 / / 64(A . MiillerKirsten.3 1 A + 8) ( A . W.1) .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. Substituting these expressions into Eq.9b) about A = 4.9b) v — 2 + <5. H. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.1)2/A TT2 +h8 (15A2 .
We now derive these boundary conditions..W4W=UtWSta±I(.w^. Fig. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. 28 .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.21b). "These conditions.6.10) applies. 17. M. i. Taking in particular Eqs. Arscott [11]. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. as also conditions (17. may also be found. = and s+w 2tW±I.350 CHAPTER 17.21a) and (17. pp. (17.4) into account.29. These are defined by specific boundary conditions. (17. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. . for instance. in F. solutions for integral values of v. 17.1 Boundaries of domains of stability.2. We verify these for the large/i 2 solutions in Example 17. bV\ y+(0) 2/40) 07.e.
1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. we obtain y+(7r) _ i  2y_(f). /7r\j/!_(7r) y L l 7TJ . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . . ^\y±M /vr\yL(7r) .y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. which after some rearrangement can be written y+(vr) _ .4(f) (17.17. . o y+(fy(f)ob\/A _ = 1 + 2I&A/X . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f.21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. /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. Then y+ y 351 U .
22) * a2n+i(h ).1. (17. where the notation for the corresponding eigenvalue is put alongside on the right: y+(f) = 0 with y+(f) = 0 y'{\) = 0 with y _ ( § ) = 0 with with A .T + . was treated in detail in Example 17. y > ce2n+i=qo.. IT. These may be subdivided into classes depending on whether the integral value n — 0.— — — • • • . This is shown schematically in Fig.3. A > b2n+i(h2). (17.2): hA 7h8 29h12 + Y l28^04. of v is even or odd.' fh2 V2ceo = l . (17. 2K.• a2n(h2). y > se2n+i=qo.5) that the left hand sides of Eqs. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. We see that there are four possibilities and hence four different types of functions.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. Mr7nA (17.2. (17. 17. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . These are defined by the following boundary conditions.352 CHAPTER 17. 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. 19.1).2 we require this relation for nonintegral values of v. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .24c) 8 h2 sin 4z \ . ce± = cos 8 64 h4 5h8 b2 = 4 1 .. 17. the case of ai with solution cei (of period 27r). in this order. . 27r.+ .. (17.24b) 8 h2 z —— cos 3z + • • • .. (17. There the second term on the right is nonzero.24d) v 12 ' **To avoid confusion we emphasize: In Sec. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. A 2 y • ce2n=q0i.21b) must be ± 1 .24a) ao = hA h6 b\ = 1 — h —— + — — • • • .21a) and (17.. y > se 2n +2s 9o +i The functions so defined have respectively period 7r.23) We can now see how the domains of stability and instability arise. (17. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. A *• b2n+2(h2). ..
1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig.> (17.17. 17.2 The cosine potential.2cos4.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . The cosine potential cos 2z depicted in Fig.3 we show schematically how the low energy solutions for integral values of the Floquet exponent are related to the asymptotically degenerate harmonic oscillator eigenvalues. 17.24e) 17. In the case of the cosine potential. W. .17„. Thus in Fig. 17. 17.z3 ce2 = cos 2z .1 that the boundaries of regions of stability merge to lines. of course. 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. . 17. B. .2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. the walls are not infinitely high and hence tunneling occurs from one well into another. We can see from the inverse of Fig.3.3 17. Miiller [73].hz — 12 . 353 . J. 13824 .2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R. 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. Dingle and H.
i r ) ~ 1 — 21 z ± 7r Changing the independent variable now to the equation is approximated by A + 2/t2 dx2 + 4/i 2T y = 0. of course. Thus in this case the quantity on the left is not an exact odd integer.2. (17. we therefore set 9 A A = 2h2 +2hq + —.26) 17. Taking the remaining terms of the cosine expansion into account. where A / 8 is a remainder.3) and is. Ion (17.354 CHAPTER 17. Sees.1. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. (17. 8. Periodic Potentials minima at z — ±ir/2.3...2/i cos z V = 0. and we set this equal to q.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. n 0.. but only approximately so. In the case of finite heights of the potential barriers. simply the onedimensional Schrodinger equation for the harmonic oscillator.27) .. there is tunneling from one barrier to the next. 4 (17.2 The solutions We insert Eq. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . as in the case of the cosine potential.2 and 8.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . We therefore wish to expand it about these points.
31) We make the important observation here t h a t if one solution of the type (17. 16/i/ (17. (17. and substituting this into Eq.17. Choosing y(z) = A(z)e2hsinz.30) D A): COS « = 4z + 1 iq sin 2.27) we )A = 0. can be . the linearly independent one. another solution.29) is known. 17. 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. (17.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. (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).3 Schematic picture of Mathieu equation eigenvalues.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . 2 (17.
30) amounting to Ri ] q = . 2 A A = («. (9. ? .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. ? . (9l)(93). ._ 4 . when i = 0 — the terms not involving Aq in (17.9) (9. Aq+4 2Jh .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.e. D{q%Aq+4i = 0. (17. ^ ) A + ( ? . (17. 2 /i D^Ag = 0.9) = 0.g + 4) .4)A.36) Then ^4.94) = (9 + l)(9 + 3).94) \ — Aq4 (17. i. of course. With some algebra one finds that 16^ + where (9.35) But since and so a term iiAq+n on the right hand side of Eq. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout.Q)Ag + ( ? .\z To lowest order the solution A is A^> = Aq.33) (17.356 CHAPTER 17. (17. 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.^ 6T [(«>« + 4 ) A ? + 4 + (Q.35) not yet taken care of is set equal to zero.32) cos^'+i) [\K . (17.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17. (9. 2[(92 + l) + A]. 1 + 4 ) ^ + 4 + (^.4 ] .9 + 4) = (9.4 ) ^ .
. (17.4.38) Aq8 + 1 2 Clearly we can continue in this way.37) and so on — left uncompensated — must vanish. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17..30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. except those involving Aq. q. (17. . A. q .g4) 4.q + 4)(q + 4. q . Then for A = A(0)+A(D+A(2) + . . can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.g + . q. (17. Thus in particular A(z.40) . (17.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation.4 .17.4) 1 1 (q. h).4. (q. q. Thus 0 = (g. +4. ) ^ 4.35) and (17..4) (q . q. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known.39) which is the equation from which the quantity A and hence the eigenvalue A is determined. Q . h) = A(z. h). h) = A(z.q + 4) (g. A(z.8)A. the sum of the terms in Aq in Eqs..4) (q ._ 8 (17.g + 4)(g + 4.)+ & ! = « ( . q .30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .<J+4 Aq+& + 1 2 1 . q (17.q 4)Aq+4 q + 4 1 q + + + + + <^(.37) Here all the terms.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17.q 8)Aq+8 (o)' ^±V(q 1 4. to satisfy Eq. (q. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.
" Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (91) H (17.41) so that the early successive contributions decrease in magnitude. i. We return to Eqs. (17. We observe that for the large values of h we are considering here. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). Periodic Potentials Finally we note t h a t the above trigonometric solutions of t h e Mathieu equation are valid in the domains cos 1 4 1 Z 7T ± » 1 2 VK' (17.42) (17. 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. it is found that D(B) z+ — ]B = 0. where Hen{w) is a Hermite polynomial when n is an integer.i ) We encountered Hermite polynomials earlier in Chapter 6. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. (6.45) {B) We recognize Dq as the differential operator of the Hermite equation.43) 9 2eh . Before we study the eigenvalue equation in more detail.46) 2s(«.358 CHAPTER 17.67). 2 w2 + H* (17.e.44) where D(B) 9 = dw2 $_ !)• (17. Note different definitions in use. and discussed there. (17.28) 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 .d2B w dw2 + w(l w dB_ dw d dw 1. The differential equation there is Eq.) i ( Q . .
.3 ) \He* ! (9+1) H.47) where the coefficients are the same as before. y — Cexp(2hsinz) and Cexp(—2/isinz). (17.e.e.e. A{wz (17. i.48) This shows that this solution is valid in particular around z = ir/2. i.. i.e.. The solution with contributions B^°\B^>. which is an amazing and unique property of the Mathieu solutions — cf.7 T H 1 2 Z 4 <1.qi)Bq_4.17.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. Again a change in the sign of z throughout yields the associated Mathieu function y = i?exp(—2hsinz) with B[w(z)] = B[w(—z)] and domain of validity in particular around z = —ir/2. q)Bq + (q. the expansion for the quantity A being identical with that of the previous case.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. 16hJ (17. Finally we obtain a third pair of solutions.51) .49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .q + 4 ) £ g + 4 + (q. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. .34). as in Eq. (17. with the equation C" + 4h cos z C + 2h ( q .4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. (17. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w). 1 COS I .50) C = 0. also the same coefficients in the case of functions C below. i.
A Hen .q. show that the perturbation remainder of Eq. q)Cq + (q.44).52) and the coefficients are again the same as before.54) where the proportionality factor is complex.53) This solution is valid in particular around z = —7r/2. (9+3) (2n 2 + 2n + 1) + .(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .(w2 + A)Hen. These domains of validity are indicated in Fig.h) ocB{z. (17.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. 2 J where n=(ql).h). (17. (17. 17. (17. He'n(w) = wHen(w) Hen+i(w). (17. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.4)C g _ 4 .7 T 1 2 Z 4 <1. ie 1 COS I . . 2' the remainder 72.360 CHAPTER 17.w2)He'n . Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.q. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq.q. We note that C(z. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.4. q + 4)C 9 + 4 + (q.4: Hermite function linearization of perturbation Using the recurrence relations (6. Example 17. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). A closer look at that expansion is our next objective.59d) of Hermite functions Hen(w). can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He.n2(n l)Hen2 . We have thus determined three pairs of solutions of the Mathieu equation. each of them associated with one and the same expansion for the eigenvalue. 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.
g4)(q4. l)Ag4 + P2(q.q4) (17.)Bqi(w).57) ..q + 4)Bq+4(w) + (q. .j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V . with (.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K .4) = (q . 17.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. q) = 2[(q2 + 1) + A].0 and —4.q + A){q + A.q 4. (?) 1 1 + (9.56) (94. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.3).g. e. .3. no intermediate move to or from q being allowed.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). ^ 1 Hv" '^ ^' ^ ' .55) M2 = M. (q. the solution y = exp(2/i sin 2)^(2). r j=r. q + 4) = (g + 1)(? + 3). In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.. 2)Aq_8} + • (17. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.17.1i . q)Bq(w) + (q. ) . q .q + 4) (q + 4.i ) ( w ) 2l(. (q.l)[l(q _!)]. 2 + ^ M 3 + (17.l)(q .' 361 in 2*f*1>[i(gl)]! = (q.g) 1 1 and so on. (q. in the following compact form with coefficients Pr(q. (b) The allowed moves are +4. We can write the solutions obtained previously.
j)Pr+1(g. (17. P2(q 2) Pl(g._1) = Mzil.i (g.g + 12) 1 2 3 (g.Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j.g + 8)(g + 8. g + 4j)P r _i (g. for instance.j) = (l)rPr(q.4. Dingle and H.e. (17. From the coefficients (17. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g.j).34) we deduce that Pr(q.g + 8)(g + 8. x)_ (g. and in general* M2r = J]j[P2(g.3) ^3(9.g + 4 ) ( g + 4. . Thus we have. M2 = P 2 ( g . j + 1) (17. g + 4j)P r _!(g.g + 4) + 1 1 1 and so on. Miiller [73]. W. Periodic Potentials PlM)=(M+i). j . .g + 4)(g + 4.g + 4)(g + 4. J.j). the factor j removing a duplication of factors in the denominator at the junction. 2) 1 2 ' ' ^ ' ^ 1 (g.j)^2(9.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.60) M2r+l = = *R.g + 4)(g + 4. 0) = 1 and all other P0{q. By the above rules we can write down the recurrence relation for the evolution of a coefficient Pr by steps from the coefficients Pr\: jPr (g.58) with P 0 (g. B. j) = (g + 4j . p = (<?><? + 4 ) ( g + 4 >g + 8) .j).j)] = 2j]jPr2(g. as jP r 2 (g.g + 4)(g + 4. i. j) +(g + 4j + 4.P 2 ( g .1) + (g + 4j.g + 4) ^3(9. and the contribution from a sequence of r moves from q + 4j back to g. l ) .j ^ 0) = 0.l ) .g + 4) 2 1 (g. g + 4 j ) P r .362 Clearly CHAPTER 17.j).j).
These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 .3. S.62a) f R . h — —h.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. 17. [116]. as a result of tunneling between wells.284g + 57) + (17.92q3 + 70g2 . remains to be determined next.2 * +3) + ^ 2 (9g4 .( 3 ^ .^q(33q4 1 220 hA 1 (63g6 + 1260g4 + 2943g2 + 486) 225/i5 rg(527g6 + 15617g4 + 69001g2 + 41607) —3T76 (9387g8 + 388780g6 + 2845898g4 + 4021884g2 + 506979) —WTfs q ( 175045qS + 9702612 ^ 6 + 107798166g4 (17. as pointed out earlier. W. 17.61) +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. . It will be seen in Chapter 18 that in the case of anharmonic oscillators.17. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness.5 the following proportionalities there: B[w(z)] = aA(z). Dingle and H.* 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. Ince [136]. B. We observe that there are regions. In fact we demonstrate in Example 17. where solutions overlap. a= (8/i)3(9x) [1(91)]! 1 . Miiller [73]. of course. J. [137]. Up to > > this point. the solutions and eigenvalue expansions again have the same type of symmetries.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . *E. L. Goldstein [115].4. Its precise deviation from qo.^ .
32). 17.62b) *z domains of overlap Fig.1 4 . The first term of the expansion of A(z) is the function Aq(z) given by Eq. and determine the constants a and a.5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .1 ) ( 7T + .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. a .364 and C[w(z)] =a~A(z).(17. We write and expand this function as follows: .7T H Aq(z) = = C O S ^ .4 Domains of solutions and their domains of overlap. Solution: We begin with the solution containing the function A{z).1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . Example 17._ ^ l CHAPTER 17.
2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 . These are the conditions of Eqs.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.22). with e.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.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.g.!)(. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.53) for the expansion. i. the coefficient of the dominant factor c o s ( . (17.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. we obtain Bq[w(z)} H. (17. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73].46). A*1) given by Eq._i)]! w*te» 2l(91 (. 17.17. Since the solution with A is obtained from that with A by a change of sign of z.92q 3 + 70q 2 . these are conditions imposed on even and odd solutions at the point z = it/2.2. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.i ) 1 (g. 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 .36).284g + 57) + • The other relation is obtained similarly. with (8h)li''1) (3g2 . we have to impose the appropriate boundary conditions.61) for the connection and Eq. (6. As we saw there. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. \ql)/2(w) 2i(9D[i(.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . (17. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.g + 4) (q + 4. (8. s .63) In Example 17.
22) are therefore given by the following boundary conditions: 1T y'+ v+fao I'=° y a a\dwjw=0 B[w = 0]c2h a a 1h dw\ ! ' . 80. He2n+1 (0) = 0. Thus y+ = ce and y_ = se. He'v(0) = Hev+1{0). (17. i. The functions defined as in Eq.e. Periodic Potentials We can extend the solutions (17.65) 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' (17.62a) and (17. Magnus and F.62b). With the help of the duplication and inversion formulas of factorials. Here v is the not necessarily integral index of the Hermite function Hev{w) with Hermite equation ( jir w—+v)Hev(w) \ aw1 aw J = 0.366 CHAPTER 17. one obtains the first of relations (17.e. The second of relations (17. Oberhettinger [181]. a ± C[w{z)}c_2hsinz^ a (17. give the following expressions for polynomials with argument zero: ~19)"(^)!.63) to the domain around z = ir/2 by using the proportionalities (17.66).66) then follows from the recurrence relation (6.g.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. . W. p. Then y± oc ^ M e ^ i n .22) and to correlate the solutions to those for small values of h2. i.66) ^E. and then replacing 2n by i>. He'u{w) = wHev{w)He„+1{w).59d).» a \dw _ C[w = 0}c_2h a w=0 = Q (17.
from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.39 2~^h? (17.67d) Considering now the second of conditions (17.67b) <f ~ 82g^ .106g . B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .Qo) + 0[(q .17.» 2 6 /i the condition reduces to the equation cot{ .82g2 .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative. Hence expanding the cotangent about these points.68) Inserting here expansions (17.14) from which one obtains 1 (q + i) TV cos{f(gl)}[i(g3)]!' and the duplication formula (2Z)\\/TT = z\(z — l/2)!2 2z .! ) ] ! [ .( g .7.136g + 9) + • • • .1) 7T .g0)3].67a) and (17.87 4 2 ¥w (17. we have cot j ^ ( g ~ 1)} = ~\<(l .69) If the right hand side of this equation were exactly equal to zero. . (17.67a) g 106g 87 2 14 /i 2 (17. and similarly for C and its derivative. and using the reflection formula (8.40g3 + 18g2 .67c) 4 2 \dw)w=Q C[w = 0] [\(q . the solutions would be given by q = go = 3.67c). f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .11.39 (17.
The remaining condition (17. Thus we have 2 (16/i) qo/2 T2 4. = _a a 6 _4h (17.40gg + 18gg . .120g£ + 467g . Considering now the first of conditions (17.528gg + 3307Q .40gg + 18gg . (17.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg ..e..136q + 9) + .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . i.. .65) leads to the same result except for a change in sign.136g0 + 9) + • Clearly the last of conditions (17.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17..73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.72) This leads again to the result (17.(17.408g0 + 1089) 2 /i +• (17.65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i . we use \(q)~\(q0) + (qq0)(^J .5.70) except that now qo = 1.9.40g3 + 18q2 .65) yields the same result except for a change in sign.70) + ^ 2 (9?o .h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .
J. quite apart from applications. . (17. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. The dominant contributions were also found by S. (17. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. example 3.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . results from the boundary conditions. 20. this becomes 369 = (16/i)^+1e4h l .74b) The above results for the cosine potential are in many respects important.A_(g0) ^ . See also F. W.t From Eq.61). i. Dingle and H.2... In the literature that we follow here* in all cases several more higher order terms have been given. (17.e. We see here that *R. The tunneling effect.e.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . though approximate. Arscott [11]. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.3 we call this difference 26E„n. 17..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. + ^ 2 (9<?o + 89o " 78g02 . 9o = 2n + 1. and not the separation of these as a result of tunneling. B. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. n = 0. i.3.87) = Azpteo). just as any solvable problem is important for the insights it offers in a concrete and transparent form. M.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. 120.3. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. Miiller [73]. and that for the odd Mathieu functions se qo+ i and se9o (lower. The complete. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels.17. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. p.1. Goldstein [116]. (17.88q0 . minus sign). . *In Sec.
The splitting can also be derived by the pathintegral method as will be shown in Chapter 26.370 CHAPTER 17. (17. Example 17.74b) is also important for various other reasons. the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). yA{z) = Aq(z)e2hsin '.6: Translation of Solutions For the largeh 2 even and odd solutions (17. In fact the perturbation theory developed in this context and tested by application to the cosine potential is of such generality that the recurrence relation of the coefficients of the eigenvalue expansion can be looked at as a difference equation. Then we obtain first y±(z) = yA(z)±y^(z).74b). in effect. Naturally one expects the solutions there to be related to those of the periodic case considered here. Thus we do not reproduce every step.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . evident through the exponential factor exp(—4/i) in Eq. Periodic Potentials this nonperturbative effect. The exponential factor represents. y^{z) = Aq(z)e~2hsin *. (17. and one can compare the methods. The explicit derivation of the level splitting (17. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. _3/+(z) + ____2/_(z).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. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). . 17. the Schrodinger equation with the cosine potential is a special case of several other more general cases. results from the imposition of boundary conditions. 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. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs.2. As mentioned at the beginning. The splitting can also be obtained with some methods of large order of perturbation theory. . 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. T y'_(0) y'A±*) y{z). M±TT) . 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.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). We begin with the solution containing the function A(z). From Eq.
c o s  ~(<1 T 1) f y .T ( ? =F 1) p + (0). A ?q( Z = ±TT) = .4 17.[A'qM + 2hAq(n)] = iV2(q . ( . = ±ir).Ah) ~ 32ft.4ft) ~ 8(q . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . 17. 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+. where Aq(0) = Aq(0) = V2.1 )  .3 The Ellipsoidal Potential Replacing VA. and y+(z = ±TT) = V2(q .4ft) sin j ^n(q . (l^ifr^i^O + Cl^ifo^j/aCO cos  .1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials..q).7 r ( q = F l ) } j / .( ± 7 r ) = c ° s  .7 r ( g ^ l ) U + ( z ) q : i s i n { . which is easiest by comparison with the foregoing treatment of the Mathieu equation.) cos < 7r(q 1) [ = .17.( 2 ) .z ) . y'_ (0) = \/2(4h .• Then y'+ (0) = 0. Replacing sines and cosines by the functional expressions. the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). (17.2hAq(n)] . Aq(z) The derivatives can be handled with the help of Eq. Similarly we find y(z±ir) = = and by putting z = 0: V.g A ? ( z = ±TT). A' (z = ±ir) = (.%/2(g . we obtain the conditions stated at the beginning. etc. y+(0)~2V2. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .y]z=„ = [y+yL .4ft) cos i TT(<J + 1 ) 1 . The Lame equation has not been a widely known equation of mathematical physics so .31). but also to enable a brief familiarization with the Lame equation.( z ) From this we deduce for z = 0.VA DV 371 1 V±. — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .yV+\z^ = [j/+(0)] 2 (<? .4.4ft.
5. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940..0. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . but also cubic potentials.0. But more recently it has been observed to arise in various contexts. so that Floquet's theorem cannot be immediately applied . Arscott [11]. *A. p.75) ' T h i s means double well.0 and . k) for k = 0.98. Periodic Potentials far. and was continued by Erdelyi. § F . In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.. H." Fig..372 CHAPTER 17. Erdelyi [85]. J.4sn4u]y = 0. however..0... L. Ince [138].fi2A. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). .* As Arscott [11] (p. (17.1. W. 194) remarks to parallels with the Mathieu equation:".* and the work of both prepared the ground for presentday investigations.5 < z < 5. H. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane. the nonoccurrence of the Floquet exponent. M.. This line of investigation. 194) remarks. 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 .K2sn2u .Q. f E . Liang. as alluded to at the beginning. The most conspicuous difference compared with the Mathieu equation is. inverted double well and cosine potentials (J. 19.75. 17.5 The potential sn2(z. MullerKirsten and D. Tchrakian [165]). Arscott [11] (p.
Jianzu Zhang and Yunbo Zhang [206]. in front of the second derivative has to be kept in mind (mo being the mass). Although the results have been calculated for the ellipsoidal wave equation. (17.17.75) reduces to Lame's equation and one writes K.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k. i. cmx and dnu. 17.4. i. the usual factor —h2/2mo. (17.5 for several values of fc. Eq.77) into Eq. and of the derivation of the level splitting'.77) where q —> go = 2iV + 1. (17.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. The first step is to write the eigenvalue A as A(q.2 = n(n + l)k2. (17. . The function sn2u is plotted in Fig.e. The range of the independent variable u is 0 < u < 2K. W.K2sx?u}y = 0.78) We follow in this brief recapitulation the description in H.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. J.e.K) = qK+^±. (17. . If we put fl = 0. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). N = 0. K being the complete elliptic integral of the first kind. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects.1. the equation ^  + [A . .2k. In order to distinguish the above equation from that with the Lame potential. 17. The second step is to insert (17.\k\ < 1. 2 .3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). we consider mainly the Lame equation. MiillerKirsten. is the elliptic modulus of the Jacobian elliptic functions snu. In comparisons with the Schrodinger equation. . in the case n2 — oo.
which are respectively even in u (or snu) or odd. B and C.K). A into equations in terms of the variables z(u) = ^—{k' .e. (17. Thus in the third step two more pairs of solutions B. one pair in terms of Hermite functions of a real variable.80) The domain of validity of these solutions is that away from an extremum of the potential. i.77).79) The very useful property of these solutions is that for the same value of A (which remains unchanged under the combined replacements q —>• —q. i. but solutions B. /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. Periodic Potentials . A are derived. . A second solution is written y = A(u) exp < / nsnudu >. dnu± cnu K Thus one can construct solutions Ec(u). C replacing yl. the other in terms of those of an imaginary variable. more precisly for dnu = cnu F 1 » . determine their proportionality factors) in domains of overlap (their extreme regions of validity). by transforming the equations for A.K)=A(u. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. (17. k' = Vl~ k\ \dmi±cnu/ 17.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). A(u. ^ (17.q.e. \/8K (dmz = cn« \ ' F .82 . dnu =F cnu . A. (17. . / z J . one has to derive new sets of solutions there and match these to the former (i. one obtains again the same expansion for A. C which are valid for dnu±cnu <1. _ Solving the resulting equations iteratively as before.Es(ii).q..e.K — > A(u)=A(u + 2K).374 where CHAPTER 17.
Tricomi [87].e. . Oberhettinger and F.^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) 8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) 64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£.38g2 . i. a of _ _ B = aA. p. Es(2f0 = Es(0) = 0. W. *A.85) The first three terms of this expansion were first given by Ince [138].6(5g4 . G.17. as well as \duJ2K 11 (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) . (17.384n2k4(q2 + 1)} q r {(l + A.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.4. (17. C = a~A. Magnus. who obtained this expansion for the eigenvalues of Lame's equation (i. F.e.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . £2 = 0).63)} + ••• .86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. This expansion is found to be in the more general case of Eq. 64. the ellipsoidal equation. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.75). Erdelyi.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. Miiller [205]. (17. W. 17. J.
W. . from which the difference q .91) tt H. Muller [205].Es^ 0+1 .136g0 + 9) n 3.4k2(q2 + 2g0 + 5)} + • • • ].2 n K 2 {3(1 + 1 +128fc2(2g^ + 9ql + 10g0 + 15)} For the two lowest levels go = 1 and one obtains for their separation AA(1) 2(4^) 3 / 2 (lfc)*.376 CHAPTER 17. J. and the lower to Es*>+1 or Es*5.:J {3(1 + k2)2 (9q* .89) one obtains the eigenvalues from which the level splitting can be deduced. (17. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .j 26K2 (1 85) = A(g0) + (q .88g0 .1 / 2 (2TT) 1 /22« l + (l_fe)^«( i . Periodic Potentials These conditions define respectively functions EcX°.i Here the upper sign refers to Ec*> l or Ec£°.78g^ . Finally expanding A(g) ~ A(g0) + ( g .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) . 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). 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. 2K and 4K respectively.2 /c +256fc2g0(g2 + 5)} (17. One obtains with go an odd integer QQo =F2 2fl + ir\lkj 3(g02 k\K/k 8K qo/2 \lk ) 2 1 [i(g0l)]!L + l)(l + fc2) 25 K ^ 2 J J .g0 is obtained by expansion around zeros.90) + 3.87) (17.Ec£ 0_1 and Es2° of periods 4K.40g03 + 18g02 .g 0 ) ( ^ . 2K.l n 2^+ i J + 0(«i) (17. go being an odd integer.
K2 = h2 (say).90) one has in this limit: /l±±\ ~K/U = e .4. Ince [138].74a). then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. V.J1 O. Miiller [206].f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. (17. in which they dubbed the classical configurations Lame instantons.4 h ( i e M 9 0 7 2 \. Hence the conditions fi = 0. Rao [79].e. L. J.2h2 . but also more easily generalizable.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. i.93) (17. Replacing u by x ± IT/2.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. 17. (17.* M G . (17. h . K = 2/i.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. this equation becomes y" + {A . Without going into details we mention again (as at the beginning) that there is also a specialization from the ellipsoidal wave equation and its solutions to spheroidal wave equations and their solutions. A = 0.Ah2 sin2 u}y = 0.* Apart from the choice of notation. however. (17.86). Dunne and K.2h2 = A.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. One can see.2h2 cos 2x}y = 0. ./2i and hence A^ A/ A AU I2 . Thus in the case of the dominant contribution of Eq. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. A . it = x ± ^ .17. the conditions (17. + E.87) agree with those of Ince. W. *H. that calculations with the Schrodinger equation are not only easier. .
Cho. MiillerKirsten and D. Khare and U. J. This is an interesting problem. Ouvry [276] and Jianzu Zhang. Ganguly [103].5 Concluding Remarks The potentials we considered above will reappear in later chapters. 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. W. . J . A generalized associated Lame potential has been considered by A. MiillerKirsten and J. M.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). A. all basic potentials. 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.W. A. Shiraishi [49]. § .e.^ The equations considered above also appear in diverse new problems of physics. See e. N. 1 1 See Z. i. See Chapter 25.378 CHAPTER 17.Y.Q . Gu and S. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. J. Quian [123].§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. Thus. D. for instance. H. H. W. Rana [287]. Periodic Potentials 17.g. Kan and K. double well. Sukhatme [148]. H. Liang. S. inverted double well and cubic potentials. **Y. Tchrakian [165]. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. de Veigy and S. A host of related elliptic equations has recently been discovered and studied.
The recent work of R. W. In particular the investigations of Bender and Wu. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. M. This seemingly simple problem revealed extremely rich internal structure . Weinstein [281]. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. both of which make the calculation more difficult.Chapter 18 Anharmonic Oscillator Potentials 18. J.". T. MiillerKirsten [163]. Wu [18]. ' T h u s A. MiillerKirsten and A..t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. f C . [19]. 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. Wiedemann [4] and a revised version of parts of this reference by J.. T.Q.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory.* 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. Wu. J. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. W. Bender and T. D. Bender and T. Achuthan. Priedberg and T. Liang and H. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. 379 . in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. H.
V(z) = \\h*\z* + \\<?\z*. with the potential described as an inverted double well potential. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. These contributions may be given different signs.1 The three different types of anharmonic potentials. described as the case of the double well potential.380 CHAPTER 18. *~z (1) (2) (3) Fig. 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. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. and thus lead to very different physical situations. 18. 18.1. The three different cases are: (1) Discrete eigenvalues with no tunneling: In this case V{z) = \\h'\z* + \\<?\z\ (2) Discrete eigenvalues with tunneling: In this case. V(z) = \\h*\z2 I 2i 4 \cr\z . Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . (3) Complex eigenvalues with tunneling: In this case.
In this case our aim is to obtain the aforementioned complex eigenvalue. The level splitting will be rederived from path integrals in Chapter 26. 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). that the general applicability of the method becomes evident.1 Introductory Remarks 381 normalisable wave functions. as is sometimes hoped. however with decay as a consequence of tunneling. (18.e. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). We begin with the latter. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side).18. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. since the potential decreases without limit on either side of the centre. Case (3) is seen to be very different from the first two cases.86) below. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. The result is an expansion in descending powers of h?. 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. i. The shift of the eigenvalues is best calculated with straightforward perturbation theory. except for a change of sign of c 2 . The imaginary part will be rederived from path integrals in Chapter 26. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. Thus we are mainly concerned with the double well potential and its inverted form. 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.175) below. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17. The eigenvalues of this case are given by Eq. It is this expansion which led to a large number of investigations culminating (so to speak) in the work of Bender and Wu who established the asymptotic nature of the expansion. (18. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. . The eigenvalues of this case are given by Eq. it will not be possible to obtain the exponentially small contributions with convergent expansions. however the central hump with troughs on either side permits tunneling and hence (if the hump is sufficiently high) a splitting of the asymptotically degenerate eigenvalues in the wells on either side which vanishes in the limit of an infinitely high central hump. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way.
1 refer to the two cases. (18. 18.2. . The potential in this case is given by V{z) = v(z). 18. If suffixes 1/2. 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 .2 18.2.1 and more specifically in Fig.382 CHAPTER 18.z Fig.3) "1/2 harmonic oscillator h8/25c2 . 18. We take here h = 1 and the mass rriQ of the particle = 1/2. h 4 2 2h1> ~1/2 2c?.2 The inverted double well potential with (hatched) oscillator potential. a point which has to be kept in mind in comparisons. v(z) = ^h4z2 + ^c2z\ (18. (18. Anharmonic Oscillator Potentials 18. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.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.1) for h4 and c 2 real and positive.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature.
The result will be that derived originally by Bender and Wu. 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.. (18.. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.4) we can rewrite Eq. to match these in domains of overlap. 18. (18. (18. i.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.2. The problem here is to obtain the solutions in various domains of the variable. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.e. 4 .2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18.2. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different..7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). (18. \h?\ — oo and c finite. /i). n = 0.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite. (18.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0.1.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. and a variable w defined by setting E= b* 1 + A A and w = hz. y{w) oc Di.18.1/7  Vg(w)y(w) = o(J^j.5) becomes . _i\(w) and q = qo = 2n + l. the harmonic part of the > potential dominates over the quartic contribution and Eq.d2 ..
as we shall see.10) into (18.10) E= qh2 ^ j. h) is obtained from the perturbation expansion of the eigenvalue.384 CHAPTER 18. •A~2 hrz" 2^\ c^z V2 —— + (18.14a) 1/2 VA( ) = A(z)exp (18.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18. as encountered and explained earlier. (18. these solutions are not valid around z = 0.+ ^ . Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. (18.14b) .2 A (18.13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0.11) 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 ) . This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. In order to arrive at solutions we set in Eq. Inserting (18. This observation allows us to define the pair of solutions yA(z) = .9) z =0 these Thus these c2z4 1/2 + • 2 (18. 1 d r A!{z)±iA{z)±{ 0.2.9) we obtain 1 h4z2 c2z4 A L2 (18.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.2 We are concerned with the equation d2y(z dz2 where ^ 1 .12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^.4(2. Before we return to solutions we derive a new pair which is valid in the adjoining domains. We observe — before touching the square roots in Eq. be and A = A(q.
Aq(z) approximate the solutions of Eqs. / c z Aq{z)+0 1/2 .4 . Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. (18. /H 4 Z h +~C Z 21.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).20b) .e. 1 (18. 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.16) For large h2 we can write the Eqs.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18. (18.13) and A(z) that of the lower of these equations. (18..13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. Since these higher order contributions are of little interest for our present considerations.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.18.19) We see that one solution follows from the other by replacing z by — z. (18. i. 1 2 4 2 Aq(z) + 0 (18. Clearly Aq(z). 385 (18.zh2 =()*2"(18.34) below) — to be derived in detail in Example 18. (18.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.1 and as a verification again in connection with the solution y# — as the other solutions. We take the square root by setting z2h^1/2 +.15) where A{z) is the solution of the upper of Eqs.
one observes that with £>.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.. 03 = . (18.21) Example 18.1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. to obtain in leading order the eigenvalues E. X » / A q + 4 i = Mg+4i . procedure.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . A A _ 3 .20a) and (18. . (18. Vq+ii = Vq + 2i. W i t h proper care in selecting signs of square roots we can use the solutions (18. h2. Anharmonic Oscillator Potentials i.v ' = o (« . z) ± yA(q.17) and (18.• Using Eqs.20b). a2 = 2 1 .18) we obtain A » = H^rL^z) 2 z . (18. . h2 —> —h2). zA'(z) where the expansion coefficients are given by ao = l. h2. * = — .f .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. i. (18.. A = .20a). (18. 1 . 04 = 5 . away from the central minimum. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).! ) ^ . we write > V±(z) = . „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. ft = .[yA(q.1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"^ The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z). z)).2 .. i. 128 8 16 A = i.e. 2 . „ . c*i = 1 1 .2 ( * ) 2 and from this or separately A"(Z) : 1 (q .e.:=*£ +i(5l).386 CHAPTER 18..l)A(z) 1 = tfA"(z) A .e.
w = hz. [1(9 + 1)]! .w — > > ±iw) or functions R B w q( ) .18.5). i.(q + 3)(q + l)Aqvri(q '^ ' .. of course. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B.2 The Inverted Double Well Potential Thus a term /j. The solutions yq(z) of the equation Vq(w)yq(w) = 0.'2H9i) ^faDfr™) and .. _1J±w) and D_\. r ( CJw) = ^ . Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= .22) are parabolic cylinder functions D i .Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. = •>— r. (18.—77 [i(9l)]. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ . ^f( g + i)(±H2^ + 1 > \ — 1. when i = 0.22) is invariant under the combined substitutions q — —q.. give an equation from which A is determined. and the coefficient of terms with i = 0. (18. Rq '. 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. We return to Eq. those in Aq(z). In this way we obtain the next order contribution A^1' to A^°>. (18..e.23) . +1J±iw) (observe that Eq. (18. 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).
±4) S4(9. (q±2)(q±S). Tricomi [86].0). pp. and so Vqyq+Aj(w) = Ajyq+Aj(w). 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 .25) where in the case i = 2: SA{q.l)2/?4.Aj)(18. (18.3. Comparison with our notation is easier if this reference is used.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.0) = = = ±(q±3)(q±7).q + 4j] = c2SA{q.27) Now. Magnus. since Vqyq = 0.24) The extra factors in Eq. W.123. F. S2i(q. See Example 18. Actually these factors can also be extracted from W K B solutions for large values of q. we also have Vq+^yq+tj = 0. (18.^ For higher even powers of w we write i w2l V<i = Y. Vq+4j = Vq + 4j. Oberhettinger and F. (18. (18.q + 4j}yq+4j. .29) ' A . (18. Erdelyi. G.3)y g _ 4 .±8) 5 4 (g.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. As an alternative to Bq(w) in Eq. 115 . j=i (18.( < ? . l(q2 + l).23) one can choose the solutions as Bq(w) with B QM = ~ 77[(93)]!24(91) These satisfy the recurrence relation w2yq{w) = (q+l)yq+4 + qyq + .q}=A + c2S4(q. (18.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.388 CHAPTER 18.4j)yq+Aj. and for j ± 0 : [q.
27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. (18.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1.e. Equation (18. h2 —> h2. (18.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. We can now write the solution y(w) in the form y{w)=yq{w) + Y.) ty^^w) to be a solution to that order we must also have to that order [q.—v*+*j ( 18 . Thus the next order contribution to yq is y H = ie 2^ —z. ( 18 . .^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.10) we obtain E(q.18. p i(9»9 + 4 J'WK7». so that the entire expansion is invariant under the interchanges q .1. (18.[rp ) Y. (18.32) Evaluating this expansion and inserting the result for A into Eq. A = ^(q2 + l)c 2 + o(J^. h ) = qh . (18.—A(q + 1) .q]=0.31) Proceeding in this way we obtain the solution y = yW („. This type of invariance is a property of a very large class of eigenvalue problems.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. . In Case (3) the parameter q is only approximately an odd integer in view of tunneling. and even powers of q in combination with even powers of 1/h2.) + j.(!) («. i.e.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained. and c 2  replaced by — c 2 . i.3 °) For the sum y{w) = y(°)(«. 2 . n — 0.35 ) . .• q.
we may infer that given one solution y(z). / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz.390 where for instance v J CHAPTER 18.i 1 + 4j) in complete analogy to other applications of the method. > "p. i.g±4] ±4 ±4 [g. Again we can write down a recurrence relation for the coefficients Pi{Q. Achuthan.q±A) = MM!i±MM+[M±8][g±8.argz=0 (18.11) is invariant under a change of sign of z. ? T 4 ] [ g T 4 . g ± 4 ] T4 ±4 and so on.—. there is another solution y(—z). Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. We thus have the following pair of solutions VB(Z) VB(Z) = B w . ^VO(<L<?) ^i(?. MiillerKirsten and A. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t.36) with the boundary conditions p o(q. W. We observed earlier that these are obtained by making the replacements q —> —q. J.)]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. H.37) For further details concerning these coefficients. q) = = 1. their recurrence relations and the solutions of the latter we refer to the literature.e. q + 4j) = 0. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. (18. q + At] (18. o. Wiedemann [4] . w — ±iw." Since our starting equation (18. and for j ^ 0 all other P0(q. 2 4tPi{q.38) [yB(z)]argz=7T = [?/s(2. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1.
z2 .g. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. one has to stretch each by appropriate expansion to the limit of its domain of validity.2.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.q + 4j) as in ys. . h2 — — h2 as long as corrections » resulting from boundary conditions are ignored. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e.h^ih. (18. Vc(Z) = [V!B(Z)}q^~q.39) are with the same coefficients Pi(q. in fact. Thus in the transition region some become proportional. In order to be able to extract the proportionality factor between two solutions.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.40) "Variables like those we use here for expansion about the minimum of a potential (e. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.g. (18.h^ih 391 (18.4)).5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. h2) in which odd powers of q are associated with odd powers of h?.18. Aq(z)) does not appear in any of the higher order terms.g. see e. 4 . Integrating and expanding as follows since h? is assumed to be large. like w of Eqs. Mauss [192]. with normalization since the normalization constants are also asymptotic expansions. the solutions of type A being valid away from the minimum.The solutions yc.Vc suitable asymptotically decreasing expansions in one of the domains \z\ < O 7T argz We emphasize again that all three pairs of solutions are associated with the same expansion of the eigenvalue E(q.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 .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. Such contributions arise. J. 18.
392 CHAPTER 18. 2 ^i![i(g4il)]!(2«.44) In the solution y~g(z).0x ! [ £ ( g . y^(z) we see that in the direction of z = 0 (of course. we would have to substitute correspondingly the expression (18.z). We do not require this at present. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . Anharmonic Oscillator Potentials Considering the pair of solutions ju(.42) H W2 (ql)e$W I „ .) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.42) we obtain for the solution ys(z).(«.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18. with z in ys(z) replaced by — z.l ) ] ! ( .ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw. (18.45) . The function Di.43) W + 1)]! w^+V A . > n. eh* /12c* e\z*h? (18.. 4/T 4 Prom (18.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.43) (since z — — 2 implies > argz = ±7r).7 r . . 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.w yB(z) ~ B. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. (18.41) with the solution ys(z) of Eq.2 « . (18.4 i . Comparing the solution VA{Z) of Eq.
50) and y+(0) ^ 0.5.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.49) d yc{z) Again there is no such simple relation between yA(z) 18.21) as even and odd about z = 0. large probability for the particle to be found thereabouts.e.41). i.9.yc{z)pansion (18. Looking at the potential we are considering here — as depicted in Fig..y'_(0) ^ 0.1 ) e 12c 393 Kgl)]^"1) l + O h? (18. We proceed similarly with the solutions yc(z).18. the ratio of yA(z)iVB(z) i s n ° t a constant. Recalling the solutions y±{z) which we defined with Eq. hence we have to .2. 18. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18. (18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18. we see that in their common domain of validity yA(z) where a = =Vc(z). 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. At z = 0 the solutions of type A are invalid.42) into y~c(z) w e obtain Vc(z) = (h*.50) will be seen to imply qo = 2n + 1 = 1. The first of the conditions (18.4 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions The more difficult part of the problem is to recognize the boundary conditions we have to impose. (18.2 — we see that near the origin the potential behaves like that of the harmonic oscillator in fact — our largeh 2 solutions require this for large h2.46) However. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0.11.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .. and the second qo = 3.7.. (18.
We leave the detailed calculations to Example 18. Abramowitz and I.(0) (18.(«.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.oi(.! ( .i ( g + 3)]l (18.(0) = .' '.g. IT 4 dio In fact.i ) . A.. + 1)]!' D ' j . e.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18."*(«!) ^ ' ~ [ . y c ( 0 ) . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1. y'B(0).(0) = and htoi)(°) ^—'. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.394 CHAPTER 18.5 ( 9 + !!)(*"')• i ( 1) . Expressions for C 9 (0). Z (18./o^Tpif (q+i) .F ? . Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. 1 [7(93)]! n «(°) = . ( 9 + 3)]! s i n { . Stegun [1].56) D 4(. M. Example 18. C.s i n { .i)H = e.( g + 3)}.51) and 0 = V(0) = Jim i»xM .SxW) = \ i»fl(0) .55) Solution: Prom the literature.l ) ] ! [ . — C. [C.(iu)].1)('») + ^ r + l)]! ^.58) diu [  ( g .2.fc(0) Thus we obtain the equations (18.57) [i( 9 _l)] ! 2 i(9i) B.( g + 3)}..^{(9+1)}. we obtain i V7T24( g .54) dio q(w) . with the reflection formula (—z)\{z — 1)! = IT/ simrz. [ . (18..(0) + ^ ( 0 ) l z—>0 2. one finds that [ 1 IT (q + 1)1! = .l)(°) Thus with the help of the reflection formula cited above: D B. ..) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}.2: Evaluation of yB(0).
(18.3)}. (18.46). + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.+i)(°) [ _ I ( ._l) From this we derive C. (18.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.62) . 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) _^_ .53) in dominant order and insert the appropriate expressions for a and a from Eqs. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.18.56). 4 r T (18. Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .(0) = *>i(„+i)(0> 2 . W .i ( « + i ) [ _ i ( .)[i(g3)]! and ^i(. using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .i . we obtain 2 i(»+i)[_i(. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.6 ) } ^ — i t a n < — (q + 3) >. _ 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^.49).sin{^( 9 . isin{f(g + 3 .+1iH]«=o = .sin{^(g+ 1)} = J .1/2)!. ~ ^ 1 5(9+1) 2i(''.
( g 0 + 3)  + • • • ~ (g .e... Thus we have to determine these conditions first. In fact the left hand side of (18.2) with potential (18. ^ . (18. Anharmonic Oscillator Potentials Then the derivative relation of Eqs. 18. + 3)} = .64) the right hand side is an exponentially small quantity.go) ^ (1) It follows that we obtain for the even function with q — go = 1.^ >^_>m e J*.5. (18.. (gg 0 )^±^ y e5?. 7.9.65) 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) vanishes for q = go = 1. (18.. (18. . z^ ein/2z.. .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.1).L ^ i . 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. With a Taylor expansion about go the left hand side of (18.63) becomes (? . z2 > z2.64) In each of Eqs.. (18. we obtain cos((g + 3)) = . The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or.5.53) becomes 8 m(I(. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.396 CHAPTER 18. and the left hand side of (18. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. equivalently.63) and (18.^ .. Recall > the original Schrodinger equation (18._ e =*. the energy E is real.9o) ^ cos  . the vanishing of the wave functions at infinity)..64) about go — 3. 1 1 .9.63) Proceeding similarly with the second of relations (18.64) for q = g0 = 3.1. 7 . we again obtain (18..11. i. . 18.53). (18.65) Expanding similarly the left hand side of Eq.2.. . This is Case (1) of Fig.i . by the rotations E > einE = E..
e.66) z3 = \z\3em.e. 1/2.e. Now. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase.cos 36. 2 c2>0. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. i. i.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. the case of complex E). —}. in the domain — TT/2 < 36 < > 7r/2.68) . 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. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0.e. r / c 2x 1/2^3 = exp{ ± i — —\. or — < 6 < . the resulting wave functions vanish at infinity and thus are square integrable. replacing sin 30 by " 0 + ^) 1 IT = . i. y{z)~exv\i[) 7T dV'V. we set z w r = \z\eie. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. (18. i.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. if TT/3 < 6 < 0.67) Rotating z by vr/2.18.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. (18. 6 6 In the case of the inverted double well potential under consideration here (i.e.
This is the boundary condition also used by Bender and Wu [18]. we match the WKB solutions .e. Anharmonic Oscillator Potentials V(z) Fig. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq.398 CHAPTER 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.3 The inverted double well potential with turning points ZQ. for z > Too. Eq. 18. Our procedure now is to continue the even and odd solutions (18. We therefore explain our procedure first. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. 18.68) for c 2 > 0. We have to remember that we have various branches of the solutions y(z) in different domains of z. We do this extension with the help of WKB solutions.Z\.21)) were defined in terms of solutions of type A which have a wide domain of validity. Looking at Fig. (18.Z\. (18.3 we see that at a given energy E and to the right of z = 0 (which is the only region we consider for reasons of symmetry) there are two turning points ZQ. i. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0.68) 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.21) to + infinity and to demand that they satisfy the condition (18. Our even and odd solutions y±(z) (cf. and there impose the boundary condition (18.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).
291.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . ^ 2 4 •K + —(r. The distant turning point at z\ as indicated in Fig. 291. z\ ~ .zi)/ \ 1/4 ywKB\ ) z — g/i* .3 is given by (using Eq. B.4) for E and ignoring nondominant terms) t o . i. 18. Sec.2..e.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . Dingle [70].18. p. Messiah. .2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. We begin with the exponential factors occurring in Eqs. Vol. (22) or A. 4 2 T ' i. ^2? 1 2qc2 /2? . 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18.e. t To the right of the turning point at z\ these solutions match on to (r. f R .*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . h2 (18.4. (18. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. 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\. p. 6. Dingle [70].70). z2h4/A > c2z4/2.B.^ 4 1/2 + icV 7T Jz + (18. equations (21).71) We now come to the algebra of evaluating the integrals in the above solutions. I [195].70) where z < z\.4 h + c z 2 1/4 < . A J24 1 ? 2 • "^ •z h \—c z ~ — q h . *R. (18. 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.69) The W K B solutions have been discussed in Chapter 14.^ T rz\ 1 + yh* .
(18. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.400 i. Thus the above factor yields '2c2\l/2 h4 so that (cf. zi 2q 1 c 2 [1 .2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . Eq. Eq.40)). CHAPTER 18.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.^ I 2 2 1/2 zV = F 8c2 Jz [i . (18.2.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .\qh2 + \z2h± .^ i ^ ^ c 2 .40)) E± = exp _/^_2 f ± In A 2c2 .. .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.2 \ 1/2 .±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18.e. (18.73) .211/2 1 \ /j.2.69)) 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. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). (18. Anharmonic Oscillator Potentials E± = exp = exp .
we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) .*l) (18.20b).76b) apart from factors [ l + 0 ( l / / i 2 ) ] . In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.*l) (3y^(z) (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.74) Comparing these solutions now with the solutions (18.\z2h4 + J Z\ nl/2 \M fZ .77) Now in the domain 2 — oo we have > \qh2 .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.76a) P P r 2/1 2 " 9 ( .Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.(J*i) VA 0 ) = PVwKB (Z C.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.78) Inserting this into the solutions (18.21 1/2 .1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.71) and these into (18. h2 — — h2 is maintained.79) +S_(±)exp . (18.18.20a) and (18. > » Returning to the even and odd solutions defined by Eqs.
402 CHAPTER 18. (18. (18. (18.81) Inserting expressions (18.^ j .82b) This is our second condition along with Eq.^ § ( 4 g 2 + 29) + O ( ^ ) .68)..82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.85) + .. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.5.. h2) = \qh2 .^(q2 + 1) . (18. 18. i.e. Eq.3.34).6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.2. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J .h2) + (qqQ)h + (18. i./i2) = E{qQ.84) Expanding about q = qo we obtain £(g..h2) + {qqo)(^pj = E(q0. (18.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq./ ? = 0. . (18. (18. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .. E(q. i / 3 ± .e. (18.65). we see that we have to demand that 5 + ( ± ) = 0.76a) for (3 and /3.
36) of their Ref. zo} means "to the right of the turning point ZQ" .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .86) will be dealt with in Chapter 20. i. Since this is not without interest (e. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/.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. !*WKB(Z) ~ 2 . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.3. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. 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. where the superscript {r.l ) ] ! > + = ^ 2 .~p{q. above) (1. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).18.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.g.ZQ) . Example 18.e. h2). o 2 + l)^(4.yA{z). ft6 ^ = e. h2) of the relations KWKB (Z) =PVA(Z).4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .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 .2 . S W K B ( 2 ) = PVA(z).h6/6c2 E = E(q0.g ( .
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. the (approximate) solutions* 1/2 exp • ± i ( . Anharmonic Oscillator Potentials where zo > z. 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.zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . 2 z 2 ) / 2 e \ . cf.ic224 2 4 qh2 + i^fe 4 ../ ^ 2 X 1 / 2 exp(ifi.O.404 CHAPTER 18. N. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20." / 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. f x Zn 4) h2 2.g In other words.1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation .] { hz combined with the particular constants contained in y ^ ' ^ g . On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20. Schwind [247].zo)/ \ qh2 2 ZO + z2hA 4 2 . and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . . 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.
J ' i J . ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.89) .^ 1 .zo). J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . (18.e.23)) appear quite naturally. in fact.18.1)! = ( 2 7 r ) 1 / 2 e .3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. h4 > 0.5 "l _ o f .kyil(.g. as in the case of the Mathieu equation. (18. in Eq.41) for yA(Z). 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.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(..(r._ .yA{z) with expressions 27WKB i J'WKB ' that the factors p.1 The Double Well Potential Defining t h e p r o b l e m In dealing with the case of the symmetric double well potential. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0.DfJ»v we see Comparing now the expressions (18. Then . 4" ) (h?z2)^q+1) for q j£ odd integers.3. But there are significant differences.88) + [EV(z))y{z)=0 (18. (j . (18.z ± ) = .3 18." / 4 ( ? « / 4 . we shall employ basically the same procedure as above or. 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.13) of Bender and Wu [19]. „ / M 1 / 2 i . We can reexpress these relations with the help of Stirling's formula. i.
z±) y = 0.4 The double well potential.z±). Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.95a) . 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.i>5. (18.91) becomes T>q±(w±)y(w±) = o(^)y.z Fig.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. vW(z±) = ±Qch2.406 CHAPTER 18.90) V(i)(^)=0.91) (18.V(z±) l{z z±fhA + 0[(z . the previous equation (18. 18.93) w± = h±(z . (18.94) (18. . (18.
we conclude that in the dominant approximation q± is an odd integer. we obtain another by replacing z by —z. Inserting the expression (18..i 4 f W » = o. (18.95a) and (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z).96) dz2 + 5 « 4 +  .3. Eq. (18. (18. (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.99) 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) = 0.98) 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. is again seen to be invariant under a change of sign of z. qo = 2n + l. given one solution.e. 18. (18. n = 0. d2u{z dz2 + v+ 2~r u(z) = 0.1.2.96)..87)..100) . (18..97b) Our basic equation. we obtain d*y (18.93) for E into Eq.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 (18.97a) U(z) = (zz±)2 + 0[(zz±)3}.95b) By comparison of Eqs.18. Thus again. i. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].
(18. these equations can be rewritten as + (qz+z)A{z) = . Vq = {z2+z2)— + (qz+z). ~h2.101b) To a first approximation for large h2 we can neglect the right hand sides.h2.^q(l7q2 6 4/i + 19) + (18.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. z) may be obtained from the solution yA{q. z) = yA{~q. We leave the calculation to Example 18.yA(z) are associated with one and the same expansion for A and hence E. Aq(z) = Aq(—z) = Aq(z). z) = yA{q.101b).101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. i. i.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .103) Looking at Eqs. h2.408 Since CHAPTER 18. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —. (18.c W + 1) . z).4.e. Integration of Eq. n± = V2h2. but the solution is a different one. (18. h2.104) Both solutions yA{z). 8hw (18.1 6 ) .h2) .^ A"{z) + 2c V2 *A(z) (18.105) "8 + 4 ^ . The result is given by A and E(q.e.101a). h2 h4 2c' with g+ = q. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. (18. (18. yA(q.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). we observe that the solution y^iq.106) . h2.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18.
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.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.{z2 .109) We observe some properties of the function Aq{z) given by Eq. J = 0. ± 2 . (18. Thus it is natural to explore analogous steps.113) 2 . .qz+)2 .103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. The first such step would be to reexpress the right hand side of Eq.103): . First.e.18.2 z\) • Aq$4 £Aq + Aq—4.jf2i(2) We also note at this stage the derivative of the entire solution yA(z) taking into account only the dominant contribution: yA(*) =* A*2 V2 . . (18. i.z2 ) N t.qz+) [2z2 . .102). (18. + Aq+2 + Aq Aq+2 . = Aq+2i+2jAq.106). (18.3 The Double Well Potential 409 Example 18.112) From these we obtain.z\) + 2z(z .(l + + 2  z y* .108) We wish to rewrite this expression as a sum y coefficientiv4. <4q+2 — z — z+ Aq+2iAq+2j (18. ± 1 .110) Aq+2+Aq2 9+2 A. . z • . Solution: The structure of the solution (18..Aq 1 + 2^± 2 Aq+2 (^M^f (18. .101a) with A replaced by A . as a linear combination of terms Aq+2i. A Differentiation yields » = ^^w (18. (18. 92 ZZJf (18.111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . for instance by componendo z z et dividendo. . we reexpress A'q in terms of functions of z multiplied by Aq.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .4zz+q+ z\{q2 + 1)]. however.105) and (18. (18.+ 2 ^ ± l + 2^±i + . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.114a) . We know the first derivative of Aq from Eq. such as periodic potentials.
+6 + • • • ] .108). cf. (q.q + 2)Aq+2 (18. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.q + 2) = 4{q + l ) 2 . we obtain 5 K = ( ^ f ) [ ( 9 . leaves uncompensated on the right hand side of Eq.114b) .l ) ( g .101a) the contribution R . Since VqAq = 0. .z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.2Aq + Aq4] 1.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 . (17. In particular the dominant approximation of A is obtained by setting (q.4Aq+2 + Aq+A Inserting (18. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.34).112) CHAPTER 18. Anharmonic Oscillator Potentials A 1 . 1_4^±^+8^±±12^±^ +. (q.4 ( g . _ 4 . QA?+4 _ 10 A + 6 . Eqs. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.q2)Aq2 + (q.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.410 and from this = Hence with Eq. A(°~) = Aq.e. (18. (18.q)Aq + (q. (18. (18.3 ) A . •••]. q + 4)Aq+4 + (q.<? +{q. (18.4]j.116) into Eq.116) [Ag4 .4)A„_ 4 + (q. q) = 0.112). where the lowest coefficients have been determined above as (9. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.aAg+S 12+ 16 (18.115) and (18. i.(0) V2 ' 2c K + h4+Ao .119) It is now clear how the calculation of higher order contributions proceeds in our standard way.q + 6)A<.4 4+* + .117) 4(q + l)2Aq+2 Here the first approximation of A. (18.2\2 (z* .4A ? _2 + 6Aq .l ) 2 A .118) 2s/2c ~^j[(?. .33) and (17.2A g + A g _ 4 ] .115) (z*z%) 2 i2 (z 2 .9=F4) = (g=Fl)(?=F3).
y'+(q. y'_(q. 9 + 2) Aq+2Az+ (9. i. (g.123) ^[yA(q. A' (0) = q/z\) y+(q.0) = ^ .2 ) (0) (18.0)=0. h2 2y^ Age? h4 (18.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.4 ) ( 0 ) ( g . \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.(0).z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.h2.h2. of course.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. where Rq is the sum of terms left uncompensated by AW. z)\ = 2^A^ h2 '>z)± VA(Q.124) .z)±yA(q. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.h2.g) 8z+ (g.h2.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.0)=0.h2. g . g + 4)(g + 4.94)(g4.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.94) 8z+ (9.g ( 1 7 g 2 + 19). 9 + 4) ^+48z+ (18. h2.g + 2)(g + 2. h2.105).g2)(g2. g) 8z 4 .0) = y^(q. for i = 0. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .101a) the contribution Rq .28.e. *)] (18.g) 4z+ {q q)+ 2r (9. thus yielding the next approximation of A as given in Eq. (i) _ / V2c\ [ ( g .g) Az.122) The solutions yA(z). in the domains away from the minima. (18.yA(z) derived above are valid around z — 0. i. Terms jj.{^ h2.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9. (18. g . (18.e.
2(9+) . yB(z). We see already from Eqs. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). (18. (18.91). in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum.125b) It is clear that correspondingly we have solutions yc(z).Evaluating the exponential factor contained in VA{Z) of Eq. Inserting (18. we have w±(—z±) = —2h±z±. the equation is — with differential operator T>q as defined by Eq. Anharmonic Oscillator Potentials Our second pair of solutions. (18.2A y(w±). Bq[w±{ ^ .95a) and (18. (18.93) and setting w± = h±(z — z±).412 CHAPTER 18.99) for z — z±. Moreover. Since w±(—z) = —h±(z + z±).1 ) D (18. and another VB(Z) ± 25^ch3wl + c2wi .125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. This means. (18.23) with appropriate change of parameters to those of the present case.We draw attention to two additional points. but w±(z±) = 0.95b) that the solution there is of parabolic cylinder type.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z). (18.126b) These are solutions again around a minimum and with yB(z).yB(z) is obtained around a minimum of the potential. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. 18. we > .95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). + 0(hl2).^ % [i(?l)]!24(«.3.yc(z) with complex variables and Cq(w) given by Eq.
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. (18. ein+z+e in+(z 22+(<?+i) . 24(9D[i(gi)]! hj (18.ZI and find that these are given by z+ + 0(l/h) for finite q (cf. Eqs. . 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 .^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. Allowing z to approach z+ in the solution IJA(Z). (18. (18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.147)).i*l(«^ 4'"+^+ 4"'+v ( « .. .18.172)) we add here comments on this approximation.97b) ~ exp .z±)dz .128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).• • " .127) Recalling that around arg w± ~ 0.146).164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.frW. . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. .172) obtained later. 2 An+z+e 4w+ (18.« + ) 2 (9+1) e (18. Thus for h very large these turning points are very close to the minimum at z+. (18.129) . Later we calculate the coordinates of turning points ZQ. (18.3 The Double Well Potential (cf. a= ^ .^/w + ± 1 . (18. . and comparing with Eq. Here z± = ±h?/2c are the positions of the minima of the potential.
We have to impose boundary conditions at the minima and at the origin. D. S. but it is clear that none of the above solutions can be used at the top of the central barrier.(18. R. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . K. H.3. The solutions in terms of parabolic cylinder functions have a wide range of validity. g .131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. Rau [22]. S. and this means at both minima.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. J.4 B o u n d a r y conditions at t h e m i n i m a (A) Formulation of the boundary conditions The present case of the double well potential differs from that of the simple harmonic oscillator potential in having two minima instead of one. basically. Furry [102] . Bhattacharya and A. The next aspect to be considered is that of boundary conditions.7r and the solutions of type C around argz = ±7r/2. Thus 1 E . 18. Various investigations^ therefore struggle to overcome this to a good approximation. P. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. we naturally expect the wave function there to be similar to that of the harmonic oscillator. Bhattacharya [23]. de Deus [66] and W. 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.z+)]fr+V ' (18. The pair of solutions of type B is defined around argz = 0. even above the turning points. 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.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . K. since. Since it is more probable to find a particle in the region of a minimum than elsewhere. The involvement of these WKB solutions leads to problems. they assume large quantum numbers.414 and CHAPTER 18.
5.Zii yB{z±) a ± a yciz±) (18.5.136) . The odd wave function then exhibits a correspondingly opposite behaviour. an even wave function can also pass through zero at these points. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. ^(*±)#0 (18.5 Behaviour of fundamental wave functions. and y+(z±) = 0. as indicated in Fig.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. y'_(z±) = 0.18. (18. as indicated in Fig.135) Hence the conditions (18. 18.132). as indicated there. However. y(z±)=0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.133) y+(z±)^0. (18. y{z±) £ 0. We have y±(z±) = ^VA{z)±yA{z)]z^.134) y'+{z±) ± 0.132) Fig. 18. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. 18.
(18.136) as  M . +1)} = 4 m.141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i. also (18. .„*{!<. Thus sin {^ +1) } = ^^>^30fe^i.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. 1 1 . . . 7 .134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 . Eqs.5.2 we can rewrite the second of (18.. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.. .136) the dominant approximations we obtain (cf.142) (<Z9o)J(l)^°+1)(l)1/2 .416 CHAPTER 18. (18. (18. (1. (18.m Using Eq.9. _ 8)} s .«*{=(.138) Using formulae derived in Example 18..
y'_(0)^0.» ^ B ^ ' .3.z\. Hence we have to impose at z — 0 the conditions j/_(0) = 0. V(z) — • z E=V Fig.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.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. 18. (18. Since the type A solutions are not valid at the minima.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.3 The Double Well Potential Thus altogether we obtain 417 ^ • . We emphasize: We needed the solutions of type A for the definition of even and odd solutions.18. we must also demand this behaviour here along with a nonvanishing Wronskian. y+(0)^0. y'+(0) = 0. We do this with the help of WKB solutions.6 Turning points z$.1 " <813 1 4)  In Example 18. 18. We deduce . 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.
the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .e.148) ("f * "The superscript (I.144).24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. In the domain 0 < z < ZQ. i.e. Anharmonic Oscillator Potentials from Eq.^1 25c2 + V2 2 Thus again we see that for large values of h2 the turning points are very close to the minima of the potential for nonasymptotically large values of q.145) \qh\ + \zW Using z+ = h2 /2c. one finds that . h8 25c2 (18. ZQ) means "to the left of 20" .418 CHAPTER 18. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. We also note that the height of the potential at the turning points is h8 qh V(z) 20.147) Since the minima are at z± — h2 /2c with /i 2 very large. 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.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. + £\h%U(z)=Q.\qh\ + \h%U{z) (18.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. As a consequence. to the left of ZQ where V > E. (18.
150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .18.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.148) and (18.^ e therefore have to consider the exponential factors occurring in (18.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.^ + ) 2 + igln2(zz+).i (2q ln (18. 20 dz \qh\ + \h%U{z) .^ j O e .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+. (18. )+^*+)2lr} rnD h\z'2)}z Q )\ = .y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . we have to match yA(z). Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.* 4+ .(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .149) In order to be able to extend the even and odd solutions to z = 0. (18.S2 ) dz rzo 1/2 'Ihl + ^Uiz) I 1/2 / Jz dz (z~z+)  w + 1/2 1 = ±lh+ 1/2 ZQZ+ Z — Z+ 2 i^$) 1/2 ~4 >'^. 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) . .
154) .420 CHAPTER 18.125a) and (18. l/2 ~h\U{z \(zz+?h\ (18. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.126a). t h a t the Stirling approximation is amazingly good even for small values of the argument). Stirling's formula is the dominant term of the asymptotic expansion of a factorial or gamma function and thus assumes the argument (oc q ~ 2n + 1) to be large (it is known. using Stirling's formula (and not the inversion relation) — we have —(1. Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. This is the aspect investigated by Furry [102]. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. We see therefore. the results necessarily require adjustment or normalization there in the q—dependence. In a corresponding manner — i.ZQ) . Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18.152b) [\(q 1)!] V2 K7: hi Here q/A was assumed to be large but we write [\{q — 1)]! since this is the factor appearing in the solution (18.e. 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 (18.125a). of course. However.
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 (18.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. this result is somewhat imprecise.103) and (18. However. Comparing for z —• z+ the W K B solutions (18.99). it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.e. \hl.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.153). (18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4. 27T1/2 9/4 1VA{Z).159) 7 i(«i) r(9^3) ^2~ j(9l) . we obtain in leading order (i.143) T^^*)) 27T (18.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. .18.154) with the typeA solutions (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large. (18. 7 of t h e matching relations y^NKB\z) i. The relation (18.e. multiplied by (1 + 0(\/h\))) the proportionality constants 7 . and is shown to reproduce correctly the result (18.104).144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.e. (18.K?+1). Returning to the even and odd solutions (18.157) is used in Example 18. .155) (2z. linearly matched) W K B solutions.157) Since the factorials [^(q — 1)]!.5 for the calculation of the tunneling deviation q — go by using the usual (i. as the results also seem to support.158) Applying the boundary conditions (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18. (18.
160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. Friedman [40]. n + uY 2c „ ^6 8V2c2 (18. p. Byrd and M.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.19.161) we can rewrite the integral as r I2{z) = = dz^{a2 .163b) The integral hi?) . G2 K [i 4. F. 3G2 12V2c ^See P. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. 213. .3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18.05.163a) (18.. (18. (18..149) near z = 0.422 CHAPTER 18.148).. D.164) / 2 (0) z ca [(1 + k2)E(k) . formulae 220. We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^2 —pz I V2 dz ^?^ h+ Z 4? + ^f ^2<«2 (18.z2)(b2  z2).iV2 ! i .k'/2. .162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. 60 and 361. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18. p.
( ft 7.r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . where the result is shown to be (with q ~ q0 = 2 n + l .1 " e " O T (18 169)  Y^V / 4 tfo.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . n. /2n + l 2 ± — . ^ ^ .1)]' Gfo.± ' 3G 2 ' T ( ^ (18. [167].. 1.. .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .2.165) in agreement with Ref.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. .. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 .3 )l ! . n = 0.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. It then follows that (again in each case in leading order) '2_hl)9/4 ^ ^ * > l * = ° ^ 23g/S(_g:)t/4(!)g/4eg/4 "1^Using Stirling's formula we can write this e ( 18 . [4] misses an n—dependent power of 2 in the result.1.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( . ± . which implies small G2.18. The nontrivial expansions are derived in Example 5. (G\ . . ..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 \.
h2) + (qq0)^=. (18... (i*™) We obtain the energy E(q.174) .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. (18. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. .173).3 concerning the exponentials. 2 . . with the help of Eq..i ) ] ! e ~ s A ' .3. Inserting the expressions for h+ and z+ in terms of h. we obtain for E(q0. we see that the relation is really an identity (the preexponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton). (18. h2) the split expressions 29o+1/i2(—W2 h6 (18. (18. 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.. 18.175) ~E0(q0.n = 0.157).143) with qo — 2n + l.^ W ^ / 4 t o . Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.1. 18.»W.106).172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z. (18.143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.h2) and hence the splitting of asymptotically degenerate energy levels.424 CHAPTER 18. (18.3. and obtain ( * . . we can therefore impose the boundary conditions at z = 0 by making in Eq.6 Eigenvalues and level splitting We now insert the replacement (18. (18.
176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.178) "In S. i. the following expression for large q ~ 2n + 1. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.fr2) is given by Eq. which evidently supplies some correction terms.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .. The result (18.e. . Combining Eqs. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i.157).1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.h2) = E_(q0. 1 7 425 2 . the level splitting..e.e. i. i n U n / ! Thus AE(q0.^ 2^ h* . the pure WKB result of Banerjee and Bhatnagar [14] (i..V2^\ \n+ i .18. K. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 . 1 u2 c 2 (3 9 2 + l) ^ « b .e.2 = 2 .3 The Double Well Potential where £0(<7o.n = 0.106).q2 7.1.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. 2h4' i.. (18.h2)E+(qQ.2.7 =( e N n+1 .143) with (18. T \ *n ! 1 ' (18. for finite h2. 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 / » : = \ .174) and (18.e. (18.
h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.3)). K. This. Furry [102].181) "W. h = 1. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. (18. ** Comparing the present work with that of S.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). We see therefore. (18. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. 2/z4 and 2c2 respectively (see Eq.. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0.180) h8 E0(qo. as is also explained by Bhattacharya [23]. h4 and c2 replaced by 2E.h2 <> — h2.6 \9o/2 e~h&l^\ with (m 0 = l) (18. Eq. however.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». A>0. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.176) together with Eq. h2)\mo=i = . Then the splitting A ^ W B of Eq. 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. as we do here. 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. the Furry factor corrected WKB result follows automatically. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. (66). H. that if this symmetry is taken into account from the very beginning. (18.4/4 here.2 / 7.2X2 V(z) = \z2tL) . supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. we would have obtained the result with E.426 CHAPTER 18.^ 2 + ^h2 If in addition the potential is written in the form A / . (18." Of course.178).6) to yield an improvement of WKB results for small values of n. . and A there with c 2 / 2 here. we identity the parameter k there with ft. Bhattacharya [23].
The comparison with Eq.11). D. Friedberg and T. D. Lee [98]. J. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.^ *See E.§ 18.W. first paper. these are presumably not of much interest in physics (for reasons explained in Chapter 20). (2. Mansour and H. H. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions.* A similar correspondence is also found in the case of the cosine potential. the level splitting is (with h = 1) This result agrees with the ground state (go = 1) result of Gildener and Patrascioiu [110] using the path integral method for the evaluation of pseudoparticle (instanton) contributions. J.c2 = A/2. See R. J. Jentschura [293]. 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.^ Thus for the case of the ground state Eq.h>) * 2V2^^^y/2e^3/3A. . MiillerKirsten [186]. Eqs. however.3. so that by comparison with Eq. Liang and H. Jentschura [293]. ZinnJustin and U. M. We considered above only the symmetric twominima potential. Gildener and A. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. §J.181) therefore implies the correspondence rj1 <> A/2. Patrasciociu [110]. Equation (18. (18. P. m 2 <• p 2 / 2 . formula (4. ZinnJustin and U.88) h4 = 2/x2. Bhatnagar [14].18. W.182) implies AlE(l. D. Banerjee and S. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. M.15).3 The Double Well Potential 427 a form frequently used in field theoretic applications. (18. ^To help the comparison note that in J.—Q.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. K. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.34) and (E. (18. and g2 is given as g2 = r]2/m3.
[131]. Mahapatra. In fact. mostly in connection with instantons. can be found in an article by Coleman [54].g. in both symmetric and asymmetric form. Pradhan [182]. like those for anharmonic oscillators. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Very illuminating discussions of double wells and periodic potentials. B. [19] — that the expansions considered above are asymptotic. Simon and Wightman [178]. Perturbation theoretical aspects are "See e. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. as he discusses. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. 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 references in this direction. we cite papers of the Uppsala group of Froman and Froman [100]. though not exclusively numerical. B. 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. has also been the subject of numerous numerical studies. Ashbaugh and Harrell [12] and Harrell [130]. The double well potential. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. . as some relevant references with their view we cite papers of Harrell and Simon [129]. There is no reason to view the anharmonic oscillator differently. Santi and N.428 CHAPTER 18. Martin. [132] and Loeffel. for instance. P. 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. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. 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]. A reasonable. N. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle.
186) (18.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.187) We now reexpress various quantities in terms of u.f + • • •) 2u\ . Gutschick. X +2 + ' ^ (18. q ~ 2n + 1. We obtain the expansions from Ref./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. fc ~ 1 . Datta.183) S o l u t i o n : We have k2=1^.2G^2q + 4G2q . 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. 2 a. Biswas.189) =2u 2u2 (18. Saxena.u)1'2 = V2u{ 1 . which is assumed to be small. Bender.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14].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 . Example 18.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . Cooper and Strottman [221]. Srivastava and Varma [24] and Bhatacharya and Rau [23].185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. T h e r e f o r e it is convenient t o deal w i t h this here. 1+ u : Gy/2q. k! (18..G2q + .^ + • (18. Thus k' and hence k' = v ^ ( l . We have 4 ~k' 2u(l . (18.192) 'F 12 16 . [122] as 13 1. (18.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18.\G\ + • and 1k2 =2Gv/2q4.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .
3).13)} + 12u 3 (u .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .u ) (^){« + 5^ 1 )} + 5..197) .u) + 2u} + i u 2 [ 2 + (1 .^ L ) H[(i _ U ){4 + 3u(l .u){24 .1)(1 .430 and CHAPTER 18.u ( l .3 u ( l . 1 3 « 2 = .195) we obtain E(k) 1 + ln uK(k) .u){4 + 3u(l .3 y .«) + 2} + V [ .3u(6u .a = (l«)(jl 2u/ 2 ) .3u(6« .18u 2 + 39M} + 12u 3 (u .u) + 2} + i u ( u .— ) .194) ln '^J + 2 In i/2u )+H l + 2u(l .« 2 + 3u] In [ .« ) ^ { 2 4 . 8 + 9u 2 .2{«(1 . 8 + 6 u 2 .u 3 ( u .4.3) .u)} u{u(l .u) 2 {24 .193) 2u(l .u) 2 {8 .u 13 \ 1 ln(± k> fc'2 +  (18.13)} + ±u3(u .3 .u ( l . (18.3) .3 « . (18.e.{ M 2 ( 1 . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .u){24 .3) = = ~ .u)} + — u ( u .3u(6u .196) Now consider the last line here without the factor 2.13)} + ~u3(u .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .3) 1 + In ^ . .u)} . ) J 4 (18.u) + 2}] 1 H 1 o u(u .1)(1 . 8 .u)(u .6w2 + 13u} + 12u 3 (« .1)(1 .3u(6u .194) and (18. (18.{l u ( l u) / W 2 + 2} .6 in the denominator and in the last step pick out the lowest order terms in u (i. 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +. up to and including u2): u(u .1)(1 .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .13)}.u){24 .195) With (18.3 u .
173) t o be evaluated.e.^HfHGH Example 18....148) (for particle mass 1/2 and in dominant order) is given by H.u H I .2 l_ln(1 JL\ V {2.it)} . i. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .4u .196) and (18.128) and (18.2 /o 2 431 (18.156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf. . Inserting here from the above expansions the contributions up to and including those of order u2. (18. l2 = ^ ( l + uf/2[E(k)uK(k)}.155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).3 The Double Well Potential Now consider the bracket [. ' 8G 2 q = 4 2 q ln f 26/2 \ o .«){4 + 3u(l . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .] in (18.2{w(l u) + 2}] = u(3u2 .196). (18.4u .J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs. / to + 3GH~ \ 8 .zo) HWKB normalized _ ( h* y^eM. i. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .3).200) Remembering that u = Gy/2q. From (18. (18. where w±(z) — h±(z — z±).2 (18. + iu^Zu2)^~u+ — +0(u3) l .199) We now return t o the expression on the left of Eq.6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18.18. we have. Eqs.uK(fe) ~ 1 + In I — = Mu2(3u.197) we now obtain 4 1 • E(fc) .128) and (18. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V . [(1 .198) 9 «.e. (18.195) with (18.3) . 2 . V^LJ 4 2 16 ^ .
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 [ .hA/2 = h^/4 this implies . (18. (26. (18.Aw) = s/2i 2("!) V (41) and [(9 . Eq.e. (18.** oo / dwD\./ .432 CHAPTER 18.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.35)). are given by Eqs. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf.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+.f hi\ VWKB. 93. pp. and obtained as Eq.158).143). by using the periodic WKB solutions below the turning points. Solution: The turning points at zo and z\ on either side of the minimum at 24.e. 1 7(l.146) and (18. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.147). i. Oberhettinger [181]. 82. We start from Eq. i. oo .l)]![i(0 . (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 .zo). y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. 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. normalized = We see that the normalization constant is independent of a quantum number. Then ((r.*o)/ ^WWKBOO (18. ° \Edzy/\EV(z) V(z)V4 VWKB. W.g.3)]!23<«D [5(91)]' >A.204) . Magnus and F. Example 18.
.207) (18.e.3.3 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 .205) We now apply the boundary conditions (18..18.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. Sec..V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .g.212) tt See e. this implies sin cos sin cos Thus e. i. from z\. (18.^r + + .e. Then at any point z € (zo. z{) we expect tt (18. Messiah [195]..211) 7 z l . We could.2. i.iK h\U{z) + / 4 (18. \h\U{z) 1/2 ..i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. . 7T\ + ^ 4 7r\ dz .209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.6. approach the minimum also from the right.( .r + ^=co S [r.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.2.e.210) Choosing the point z to be z+.g. from ZQ.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = . JV = 1. 6..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 (18.132) and (18.J^Wj +(18. A. i. 1/2 . of course.
P. K. W "Cf.146)). Bhatnagar [14]. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). 1 1 . (18.215) which results from the factor of 2 in front of the sine in the WKB formula (18. 2IO/K. (18. 7r"y in agreement with Eq.213) The integral on the left can be approximated by 1._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . Landau and E. M.215). . they become c o t { ( . (18.210) assume a form as in our perturbation theory. in fact (with our use of qo).Zi / r. and cot< (q + 1) '!}» for q = 50 = 1.217) 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. .27 9 . Lifshitz [156].149). i. 5. .3.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 .90 — =F— for q0 = 1.216) Inserting this into Eq. In Eq. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18.5. (18. and L. (18.2 . . + l ) j } ^ 1_ 27' (18. 9. the level splitting is given by AE(q0. ' /^i . One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).143).434 CHAPTER 18. (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18.e. The prefactor 2\/2h2/n is.. We now see that Eqs. Banerjee and S. 7 .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. where u> is the oscillator frequency of the wells.215) we can expand the left hand sides about these points and thus obtain . The formula thus agrees with that in the literature.214) {«+"i} 0 for q = go = 3. S. K. (18. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18.209) and (18.205).. D.148) and (18.215) we can insert for 7 / 7 the expression given by Eqs. Bhattacharya [23].
Chapter 19 Singular Potentials 19. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. 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. W. f K . W. J. Morse and H.* However. 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. Aly. H. Giittinger and H. However. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. are not as troublesome as one might expect. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. Case [45]. Feshbach [198].1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state. M. 'For a review from this perspective with numerous references see H. MiillerKirsten [9]. 435 .* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. 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. n > 2. also in P. in an early investigation Case^ showed that potentials of the form r~~n.
D. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. Hashimoto [124]. S. K. S. Singular Potentials theories in mind. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. Tamaryan. Liang and Yunbo Zhang [189]. that with the hyperbolic cosine replacing the trigonometric cosine). S. It is therefore natural that we study this case here in detail. H.2 19. In the concluding section we refer to additional related literature and applications. § 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). Gubser and A. J. some also permitting further research.e.2) § C . This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. H. Park.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. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. Recently. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. Manvelyan. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. W. R. Singular potentials arise in various other contexts. Maldacena [41]. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. N. W. MiillerKirsten. 19. J. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. for instance. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. In fact.2. G. . 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. D referring to Dirichlet boundary conditions) in 10dimensional string theory.Q. J. MiillerKirsten and Jianzu Zhang [226]. Callan and J.436 CHAPTER 19. (19.
7 that we employed also in previous chapters. Manvelyan. The radial Schrodinger equation then assumes the form 2h2 cosh 2z .mo = 1/2 and h = c = 1. using the method of Dingle and Miiller of Sec. and as we have in mind in Sec. (19. It may be noticed that since I is an integer.3 below) g2 is negative and hence h? is real for E > 0. The following substitutions are advantageous: y = r1/2cf>. In the following we study the solutions of Eq. (19. r = 7 e z . Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. 8. (19. ig h \j g (19. h = ei^y/k^. H. MiillerKirsten.Q.4) first for small values of h2 and then for large values of h2.28) below).19. J. and so =4 5 and n ez = r = ~ = e^4fir. It is clear that we draw analogies to our earlier considerations of the periodic 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.3a) In the case of the attractive potential (as in the string theory context referred to above. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. .( I + 0. 19. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. (19. W. \[\ in the correspondence is nonintegral. Eq. This is rarely possible and therefore this case deserves particular attention. R. Liang and Yunbo Zhang [189]. h2 = ikg. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. This has advantages compared with higher dimensional cases.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. We observe that with the replacements p2 z^iz.3b) 7 = % j.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e.2 The Potential 1/r4 — Case of Small h2 437 where E = k2.g. J.
.5) so that Eq.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + . (19.2.6) terms amounting to 1 „ 2 d2Zu A V (19. Proceeding along the lines of the perturbation method of Sec.438 19. and H„ (w). Aly. i.e. —Zu — (Z„_i + Zu+i).^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+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. W.11) We follow here largely H. 8. H. w 2 .2 CHAPTER 19.1. MullerKirsten and N. The solutions Zv(w) are written Jv(w).7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. (19.7. First we make the additional substitution w = 2/icosh z.I . (19.2 A / i 2 .6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. J.i l . where £>„ : = —_ + i _ + J l . Eq. 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.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. (19. VahediFaridi [10].Hv (w). The zerothorder approximation = Zv[w) (19. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. where these are the Bessel function of the first kind. (19.Nv(w). dwA w dw { wA ) (lg. — Zv+i). (19. H.9) leaves uncompensated on the right hand side of Eq..
and this means in (19.u (19. (i/.v + 2)*{u + 2.u + 2)*Z„ +2 ]. where We observe that DUZV = Q. (19. (19.!/+ 2)*4°J 2 ]. The terms (19.13) (19.2)*Zi.i/) = 2A.13) therefore lead to the first order contribution jfU = h2[{v.19.13)._2 + {v. v . ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a. I / ± 2 ) = 1.2 ) * i ^ 2 + (!/. wz The expression (19.10) in terms of functions G„ defined by Gv+a = —nZ v+a .20) 2)*ZV_2 + {y. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.21) +(i/. ..15) (19.10) is now particularly simple: i?i°) = h2[(v. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. . = §. of course. (19.i/4)*Z 1 / _ 4 +(u.17) Thus a term fiGu+a on the right hand side of Eq.6). v + 2)Gv+2]. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.2 ) * ( z / .10) as a linear combination of various Zv. v2)*{y2.2. it is more convenient to use these relations in order to rewrite Eq. v)Gv + (y. However.18) Now </>(°) = Zv left uncompensated Rv .16) ( I / . where the starred coefficients are defined by (19. v . therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . z .14) (19.u + 4fZu+A}.2)G„_ 2 + (y. when a or 2v + a = 0. v + 2)*(v + 2. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). v + 2)*Z„ +2 (19.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.12) (19.
9) 16 + + ^ j " liy '2° This expansion is seen t o b e familiar from t h e theory of periodic Mathieu functions where (Z + l / 2 ) 2 represents the eigenvalue. (19. etc. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. i/ ± 2)* (i/ ± 2.i/±2)*. v ( 4 5 a .2.. i/ ± 2)*.2)(v + 2j .v)] (19.17)).25)/i 8 32(al)3(a4) . Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . . . i/ + 2j).I/±4)*. in Eq.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . P2*o(0) = 0.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . .v2)*{v2. Ri.v + 2y{v + 2.v) + {v.1169)/i 12 . Evaluating t h e first few terms. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v .v) + hA[{v.2 + 2 9 ) ^ 64(^ 2 . v) +(i/. . .j^0 (19.4)(»/2 .455a + 1291a . . (19.l)(z/ 2 . i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.440 CHAPTER 19.24) P2i2(2j + 2){v + 2j + 2.4 + 58. p4(±2) = (i/. . j=i. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .I/±2)*(I/±2. v + 2j) (19. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. subject to the boundary conditions P2i(2j) = 0 for  j  > i.26) iy — 2. v + 2)*{v + 2. v — 2)*(v — 2.v + 2)*(i/ + 2. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. i/)] + • • • . and p 0 ( 2 j ^ 0) = 0. we obtain 0 = h2{v.g. (I/. Adding these terms and setting the coefficient of Gv equal t o zero. (19. p 0 (0) = 1. we obtain 9 h4 = a 2(al) 3 2 (13a . we obtain t h e expansion .22) where p2(±2) p4(±4) = = (i/.
(19. we can rewrite the latter as j ~ .2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.^ + e±(v~2)z. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. v)(f>v + {v.32) (19. and is therefore real for both cases g2 > 0 and g2 < 0. Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.34) .33) = = cosh(^ + 2)z + cosh(^ — 2)z. £>„ = — .h2[(j)u+2 + 4>v_2] h2 [{u. 19.v + 2)<t>v+2]. Substituting (19. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17.v2(j> = 2/i 2 (A . = 0.30) £>„&. . (19. 2 .2.19. The solutions 4> of the modified Mathieu equation are now completely determined.cosh 2z)4>. Thus v is real for small values of  h21.7) into Eq.An(u + n).i . we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .4). Thus to O(0) in h we have (jr ' = 4>v — cosh vz.2)<t>„2 + (v. so that J2 (19. (19. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .31) It follows that Dv+2n4>v+2n = 0. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. (19.29) sinh vz or e±uz. (19. v .cosh 2z)4>v = = 2h2A(/>„ . ±2.
.1 .2. 19. 8. F. Singular Potentials (I/. W.I/±2)*. (19. The use of the symbols (v. Arscott [11]. The solutions in terms of cylindrical functions are written *J.h). • (19.7 — we refer to Meixner and Schafke [193].h) = <(>„ +Y.h).37) where p M (±2) = (I/. etc.I/) = 2 A . In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2.38) (f> with (\>v = exp(vz) — Meu(z. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.h2i *=1 J2 j=i. > (f> with </v = sinh vz — Seu(z.4 Notation and properties of solutions We now introduce standard notation as in established literature. In particular the following notation is used for the solutions obtained above which we characterize here by their first terms: <f> with (j>v — cosh vz —• Ceu(z. CHAPTER 19. (19.h). we prefer to adhere to one equation with different solutions.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions.J/±2) = .J7t0 p2i(2j)<f>u+2j.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. M. > These solutions correspond in the periodic case respectively to the solutions and me^. In order to avoid confusion arising from the use of different equations. 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. Meixner and F. in the present context should not be confused with the same symbols having a different meaning in the case of solutions in terms of cylindrical functions since it is generally clear which type of solutions and hence coefficients is being considered. Schafke [193]. v ± 2) etc.442 where (I/.
As emphasized in Chapter 8. h). (19.h). Also from the expansion of Jv{z) in rising powers of z. 16.h)±Sey{z. W. <f)v = N„(2hcosh z) MJ?\z.h). Me„(z. see this reference p. 4>v = Hv (2h cosh 2) * M^\z. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv.h). as we remarked earlier.h) and oo i M^\z.(2hcosh z). 443 (19. h) = exp(inwn)MP(z.44) (19.44) implies. W.41) i=l We see immediately that Meu(z + niri. h) = av{h)M^ (z.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. (19. — oo cv2r{h2)e^+2r>. h) = exp{vniri)Me„(z.e.40) Cev{z. Magnus and F. Meixner and F. h). {2h cosh z) • MJ?\z.h) = exp(vz) + Y. h). h) are therefore proportional to each other as a comparison of Eqs. p.42) The solutions Meu(z.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) .g. Mi.45) T Below we frequently write the normalized solution as in J. 17. (19.MP(z.43) and hence similarly M^\z + inir. For the relation of Hankel functions used below.h).h2) oo = Y. (19. (19. which. h). introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. ' (z. .19.e. (19. this normalized solution possesses contributions exp(i/z) in higher order terms. as Mev{z. Schafke [193].42). whereas the — as yet — unnormalized solution (19. Oberhettinger [181]. i.40) does not.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). 4>v = Hi. 'See e. i.
These points are important in our derivation of the S'matrix below.h) Using further properties like (19. The functions Me±v{z.h2): oo Mev(z. ^ .47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. Mi 3 . h). H™(z) = e^HfXz).48) The series expansions of the associated Mathieu functions M„ (z.50) E.h) can be shown (cf. (19. p.e. MJ>>(0.h) . the first relation is obtained as follows: 2M<£1 = M^l+M^l. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19.e.exp(TM"r)MW(. i. As mentioned earlier. i.49) Inserting this expansion into the modified Mathieu equation (19.444 CHAPTER 19. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.g. Meixner and Schafke [193].8.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. we have the following for a change from v to — v. Singular Potentials where clearly MeJQ. p. Meixner and Schafke [193]. oiv{h) = (19.h2) = ] T cv2r(h2)e{y+2r)z. 130) to converge uniformly for all finite complex values of z.43). . Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. 2M™ = eiuvMi3) + e~ivlrM^\ . h) .46) H^z) = e^H^(z). we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.e. h) = M™(z. For later essential requirements (i. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions.h) on the other hand can be shown (cf. r=—oo (19. s (19.
^ C 2r+2 (19. (19.51) we have 2r (19. .2 [ A .2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . 122. Thus now c 2r . \) /. p.(/^. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. Schafke [193]. . 117. eg = c~v = 1.19.54) 2 C2r 4 ^ h.52) in agreement with Meixner and Schafke [193].[X(u 2 + 2r2) } " c 2r2 This continued fraction equation can again be used to obtain the explicit expressions of coefficients of normalized modified Mathieu functions. Eq. + 2 r ) 2 ] . see J. Meixner and F. h~2[\ ./ i 2 C ^ . W. (39). 2r (19>51) 2r For ease of reading we give the steps in rewriting this. p.^ p 2 2 ^ or c 2r+2 2r2 1 (19.53) /i2[A(^ + 2 r ) ] .(v + 2r)2} / z . \ .^ + 2r + 2 ) 2 ] .T22r[\ A . Alternatively taking the inverse of Eq.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 .* For 'Actually.
Then di 2 1 t" + 2(^1) + • • • . Liang and Yunbo Zhang [189]. MiillerKirsten..27) and truncate the continued fraction after the second step.Q .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.C + l) 3 (i/ . . 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ . .1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. + i). Spector [257]. (19.+0^"> < 19 . H.8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected. J . W.2.[v^.+4)!ll_. H. VahediFaridi [10] (this paper contains several misprints which we correct here). 128(i/+ l) 2 (i/ + 2 ) ( ! / . W. Solution: We put r = 1 in Eq.l ) 19.5 D e r i v a t i o n of t h e S .55). but follow our earlier reference H. + 0(/. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . M. J. t For this purpose *We do this in the manner of the original derivation of R. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. J. Here we insert for A the expansion (19. (19.^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10).12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy.446 CHAPTER 19. MiillerKirsten and N. Aly.1 for the evaluation of a typical case): hb + 0(hw). Manvelyan. and also R.55 ' Example 19..52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(.( " + 2)2] ~ h2 h._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. H.
Magnus and F. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19.59) which tends to zero with r. p.57) ^ 1 .56) where by Eqs. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19. Then y reg = = r rV2M®(z. (19.j7L0 P2i(2j)exp i[v + 2j + 2 2 .19. and then the continuation of this to infinity.. I / + 2 2 l\TT +E ^ i\ E j=i. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). W . we have W r = ip + ut — 7r/4' .2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. and considering the propagation of this wavefront.h) 1/2 w) (19. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\.3a) and (19. In the case of the repulsive singular potential here. .5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const. 22. (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 . Oberhettinger [181].
61) In fact.62) Also with Eq.60) Using the above asymptotic expressions for the Hankel functions. '(z. This means that the origin of coordinates acts • > as a sink.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. p.h) = exp(Mr^)il4 4 ) (z. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo.V + § 9 2 > 4hh* . . 4%r g >0 4r^r ' rg See W.j¥=o Ki(2j)exp(Tmj) (19. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z. p.3b) ln • 7T Q"*4' r fg ° Vk (19.#. We require therefore > > the continuation of Mi. h) through the entire range of !&{z).j^Q p2./l ~ 2 ~ 2z + " 4 > 0. (19.3a) and (19. Oberhettinger [181].A1"—2 z '+"2 A~~ n or rA .448 CHAPTER 19. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z./i). 2h and the square of this expression implies 2 2 ~ ' A.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. The series M^ (z. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. 17. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. Meixner and Schafke [193]. 178) to be convergent for  cosh z\ > 1. AV . '2"4 . Magnus and F. Singular Potentials so that when t — oo : r — 0.A14A. 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.(2j)<gH • (19. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). Now from Eqs. (19.h) can be shown (cf.
49). These solutions converge uniformly for all finite complex values of z (cf. 449 This condition.1. i. p.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. This interchange in the > solution My interchanges y reg with y+.3a).rl)(r2 .r2_) > 0 with r£ = rg(2 ± >/3). and in order to maintain this symmetry.62). the domain r_ < r < r+ — as illustrated in Fig. Thus we choose . z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. (r — r+)(r + r+)(r — r_)(r + r_) > 0. (19.ijt/4 Rez . i Im z z = In r/r0+i7t/4 . Since the real part of z changes sign at r = ro.19.38) or (19. Meixner and Schafke [193]. (19. we have to assign different signs to the imaginary part of z in the two distant regions.63) Thus there is a gap between the two regions of validity — i.1 The domains of solutions. r 7r for 0 < r < ro. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. is satisfied only for r < r_ < r__ or r_ < r + < r. 19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 . .3b) and (19. 19.e.e. A suitable set is the pair of fundamental solutions Me±v defined by (19. (19. Originally we chose the sign of m / 4 as in Eqs. 130). which has to be bridged by using another set of solutions.
and we determine these coefficients from this equality and the additional equality obtained from the derivatives.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.e.^)] dr z=in/4 dr (19.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. Thus. At r = ro the real part of z vanishes and then switches from negative to positive. (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_.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_.B^0. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.z=—in/4. 131). as just explained. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. p.v(z)\ —m/4) + ^[r^Me.450 CHAPTER 19. at this junction we require also the imaginary part of z to change sign. .l/2Me_i/(T. 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. (19.(2)]z=w/4 + 0Me. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. h) (as one can check or look up in Meixner and Schafke [193].66) a = (3 . to r^2[AM^ + BM^\ A. i.[ r l/2 M. h) = Me^u(—z. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3. i. Hence we match the right hand sides of Eqs.v{z)\ . (19. Singular Potentials Then starting from the region r ~ 0.
g]:=f(z)*&g(z)WZ) dz dz Thus. (19. dr BM^(r)]. Then.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.64) the relations M&Hz) jz[M^\z)] and from Eqs.69)) cancel out in view of the nonderivative relations.71) We now determine the coefficients a.70) by MeU and subtracting . (19. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.68) From Eqs.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . multiplying the first of Eqs.3b) we infer that (z = l n r + const. (19. ^ ] = ^ + ^ ^ . the nonderivative parts (from the first term on the right of Eq.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri. = aMev(z) + 0Mev(z). and we are left with relations expressible only in terms of z. (3.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19.70) by Me'_u (the prime meaning derivative).68) the derivatives with respect to r by this relation.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. (19. we use the notation W\f. (19. with the replacements of Eq. (19. (19.69) Thus if we replace in the derivative relations of Eqs.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.65). Thus we obtain from Eqs. (19.65) and (19.3a) and (19.19. (19. and the second of Eqs. (19. (19. A and B. (19. for instance.
e. = IT <*„. (19.Mev] = = = av.^ 47 9? MJ>4)] MeV] [3.A] = W[Mi3\z). 7T (19.u}W[Me. h). Me. we obtain immediately W[Z.4] = . 169) M0) = e«^Ma) _ isin U1X M(A) _ (19. using Eqs. (19.3] = [1.Mev]' P l ' From Eqs. and the second is obtained similarly: W[M?\Mev\ W[Mev.452 CHAPTER 19.46)) W[Mev.j] given in Meixner and Schafke [193] (p.73) We now use Eq. (19. p.] W[Mev.Me^} (19..M^] J™aV.\ \ ( av 2iwKav ^ v . (19.4] = . av{h) = Me. (19. (19. or obtainable by substituting e. i.g. Singular Potentials the second resulting equation from the first resulting equation.* « * r r z H .74) Moreover.u.75) With these relations and Eq. M^4)] + W[MJ>3).e.71) we obtain in a similar way A B = = W\MP\ Me.77) av I 2ism VK \ ctv av ^Thus.(0. TV W[Mev.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. we obtain the first of the following two equations. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.. we use the following circuit relation which can be derived like Eq.58) and considering large values of \z\. 7T [1. 170/171).76) .57) and (19.. and Wronskians W[M^\MIJ)] = [i.48) (or cf.Mevy W[M?\Mev] W[Mev. Meixner and Schafke [193].M^} W[Me„. i.47) we obtain (av defined by Eq. . W[MJ?\ M^] W[Mev. W[M^\Mlf)]W[Meu. 7T W[Mev.M^]= IT e~iv*av.46).M^] W[Mev. . h)/M^\o.^ ^ a „ Q _ . B = J 2 L _ e . (19. the leading terms of the respective cylindrical functions. (19.
h) + BM^\z.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#). Then f' .(ite™)" 1 { 8 J ' f(k. (19. (19.79) where the second expression follows from Eq. (19.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z.85a) .(l)le^}Pi(cos 9). Z ) e ./ ( .l) ' We note already here that with the substitution R = ei7r7. (19. B=tfD'""./) = (l)1/2R2^2i^)ei^)U^. eii/7r/2 (19. as we recall from earlier considerations (e.19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). The superposition of an incoming plane wave and an outgoing radial wave is written and reexpressed in terms of partial waves for r — oo » with z = rcos 6 as: 1 tCT" "1 OO e *« + /(*)!_ ~ 5 > ' c < f c P . (19. f c r .h) = Me_u(z.78) we obtain Ay reg « _ ._ [ / ( f c .46) together with the relation Mev(z. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq.f c . Z ) e ^ ] .80) 2^fc (1981) where \7r/ i t sin VK /(*.h).83) The Smatrix element is therefore given by ifc*"* . in Chapter 16).g. (19.
(19. we can define respectively as amplitudes of the incident wave.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.56). 19.58)) 1/2 2r e~ 2hir cosh z. R — — = 2i sin ivy.2 The repulsive potential and the various waves. V(r)=g2/r Fig. We can rewrite Eq. and with the left hand side following from Eqs.Smatrix in the form s i n 7T7 Si = CHAPTER 19.454 we can rewrite the .85b) We shall obtain the 5matrix in this form in the case of large values of h2 later. (19. 19. Thus in terms of the (3) variable z. (19. R 2i sin TXV. 9?. h) (see Eqs. Fig.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u).56) to (19. Singular Potentials r(i+k) sin 7r(7 + u) (19.86) .2. (19. we can write this (multiplied > by 2ismvTT.58)) in the limit r — 0. (19. and recalling that yreg is proportional to a function Mc> {z. cf.
this expansion is given as 1 oo MJ> Hz.2. MullerKirsten. (19. S. (19. J.88) where the factor Me„(0." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. in which v is complex. This implies basically the evaluation of the expression R of Eq. h) serves the use of the same coefficients as in the other expansions.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . H.Q. however not here. Jv+2r.h) is (cf. 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. Eqs.e. but in the 10dimensional string theory context. therefore we cite it from Meixner and Schafke [193].e. if R becomes a pure phase factor and eil/7r a real exponential.±)2 + (2sinj/7r) 2 .m a t r i x Our next task is to evaluate the expression (19. i. i. i. M^l)(0. Eq. 178). / i ) = Y.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. unity minus reflection probability = transmission probability. By inserting in Eq. W.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). (19.6 E v a l u a t i o n of t h e S . " S e e S.e. unitarity is preserved. Gubser and A. l=—oo 1 1 lite*"* .28)) and R = ey real. Hashimoto [124] and R..In Meixner and Schafke [193] (p. i. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles.** 19.( i J e ^ ) " 1 ! 2 = R2 + ^ .79). The expansion is: 4 r ( / i 2 ) M S ( z . and then exploit it for the evaluation of the quantity R.19. A derivation is beyond the scope of our objectives here. (19.e™*/R\2 ' ' . 180). Liang and Yunbo Zhang [189]. (19.h) The function Mo (0.e.84) of the ^matrix.89) . (105) to (108). (19. (p. (^)l4lu{h2)Jir(hez)J±u+i+r(hez).2COS2VTT = (R. The latter is what happens in the S'wave case of the attractive potential. J. Manvelyan.
h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. by allocating different values to r.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. Singular Potentials 4r (h )M$(0.3 9 ^ .49)./>) = Co> 2 ) £ 7 " = — OO j ± > . one obtains *. Here in the case of the repulsive potential. we obtain M&(0. setting r = 0 in Eq.456 so that in particular for z = 0 oo CHAPTER 19. The formula (19.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).2_4)(l/2_9)iv2y) + <AA./i) with the help of Eq. /=—oo (19. e. (19. h) in many different ways.g. r = 2 as a check). h2 is complex but v2 is real (cf. Next we evaluate M^1}(0.. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . As a matter of introduction we recall the definition of the coefficients with Eq.90).49) from which we obtain for z = 0: Me„(0.94) . u»^.e.90) and choose r = 0 (one can choose e.55). r = 0 and 2. (19.91) Inserting here the coefficients given in (19.g. (19. (19.6 2 ) m 6 (^_i)3(z.28)). ^ (19. /n ^ . Eq. (19. i. We require the power expansion of the Bessel function of the first kind which we obtain from Tables of Functions as I)EH^(T)Thus.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. (19.
12i/2 + 64u .95) These expansions imply for the quantity R: I/! 2v ri1 i . _2_(h\2 . 4(z/ T H ^ 3 .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy.l) 5 (19.4)(i/ .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).2 _ ^ 4 ( ^ 2 _ 4)2 (19./I) h u .2^ 2 ± 59^ . From expansion (19.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.93) and the coefficient expansions (19..!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .7) F " (i/±l)2(^^l)(i/24)V2.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .1) 2u(4u + 15i/ . + # "+" ( I / . .] 2 \v i + {v2 .9) V2 1 fh ±v 1+ 2 + • (19. In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.. We observe that this approximation involves v and h4 and hence is real.96) With this explicit expansion we can evaluate the S'matrix.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .111) / / i 5 4 3 1+ 4i/ 2 + +' (z.IV 2 2 2(v =F 3i/ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.. one obtains M£!(O.32i/ .1 ) 2 ( I / + 1 ) ( V 2 .19.
98) (cf.84): Si ST 1 (11/R2)(11/R*2) .99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2). (19. (19.^ i 4 2+^ # 1 / •.104) We now consider the case of a — 0 implying that g of Eq.2. (19.9 / .28)). In this case R — R* and so 1/i?2 ~ Ofji4). and hence allow for the general case of complex parameters v (the Floquet exponent in the case of the periodic Mathieu equation) although here in the case of small \h2\ we know from Eq.28) that v is real.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.7 CHAPTER 19.458 19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.103) Then SiSt (1+5) i . Then cos 2TT/? = 1 .101) where / is complex and g is real. where n is an integer and a and (3 are real and of 0(h4) We have with Eq.eiun/R2)(eiu*n ~ (19.R*2 ^ * #2 \ 11 tf2 R* (19. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. and sin 2TT/3 = ft/.102) (19. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case. Eq. (e iun e~iv*^/R*2)' (19. + ( i + s w (E 4 .101) is > zero. This is the case of real . (19. (19.
3b)) h2 is real for E = k2 > 0. Oberhettinger [181].1)\} 2 h \2 iv 1 + Au {y .102) we obtain Sis.6C . Here in t h e case of the attractive potential (as remarked after Eq. J. (19.106) and evidently violates the unitarity of S.107b) = Ah2[l + 0(h4)}. With t h e help of Eq.61n/i} + ~9~ 0(h8) (19.l ) 2 2 \) +0{h«) (19. Gubser and A. W. 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.28).3. (19. H. (19. = 1 .Q. (19.e. Manvelyan.107a) In this result* we now have to insert the expansion of the parameter v given by Eq.105) sin7r/3 (19. S. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].107c) Here h2 is actually Vh4. p. 1 29/ ^ (1 . Hashimoto [124]. ( I l l ) ) in agreement with a result in S. At = 2i[l + 0(h2)}. J. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .SiSf » 4 sin7rz/ R (19.^ ) ~ 2 2 459 1+ 9 / • & ( ! . f W .1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). Hence with Eq.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . "This is the result effectively contained in R. MullerKirsten.61n2 . and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. We can also evaluate now the amplitudes (19.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.19.577. Magnus and F. . i.97) this can be written A 4TT2 [v\{v . Liang and Yunbo Zhang [189] (there Eq.
H. i.3. H. (19. VahediFaridi [8] and G. This is therefore a highly interesting case which found its complete solution only recently. J.3.e. Aly. S.460 CHAPTER 19.h) 10. we again make use of the replacement (17.108) For the large values of h2 that we wish to consider here.3 19. W. 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.a }tp = 0. where the part of interest here is based on H.26) and hence set a1 = 2hl + 2hq + A{q. Singular Potentials 19. K. W. Wannier [278].109) ^We follow here mainly D.4 Fig. h) for large h and upwards I = 0.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . H. MiillerKirsten and Jianzu Zhang [226]. Miiller and N.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5matrix obtained above by using the asymptotic expansions for large values of h2. 2. . 19.h) (19. Tamaryan. J.3 The function q(l.$ q(l. completely different solutions.1. Park. 2 2 a = [l + 2 (19. We are again concerned with the associated Mathieu equation given by Eq. H. N.
112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. y'_(0) = 1. 19.21b).yJ\ — 1 and cos TTV .113) 'We are considering nonintegral values of v. . i. We also found the boundary condition (17. Therefore the second term on the right is nonzero.e.y} where VF[y+. we defined the Floquet exponent v and observed with Eq. 19. J4(0)=0.2 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 17.3. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v. and a is its normalization constant. We begin.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. with an essential mathematical step. (17.19. to which we have to return in this subsection.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation.5) that this is given by the relation cos iris = ^ M . y± oc A{z)e2h sm * ± A(z)e2n sm z .21b). (17.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. which therefore cancels out or can be taken to be 1. (19. See also Eq.110) a where y+(7r) is the solution evaluated at z = TV. (19.l + 2y+(l.111) we have W[y+. The replacement (19.§ a W[y+.3.h*y_(l. (17.63). With the boundary conditions y+(0) = l. The behaviour of q as a function of h for some values of / is shown in Fig. y(0) = 0. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively. which is even about z = 0.h*y (19. however.e. (19. i.
32). Miiller [73]. W.112) from evaluation of Eqs. Dingle and H. which is \/2 at z = 0. Dingle and H.e. This paper contains the normalization constants of large/i 2 periodic Mathieu functions. J. from y+(z) y(z) Nn BMz)] a B W oh e"" s m z + C[w{z)]~hl _• e a C[w{z)} 2h . . i.117) a a Referring back to Eq.114) a a Our first step is therefore the determination of the normalization constants No. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . Miiller [73].118a) "We emphasize again: For integral values of v the normalization constants are different.39 2 14 /i 2 + (19. Hence we require (obtained from Eq. J.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. (19. B. (17.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. W.43) we see that z = 7r/2 implies W(TT/2) = 0. We now obtain the expressions needed in 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 .462 CHAPTER 19.A(z)e2h sin 2 sin z sin z (19. B. and hence with Eqs. (19. (19. (17. See R. (17. 1 1 Or see R.114) at z = 7r/2. (19.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). in the second expression 2h is obtained in addition from differentiation of the exponential factor). K [ (z)}2hsinz (19.
2g + 3) 27/i ( 3 g 2 . (17.3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! . Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. a in Eq. . (19.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q . Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.3). 210.117) are known from the matching of solutions in Eqs. (19.19.1Q11QM cos irv + 1 ~ •==— e . p.62b) (see also the explicit calculation in Example 17.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.119a) This formula — derived and rediscovered in Park.62a) and (17.118b) The factors a.i ( g +3)]! 1 24/i {2hl>2). (19.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . Tamaryan.! ) ] ! [ .h . (19.87 2 14 /i 2 [ i ( g .! ) ] ! [ .
109) into the equation (19. Vcosh z \ 1 .2 we insert the expression (19.h)iP(q. „ 1 A „ —A cosh z——H . (19. tp(q. h.3. z) = ip(—q. i = 1.h.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q.122) Correspondingly the various solutions ip(q. for h — oo).120) to (19. dA 1. —z). z) are TP(q. —h.h).h. of course. In the following we require solutions H e ^ ( z ) . we can obtain the linearly independent one either as ip(—q.e.z) i/>(q.123) We again make the important observation by looking at Eqs.122) that given one solution ip(q. h(q.3.464 CHAPTER 19.z) = A(q. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.h) =?==. it is still a solution but not with the symmetry property ip(q. h.h.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. h. With the solutions as they stand.h..z).s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. V— 2in (19. the expression (19.120) The resulting equation for the function A(q.124) Since this function differs from a solution ip by a factor k(q. Then straightforward integration yields > inq 4 „.4. h. z). h. 17. . (19.108) and set ip{q. We define these solutions in terms of the function Ke(q.z)exp[±2hismh z\.? s m h z) Vcosh z (19. . —z). h.109) remaining unchanged. each with a specific asymptotic behaviour.h. h. z) can then be written as .2.3 Following the procedure of Sec.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) = ip(q.z) or as tp(q. . 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ .h.z).3. —h.
(44) of W a n n i e r . IT a n d 7 = 00 — * 7 > : 2fcsinh y + i<j7T h Aw 26/? T h e factor e " r ( ? + 1 ) / 2 o n t h e r i g h t of E q . Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO.fi2) can be reexpressed in terms of Hankel functions. the Bessel functions contained in the expansion of the associated Mathieu function M{.^ s i n h 7?)]g. 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].<5±1) (19 125)  (the expression on the far right in leading order for h2 large).fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . : 7r/2./i 2 ) M (19. i.h2).e. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . .126) For large values of the argument 2/icosh 2.3 The Potential 1/r4 — Case of Large h2 465 Instead. av{h2) = ^(0J ^ Mi i . * ** = > rojKe( ". We observed that these satisfy the relation (19. as we saw earlier. (z. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions.45).e. i. (0.h2) = OvMPfrh2). (19.127) sin z T /'1 M TT\ — <>{ 4 . (19. 2j<. / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo./h). i.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 . This is. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. 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.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].19. . [ \ . o u r h2 is W a n n i e r ' s — k2 a n d o u r a 2 is his a w h i c h in t e r m s of o u r p a r a m e t e r q b e c o m e s a = 2k2 + 2iqk + A / 8 . In order to be able to obtain the Smatrix. Me„(z. achieved with the help of the Floquet solutions Me±v(z. >• * F $ 0 * e".e. in performing this cycle of replacements the function picks up a factor. we h a v e 77 = 0—> 6 = 0.
h) Be^(z.q.3. h.h) = e(z) exp ihez ./i).h). (2) (19.h). KeW(z.q.h.g.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. In this way we obtain in the domain .q.4 T h e c o n n e c t i o n formulas We now return to Eq.q. z): Re(2\z. (19./i) Be^(z. He (z.h) He (z.130b) r_0 rll2exp\g/r (ig) ' 19. HeW(z.h) (3) = = = = Ke(q. (19.q.130b). (19.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.q.q. 3te»0.466 CHAPTER 19.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .vr .g.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19. 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. 3te»0.130a). (19.eW(z. (19.130a) Re^(z.q.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.z). exp[—ikr — in/4\ kr Re{2)(z.
(19.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z.134) in the region where the function He^(z.134) Considering now the first of relations (19.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z.g.g.q.h) sin irv = sin7r(7 + ^)He (3) (2.133) In a similar way one obtains the connection formulas — sin nv HeW(z. we see that the solutions (compare with Eqs.130a). and thus can be matched in this region.129)) He (2) (z.135) sin irv 'This means. t With Eq.q. h) = Ke(q.130b)). (19. / i ) + sin7r(7^)He(4)(2./i). g . q.h).(z. — sin 777 H e ( 3 ) ( 2 . (19. (19. h.h) is asymptotically small (cf. h) = sin 7r(7 + v) He (1) (z.132) into Eqs. q. these are exponentially increasing there. we obtain in the domain tfcz <C 0 the relations: Me.132) These relations are now valid over the entire range of z. and He(*> exponentially decreasing. q.q.131) where the second relation was obtained by changing the sign of v in the first.h2) Me„(z. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19.(z.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z./i 2 ) = Me_j. h) — sin 7T7 He^(z.i ^ / 2 H e ( l ) (z> ?) (19./i) — sin 7T7 Re^(z. (19.li 2 ) Oil. (19. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e . . z) and He (3) (z. Changing the sign of z.h2) a. (19. Substituting the Eqs. (19. h.19.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19./i). h) = Ke(q.g. z) are proportional there.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . q. Eqs.
137) We see that this expression agrees with that of Eq.e. is only the dominant term for large h2} 19. (19. which in our case is the solution He^4^.( .119a) or (19. (60) in the work of G.135). Wannier did not have the relation (19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. i. sin TXJ.139) ''Equation (19. Wannier [278]. (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". Singular Potentials The factor on the left.135).133) we can now deduce the 5matrix Si — e2t51. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . (19.119b) as „4/i 1 + COS TXV ~ rrl ( g .5 D e r i v a t i o n of t h e ^ . q.e'i7ri 1.135) is the equation corresponding to Eq. Thus here the S'matrix is defined by (using Eq.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . (19.(19. The quantity 7 is now to be determined from Eq.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.133)) — sin TXV He^4' (z. of course.119a) and (19.3. and therefore remarks after his Eq. H.85b) obtained earlier.Sin TXV.l ) (l„—ikr e From this we can deduce that Si (19. The latter is defined by the following larger behaviour of the solution chosen at r = 0.119b). (19.m a t r i x From Eq. cos TXV + sin TXV. (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . where 8i is the phase shift.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.468 CHAPTER 19. .1 . Squaring Eq. (19.
4 The absorptivity of the S wave (attractive case).140) From this we obtain the absorptivity A(l.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. i.141) COS TTU irvj — i s i n WR s i n h TTUJ.e. (19.19. This is different from the behaviour for small values of \h2\.e. Si xe ~ te ilir COS KV V cos2 irv — 1 — e (COS^Q COS ITU ig7r 1 + e'iqn ielln cos TXV 2 2 C O S TTU (lp/2)n (19. 9/2 (91) with the approximation q ~ h obtainable from Eq.137) together with (19. (19. i. 19. A{l.139) could actually be neglected) implying  cos m>\ = cosh nvi. . as a comparison with Eq.§ 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. we can approximate the equation for q ~ h (i.8h COS From Eq.109). for the attractive potential. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1.e. T h i s is r e a l for UR a n integer. (19. imaginary part uj (so that 1 on the left hand side of Eq.28) shows. (19. since cos TVU is large.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. h) of the Zth partial wave. the real part of v must be an integer.138) we obtain.
F. C. **R. This is sketched schematically in Fig. (19. 19. D. Lii and J. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . such as phase shifts.P. Handelsman. Apparently this is one of the very rare cases which permits such complete treatment. Tiktopoulos and S. YuanBen [285]. A. (19. N. 19. Cvetic. Fubini and G. Bertocchi. Limic [177]. B. VazquezPoritz [61]. S . A. 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. L. 11 J.e.139) 2Tr(16h)q /32\ f e Thus with Eqs.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . Y. Partial or other aspects of the weak coupling (i. Esposito [88]. J. E. . Eden [46].470 CHAPTER 19. H. Paliov and S. Furlan [20]. A." as well as other aspects. in fact. Jones [78]. The diminishing fluctuations of A(l. and G.k) in the approach to unity are too small to become evident here. Pao and J. Shabad [248]. H. I r a n [61] and M.107b) and (19. In essence.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. it seems it is not possible to guess from the above the result for the absorptivity of a singular potential with an arbitrary negative integral power. in further contexts beyond those already mentioned.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. Treiman [271].** 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.142) we can see the behaviour of the S'wave absorptivity as a function of h. Lew [128]. see also M.4 (there are tiny fluctuations in the rapid approach to 1). Pope and T. Apart from the papers already referred to. D. Cvetic. H. Rosendorf [225]. Lii. N. "G. small h2) case have been considered by some other authors. and. one may expect corresponding results for other singular power potentials. W Some of these sources can be helpful in further investigations.4 . S. **A. Masson [191]. Dombey and R. Challifour and R.
how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. B. also convergent expansions are known and a lot of literature exists on the equation. 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. and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. Dingle and H. B.g. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. Dingle [71]. Bender and T. e. Wu [18]. M. W. *R.* However. [19]. We shall also see explicitly. J.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. 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.Chapter 20 Large Order Behaviour of Perturbation Expansions 20.* In fact. Muller [74]. the asymptotic solutions in different domains all have the same coefficients. 471 . f R . T. This is indeed a remarkable connection. We concentrate in this chapter on the large order behaviour of asymptotic *C.
8. r for p = 4.2 The function ur = r\/pr vs. 20. 5. Fig. Presumably this has never been done so far. 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.7. 19.1 and cases in Sec.1 The function ur = pr jr\ vs.2. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. 8. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. 10 12 14 16 18 20 Fig.8. This approach leads .2. r for p = 4.2 and below. Since the method is also applicable to convergent expansions.6.472 CHAPTER 20.6. as evident from Example 17. 20.
that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term. The 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].2) or. equivalently for g = l/h. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. 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. 20. (20. even if each En is known. and the expansion is said to be Borel summable. this term represents the size of the error. The traditional method of using an asymptotic expansion is to truncate the series at the least term. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). The formal perturbation expansion of a quantity E(g ) .2. In some cases the answer is affirmative. It can be seen from these. i. The behaviour of such terms is illustrated in Figs.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. in the power expansion of the exponential function). i.3) n=0 Thus for any given value of g .e. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20.1 and 20. the function E(g2) is given only approximately by Yln=o En92n.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E.1 Introductory Remarks 473 automatically to asymptotic series.20.e. T V lim h2N 2 n EQ?) . .J2 2n h E (20. and the integral E(g2) is called the Borel sum. for instance. as we discussed earlier in Chapter 8.
4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: .1)? Asymptotic expansions originate from integrating a series (i. In general this is not so easy. 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.474 CHAPTER 20. n=0 (20. . (20. El . Large Order Behaviour of Perturbation Expansions Eq. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . (20. . 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.4). In fact. (20.e. (and E0 = 0).n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0.6)). (20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. . n=0 (20. . 1 we obtain oo F(z)£(*)" = — . that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]).n _ En+\ n! En+1 ^ (!)"«!. In any case. . . Thus if CO F(z) = aiS(z) + J2 anzn.5) then with the integral representation of the gamma function F(z) = (z — 1)!. In this ideal case the integral of Eq. Q. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0.
. or can we determine Ar independently? Fig. Popov and V. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).. (20.e./ \p\dr\.. we have a quantum mechanical problem where probability can leak away.10) r as shown in Fig. i. N = l.9) The question is therefore: How can we determine 9 E(h).1 Introductory Remarks 475 except for possible subtractions.. 20. 20.N = l. This is the case if the real potential has a hump of some kind.3 The potential V(r) gr». Weinberg [228]. § where 7(ff) = To exp  . e. for < ^ 0. if the potential or a boundary condition is complex such that the problem is nonselfadjoint...11) V. (20.g. (20. Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) I jQO Ar = 7T JO tir~l%E{ti)dti. .3 and g < 0.2..20. ? When ?s E ^ 0. $s E can be calculated by the WKB procedure. M. The potential (20. S.. if the potential is V(r) = +grN.2. and E represents an eigenenergy.10) is considered in detail in this work..
Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. Weinberg [83].134)). Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. V. the cosine potential being t h e most completely investigated will also be seen in the following.1: Application to the Yukawa Potential Evaluate 9 £ = —7/2 for the Yukawa potential V(r) = g— r / \ r and insert the result into Eq. The 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.11).73) in the W K B or semiclassical approximation.476 CHAPTER 20. the BreitWigner formula (10.76) (see also the comment on the comparison with the Kepler problem before 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 . of course. as we expect on the basis of Eq. .113)). Popov and Weinberg. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. However. Eletsky. Popov and V. (20. as was (and still is) not the case for t h e equation 1 V . An interesting application is provided by Example 20. (14. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. 20. (11.1. M. Example 20. the perturbation method of Sec.11) is.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. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. 8. Thus in the following we apply these methods to our typical examples. Here n is the principal quantum number of the Coulomb problem (see Eq. S. (20.9) for Ar. (11.^ The logarithmic factor in Er is a novel feature of this case. Equation (20. L.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. Solution: The solution can be found in a paper of Eletsky.
(17. Eqs. (17.h) = 2h2 + 2hq+j.J) = (g + 4 j .j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.l(q. (17.58) and (17.26) and (17. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. are given by i . obey the recurrence relation jPi(q. M2i = 2j2 J[Pi(q.12) (20. to avoid discussion of difference equations here) 2 l{q j) ^ (16)«(»l)!(2i)«/° / _ 2q 2q + 3\ f i \ ' [\{qmwm\ sr ){j)> (2(U5a) R.13) represents effectively a partial difference equation.2 A = Mx + ~M2 + 7^WM3 + •• • .13) In the special case j = i the boundary conditions stated after Eq. (17. 3)? . i.34)) that the coefficients Pi(q. (cf.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. j . h). arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.20. . (20. Dingle and H.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.55)) X = E(q. or equivalently of the quantity A.l ) +2[(q + Aj)2 + l + A]Pl. Eqs. % = Aq. W.l ) ( g + 4 j . Eqs.j) are known. J. Miiller [74].j + 1).60). B. (20.j) of functions ipq+4j.j) a n d M2i can be deduced. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.M2i+1 for large i.e. (17. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .14) Thus these coefficients can be obtained once the coefficients Pi(q. The recurrence relation (20.Cq.3 ) P i _ 1 ( g .Bq.
16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . Using this relation and the duplication formula lw('\W±^. (20.91 l)]![i(g3)]!} 2 { [ . Hence exact Borel summation is not possible. . In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.l (20. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .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. also written lCj.16a) (20.478 CHAPTER 20. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.18.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient. see also below) .*.
W. but with different boundary conditions.2 solutions obtained earlier.* However. MiillerKirsten [200].4.1 Cosine Potential: Complex Eigenvalues The decaying ground state Our intention here is to consider again the cosine potential and hence the Mathieu equation with the same large/?. Previously. remains unchanged. we emphasize a few points *M.20. f H . this difference was found to be real.5 \ \ / \A Fig. i. l . the difference q — q$. 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. We retain the previous boundary condition at one minimum.e. n = 0 . 2 . At the risk of repeating what we explained earlier.3 20. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. A means to achieve this was found by Stone and Reeve.3 Cosine Potential: Complex Eigenvalues 479 20. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. . 20. 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. when this is g0 = 2n + l . . in the calculation of the level splitting as a result of tunneling. J. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. 20.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. \ 1 \ \ \ \ \ 0.4 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. . Stone and J.3. Reeve [262]. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.
and allows its imaginary part to tend to infinity. as we found these for this potential. i. (20. but t h a t with complex argument there. which means replacing z by z±ir/2. Now we change this situation at the minimum on the right. Thus one chooses not the solution with real argument at the minimum.21) <u> + ( i « . but achieves exactly what we want. Here we choose t h e second way.e. We can distort the potential there or alternatively impose a different behaviour there on the wave function. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. 20. 8 the equation can be approximated by d2ip(w) (20. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. . X + 2h2 = 2hq+—. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. 20.480 CHAPTER 20.4 (around z — 0) to the one on the right (around z ~ ir). 20. a boundary condition which at the face looks somewhat abstract.20a) As stated. Effectively we consider the tunneling from the left well in Fig.2h2 cos 2z]ip{z) = 0. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear).4. Setting w = 2h1/2z.4.Ah2z2 • • • }^{z) = 0.T K ^ a D_i_{q+1){±iw). Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 .20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . we shift the argument by n/2. Previously this situation was the same at the neighbouring minimum. (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.
(where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. that at z = 7r. such that this deviation becomes imaginary. by a different method. A derivation is given subsequently. however. 3 . .> 0 for z > ±ioo. Thus. whereas here we start off with a solution even or odd about a minimum.23) .e. .3 Cosine Potential: Complex Eigenvalues 481 As explained above. Since the large/i2 solutions valid away from the minima were the solutions of types A. 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. One should note the difference to our earlier case of the calculation of the level splitting: There we were seeking solutions which are even or odd about the central maximum of the potential. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. .e. we naturally constructed the even and odd solutions with these (cf. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. i.63). .20. whereas here we construct these from oscillatortype solutions.22). The proper harmonic oscillator solution there would be that for infinitely high barriers to the left and to the right with a dependence on an integer n which enumerates the quantum states in the well from the even or symmetric ground state with n = 0 through alternately then odd and even states upwards with n = 1. Eq.4.24) Considering t m O i n Eq. (20. (20. A (which we can loosely dub solutions of WKBtype). Here — again as before — we take the finite height of the barrier to the right of z = 0 into account by taking q only approximately equal to go = 2n + 1. we see that ips ~ cos \j2qhz for z ~ 0. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. 17. 20. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. We consider here explicitly only the case of the ground state around z = 0. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. 2. . This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions.
28) Extracting Du{—w) we have A. Oberhettinger [181]. 92.482 CHAPTER 20. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. . (20.g.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20.(.w) = e T Dv{w) 2TT n») AvViDv^iw). Magnus and F. 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. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function.27) requires the differentiation of this expression with respect to the index v. tp(v).29) Equation (20. p.•w) + 2ir T{u) <"+!)% Dvi(iw). p. 2. (20.24) are linearly independent around z = 0 for q ^ qo. §See e. W. However.g. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. W.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.h(2h1/2 J £2=0 (20.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. this can be handled in a few lines.27) differ by an angle IT.30) 'See e. Magnus and F. 17. Thus the only nonvanishing contribution comes from differentiation of Y[—v). We see this as follows.25) qo) E2 0: (20. and then evaluation at u = 0. From Tables of Special Functions'. Oberhettinger [181]. (20. (20.3. The arguments of the functions appearing in Eq. we know from the solutions of types B and C in Sec.
pp.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. Magnus and F. (20. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. which are valid around z ~ 7r/2.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1. p.. Oberhettinger [181]. Magnus and F. We now go to the WKBtype solutions. Differentiation of this equation and subsequently setting v — 0 yields. (20. W.//n and ij>{y) = C . . 92.)e" w /4 2 l ^Dx(iw) (20.e.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. Thus calculations for qo ^ 1 are not given by M.4. dropping from the derivative part of Eq.29). For u — 0 one obtains the result (20.33) = IV2~TT. . Stone and J. and "Prom W. (20. we obtain 4>s D0(2h^z) D0(2h1/2. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. Oberhettinger [181]. "See e. 5 7 7 . .e.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. Reeve [262]. r(i/) We now return to Eq. i. 3 . we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. those of type A.20.g. i/^0 483 T(v) It follows that d dv [T(u)\ (20. . This result is » not easily generalized to <jo > 1.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .29) v=0 (in + In w)D0(w) + v2ire (m + lnu.27) exponentially decreasing contributions. including the derivative of DE2ufl(2h1/2z)).31) dT(u) r*(v) dv 1 i/>(v) ^ o l.31).
Recalling that we have shifted the potential by 7r/2.36) cos  . Eqs. We observe that in the matching domain (i.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. To this end we set z' = z — ix..484 CHAPTER 20. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). we obtain the required solutions from the earlier ones.cos z — 2/icos z ipA^ Tcos  and i n A —.36) V'WKB cie~2hehz' 9h  . near z = n the boundary condition (20.34). 2 c2e2hehz UJ2 .— 721 . cos z/2 ~ (20. i. y COS c\ J (20. Then V'WKB becomes (with cos z ~ — 1 + z' /2.e. (20. Thus the WKBlike solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^Tsinf 2/i hz2 _z_ 2 (20.34) and (20.40) From Magnus and Oberhettinger [181] (p.29) and (17.sm I We now want to match this solution to that of Eq.e. (20.40). x . c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20. (17.37) Comparing Eqs. where we can then match these to the solution ips. (20.e. that in approaching z = ir. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.41) 2TT .38) in its limiting domain towards 2 = 0.35) We construct with constants c\.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . We now return to the WKBlike solution (20. (20. smf (20. (20.36) and consider this at the other end of its domain of validity. by there replacing z by z + ir/2.39) Now. i. Thus in the present case and with q ~ qo = 1 we obtain „2ft.
3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real. (20. J Hence Eq.45) Comparing this result with tps of Eq. ^'mliw <2a47) For qo = 1 we recover the result (20.^. (20. i. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . The general result stated by Stone and Reeve [262] is therefore presumably guessed. with c\ = exp(—2h) from above.2/. .46) Repeating these calculations for the general case (i. We derive the general result in the next subsection using our own method.e. 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 .1/V)_le^_ Z With this result Eq.46).complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(.38) can be written TAWKB (20.44) = DQ{2h / z) ± (2h7r) / e. (20. (20. (20. (20. 1 2 4 1 2 8h ^^2 2 (20.20. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.39) and (20.43) Comparing now Eqs.34) we obtain E2 = i%\ = ±ih22Qe~8h.43) we obtain.e.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2.
z. At the minimum to the right of the barrier shown in Fig. 20. (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.48) These conditions provide the required quantum number qo = 2n + 1.20a). W. the result is formula (334) there. •+(=)* V(f) 0. and expanding cos 2z around the point z = — TT/2.) + 7T V> = o. (20. V(z) . We have to impose two sets of boundary conditions. i. •_(:)*<>.46) we consider a half period of the cosine potential as illustrated in Fig. J. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. at z = TT/2. i.5. MiillerKirsten and A. 20. i. Fig. (20.e. H. Eq.E curre it \ j. Wiedemann [4].5.3.()#0. Achuthan.5 The cosine potential from — TT/2 to n/2. Considering the original equation.e. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .2 CHAPTER 20.Ah2 ( z + .*.e. 20. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. .486 20. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. this equation may be approximated there as d2ip ~dz^ + E + 2h2 .
B and C.z).z)=tpA(q. 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.e. three pairs of solutions of types named A.h.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish.21)).51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. Recapitulating these from Chapter 17. we demand that ^(*)L<. (20.h. (20. (20. i. Proceeding as in our previous cases and substituting the expression (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.21) into the Mathieu equation we obtain. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.r/2 ~ e i(2hq) V 2 .50a) V'AfeM) = where (the second expression is needed for later comparison) Aq(z) cos§(9i)(iz + lyr) + W) [ t a n (±z + W)]«/ 2 [sin(i* + ±TT) C O S ( ^ + ±vr)] V' ' sinfr+Vaz (20.51b) 2i(«1)[i(9l)]'. For the continuation we use again the WKB formalism. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. as shown earlier.20. (20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.3 Cosine Potential: Complex Eigenvalues 487 (using Eq.
(20. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity. the relation q .57) where go = 1) 3. (20. (20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) .52a) = Cq[w(z)}. we obtain 1>±W = \ a a (20. we obtain our first transcendental equation.=F2 2\1/2{28h2y/4e~Ah l + O (20.56) Applying the boundary conditions (20.^ i .52b) The first solution is valid asymptotically around z = —vr/2. and the second around z = 7r/2. 5. 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> (20.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. i.488 CHAPTER 20. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2.48) and proceeding as in Chapter 17. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20.55) Extending these solutions to the domain around the minimum at z = 7r/2. Here the upper sign refers to the even states and the lower to the odd states.e. ^ W H .qo . \vw). We return to this condition after derivation of the .
Ah2 cos2 z ~ 0.[ [Ah cos z . i.e. I^A(Z) overlap parts of the domains of validity of these solutions. Our procedure now is to match these to exponential WKB solutions. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. Byrd and M. D. Friedman [40].20. These then provide complex exponentials like (20.67). 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. e x p ( . . Thereafter we extend the latter across the turning point down to periodic WKB solutions. p.49) and we can apply our boundary condition.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .58) Ignoring the correction A / 8 . and we obtain the proportionality by appropriate expansion. as we shall see with the result of Eq. (20. formula 280.5 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 .2hq] / dz) We know that the solutions I/JA{Z).61) [Ah2 cos2 z2hq]1/2dz. we see that the turning points at z± are given by 2hq . Some nontrivial manipulations are again required to uncover the generation of important factorials. F.01. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (20. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °.Ah2 cos2 z ip = 0. Jz+ ' T In principle the integral could be evaluated exactly — see P. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier.
62) as follows.490 CHAPTER 20.64) . Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. (20.» _ i y v . 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 /•«"(.sin2 z = l . we write cos2 z = l . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.59) 2h COS Z+  ~ 0.63) We handle the logarithmic terms in Eq.cos2 ( z+^) = (z+ .) +0 z+ IT Inserting this into the integral I.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 (20.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (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. (20. We have for the following expression contained in (20.
.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . (4^2)i/4 [ i( g _3)] ! V/ l (20.50a) and (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .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. (20. 1/2 (20. .50b). <$&?(*) 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\"  . 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.62).3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. We ignore this here.60) by (4h2 cos2 z)1!4).63) and (20.20.65) [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.64) into Eq.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.+ qj V2 4 A.66) The WKB solution therefore becomes (approximating the denominator of (20. (20.
?/ 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.. io(8/*)g/V^i6ft2)V4[±(g3)]! .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T . (20. i. i 27r(8^/ 4 2/t /4 ' r(±) .z)}]. We now return to the even and odd wave functions defined by Eq. S1H (2/ KZ ) / (z f 2 (2^).492 CHAPTER 20.49).U ^K^.69) l(8M9/4e2^ \2 [ i ( g .z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). 1 2 1 ±2 .67) and (20.l^l !T ^^* T2 7 r ( ^^ /. we obtain T _ ( ± ) = 0.+) + B^) HAh^/\ir. (14. Large Order Behaviour of Perturbation Expansions again using the Stirling formula.T 7 77 r )V2 /4(2 .2 [ ? ) Imposing the boundary condition (20. where _ 1 + [±) *) + £ (20.50)).55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+. (20.i ( 2 / i g ) 1 / 2 ( z + . Eqs. we obtain l(8/i)g/ 4 e.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .KW4^8 4 .e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.49).1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .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. Thus with IPA(Z) and TPA(Z) taken from Eqs. (14.™) .
h)+i%Ego.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. (20.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . i. Ex = 2h2 + 2hq0 + A 1 (<? .72).l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.26). Eqs.57) and (20. 20.qo) El Ah' (20.72) is our second transcendental equation.h)+2h(qq0) (16h)i°+le'Sh E(qQ.74) This result agrees with that of Stone and Reeve [262]. (20. (20.h) + (qq0)( BE — 90 E{qQ.73) Inserting this expression into A = E(q. we obtain E(q. (20. (20. (20.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ . We now have the two equations which have to be satisfied. Eqs. (20. h) of Eq. In both cases we wrote (cf.3 ) I = }(»« !7rl/22I(.h) + i l + O '4{[^ol)]!}2 E(q0.21) expanded about q = qo.57) by the left hand side of Eq.25)) E = A = Ei + E2.e.3 R e l a t i n g t h e level splitting t o imaginary E We can compare the results of the two problems with the cosine potential and the same solutions but with different boundary conditions. (17.47). in fact. Equation (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.h) = = E(q0. the result (20.72) We observe that with the square root as on the right hand side both sides are complex.3.75) . for q an integer both sides are imaginary.
7 8 ) i=l V . W. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. f E .9) the coefficients Ai of its perturbation expansion.W [£(901)]! < (20. i.) {I6h)™/2e4h 1 + . (20.74b).I g  ± l ( 2 „. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. by others. The same formula can be derived in the case of the double well potential. A = . B.80) *P. J. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. its tunneling properties and the large order behaviour. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo.3. (20. Fateyev [26]. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. Eq.2 below.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE.+  .494 CHAPTER 20. Setting ng AEqo:=2iSfEqo. Wiedemann [4] and Example 20. the discontinuity across its cut from zero to infinity. 25Eqo=8h[ 2\1/2 . 20. we can recalculate with the help of Eq. (17. . cf. MiillerKirsten and A.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. We had A= or _2ft2 + 2ft. Achuthan. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles.e. H. Bogomol'nyi and V.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. A.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut.
there is nothing special about this case. B.l ) ( < ? + 4 j . and the method can be applied in many other cases. 20.l ) +2[(q + 4j)2 + l + A}Pi_1(q. the integration is seen to give (20. The integral is recognized to be of the type of the integral representation of the gamma function. For convenience we write Eq.74). But as Dingle remarks.j) satisfy the recurrence relation (20. Using again n = (go .j) + {q + 4j + l)(q + 4j + 3)Pi1(q.19) for the large values of i under consideration here. *R.14). W.1). (20. J.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.81) We observe that this result agrees with the result of Eq. (20.83) One now sets ftH = ( 2 % P * ( 9 . j . J). (20. a = J + \<1> (2084) . (20. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. (20.j + 1).13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . We also observe that these terms do not alternate in sign. MullerKirsten [200].20.3 ) P i _ 1 ( c ? .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. Dingle [74].13) out again: jPiiQj) = (g + 4 j .. so that an exact Borel summation is not possible. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. f H .
(20. . g < 4 j .87) Taking terms of 0(h~l+1) of this equation.86) one can convince oneself that Eq.^(Z). 1  1 / 2 (4 (20. z2 (20.91) + Wa+lil/2(z) = (1 .85) Considering now the generating function of the perturbation coefficients. 4j(4j + 2g + 4) + / .e.1/2(z) (a1)! (20.1) + 2 (a — 1) a . Stegun [1]. Taking dominant terms. Abramowitz and I.e.K + ^ ] L + K^)WK. (20.89) Replacing here K by a and \x by 1/2.e. the recurrence relation (20.4.g. i. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . (20.90) See M. (20.l/2l Wa.83) can be written j . Now. we regain Eq.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.4 ) ^ . which are solutions of the differential equation known as Whittaker's equation.31. 0 4j(4j + 2g) pi_i(a . A.(a) oc .h) +p(a + l.85).h) 2+ 8h a p(a. e. i. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. formula 13.l. d2WKh dz2 + 1 4 K z zM2n WKtll = 0.1) + 2 P i _i(a) + Pi_i(a + 1).h). (20.h) = ^2pi(a) with p. Whittaker functions WKjfl.z)WaA/2(z) Setting V.ll{z)= L . (20.496 CHAPTER 20. p(a. Pi(a) i.a R .85) implies p(a . a. nx pi_i(a + l). we obtain (2a . (q + 4j) 2 ~ 4j(4j + 2g).27h pi(a) a = 4j(4j + 2 g .
\)V .(« 1) 2 ]. Oberhettinger [181]. h —> —h. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20.(« . h) is expressed in descending powers of z = 8/i. Magnus and F .94) Hence to insure that p(a. 89.h) = \~\Pi(&) with pi oc h l. ^Observe that if one solution of Whittaker's equation (20. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . Dingle discovered the solutions^ p(a.88) is known.93) (a1)! ' where f(h). Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).87).„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. v .g(h) are functions of h which in view of Eq.86) have to be chosen such that p(a. (20.4 Cosine Potential: Another Method of Calculation we obtain Vai. g(h)(20.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential. we have to remove a factor (recalling a = j + (7/2 of Eq. a (20.h) ~/(fe) ' (a1)! . (20.i/2(z) + Va+lA/2(z) = 497 2+ Va. p. > > ^See W. In the present case this means the replacements q — —q.i/2(z).84)) {8h)q/2( Ah for j < 0. .20. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. > (20. (20.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. i..(« . (20.e. h) is expressed in descending powers of h.92) Comparing Eq.92) with Eq. such that p(a.j — —j.
Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2.l ) ] ! 11 R. g — —q. (20. (20. .95a) Inserting (20. i).+ i f or 3 < 0 (20. B. j — —j.8 / t ) for j > 0.95b) Then for j > 0 we obtain from Eq.j _ g / 2  1 / 2 ( . we obtain the result of Dingle: (20. This expression can be obtained from the recurrence relation (20. ^ .498 CHAPTER 20. _ i s. Comparing > > > this result now with Eq.1/2(8^) ( . Dingle and H. n = 0. (20. Correspondingly we set for j > 0 / 1 A pla = j + q. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 ..94) here we have a series in descending powers of h.83) and is given by 1 1 . h\ oc (8h)q/2e~4h ( _ .84) we conclude that the original coefficients Piiflij) a r e f° r large i given by the following expression — apart from an as yet undetermined proportionality factor cq — We determine cq by imposing as boundary condition the expression for Pi(q.).94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1.96) and a similar expression for j < 0 with /i — —/i. J... 2. Muller [74]. W.1. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.
100) **R. of Eq.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.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.3)]!2 2 '+^. (20. W. E* J'=I i+M 2 [2i + 9 . J. Ryzhik [122].l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. note the misprint there: (r — s) ought to be (r — s)\. p. formula (30). is M2l = 2j2j[Pi(<l. Thus ° 9 (i + \q) 2«(i+Igl)! p .20.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.j) (i + lg)[2i + i ( g .97d) Then the coefficient M<n of the eigenvalue expansion. i.l ] ! (20.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.+i?_l]!}2 (20.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.97c) We assume that in a leading approximation we can cancel the factors (i + \q). M.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. A special case of the formula can be found in I.14).1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . formula 0.e. B.157(4).Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20.l ] ! {[. Miiller [74]. Dingle and H. (20. 5. Gradshteyn and I. l{Q. . (20.{j + \<l). S.e.0) = 1 as desired (although not really enforcible for large values of i).l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.
e. Eq. [^ + i ( g .o i • .. n = 0.17). i... (18.e. i\ M i large 1 .34) that £ is an expansion in ascending powers of h 6 /c 2 .5 20.101) This formula allows us to cancel all factorials [i. since \{q — l) + \{q — 3) = \q. Eq.105 ) the latter because we see from the eigenvalue expansion (18..4)) £ = ^ = ^+ w and y= ^ > ( 20 . the result of Bender and Wu [18]. i. (18.86) and set (cf..86).i ) ] ! [ ^ +  ( g . [19].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.102) Thus the result is M m 1 2«Fi + g . We refer back to Eq. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20.. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. i.Am i .e. (18.l.e.3 ) ] ! _ 1 Hence M or. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity. 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. i.' 1 i • (20.81) for q — qo = 2 n + l .l l ! with the large i behaviour in agreement with the result (20. . i.500 CHAPTER 20. 20.e.5.]\.2.1.
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.e. and then the relation (20.l ) r .107) we obtain (i)r r 71" JO yirZ£{y')dy'. 20. (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. in terms of K where go = 2ftT + 1. in fact as one with Borel summability in view of the factor ( .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.5. the discontinuity across its cut.86).20. i. .106) With £= E r=0 Mr (20.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . (20. We leave this to the following Example. and hence inserting here the imaginary part from Eq.109) or.4) of Bender and Wu [19].108) Proceeding in this way.77) can be verified.e. This can then be compared with the level splitting we obtained for this potential.(20. The result establishes the series unequivocally as an asymptotic series. (18.110) This is the result in agreement with the large order formula (4. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20. i.
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). R. 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.502 CHAPTER 20. Propin and H. Wu [18]. . Harris. M. K. Silverstone [52]. MiillerKirsten and A. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. M. Harrell.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. + 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. A relation similar t o (20. in t h a t when the quantity on one side is known.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. M.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. J. V. " P . S.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. Cizek. Eqs. J. T. S. J. Liang and H. Graffi. W. J . Paldus.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. J. t p . Solution: For the solution we refer to the literature. Wiedemann [4]. J. (357) to (371).77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . E. J. J. J. Harrell. V. J. The relation (20. Cizek. J. Bender and T. MiillerKirsten [164]. R. E. K. S. Large Order Behaviour of Perturbation Expansions Example 20. . H. Nakai. Propin. Damburg. Gram. t The relation (20. Paldus and H. Wiedemann [4]. Grecchi.^ 20.Q . G. J. MiillerKirsten and A. *Some of the literature in this field following the work of C. W. R. Its usefulness has also been demonstrated there. it can be used to obtain the other. Silverstone [62]. W. as in later chapters. H. Grecchi. 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. Achuthan. of International Workshop on Perturbation Theory at Large Order (Florida. *R. Damburg. [19] can be traced back from articles in the Proc. J. Achuthan. 1982) [139].
However. 'See in particular H.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman.e. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. since in such cases the probability of this system choosing a path far away may be considered to be small. R. The formalism is not so useful for boundstate problems (in this case the initial wave function is. since this illustrates the most important and most frequent use of path integrals in a wide spectrum of applications ranging from field theory to condensed matter physics. we digress later a little and consider the path integral also in simple contexts of scalar field theories which can be treated like models in quantum mechanics. P. Chapters 2 and 5. A brief introduction is also contained in R. For the application of path integrals to the hydrogen atom (i. Feynman [93]. Hibbs [95]. *R. A standard reference is the book of R.Chapter 21 The P a t h Integral Formalism 21. This formalism is particularly useful for evaluating scattering or transition amplitudes. Feynman [94]. P. in general. ^ Our interest here focusses on quantum mechanics. can be found in the book of B. Feynman and A. Felsager [91]. Probably the most readable introduction. S. The path integral is. We illustrate the method here by rederiving the Rutherford scattering formula with this method.* 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. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. 503 . P. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. However. which explains also difficulties. A further wellknown reference is the book of L. unknown). Coulomb problem) we refer to existing literature. Kleinert [151]. Schulman [245]. however.
x).ti){x.tf\ip) .ti)ip(x.5) Sn = X / fc=l mo (zfe .e—»0 Sn.1) Let the wave function i/> at time t = ti.504 CHAPTER 21. The Feynman Path Integral Formalism 21.tf\x.ti\tp). the initial time. we wish to know* ip(xf. x. ti) = 5(xf . (21. i. */ = U + (n + l)e. U. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). 2mo dx2 (21.2 P a t h Integrals and Green's Functions The onedimensional Schrodinger equation for a particle of mass mo moving in a potential V(x) is (with mostly h = c = 1) i d2 + V(x) tff(x. *In Dirac's notation we write this equation later (xf.x. we can compute the action S1.e. (21.tf.t).2) where K is a Green's function. 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).3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.ti). (21. ti). =J dx(xf. (21.tf) = / dxK(xf. Obviously K(xf. be ip{x.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo.
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.tf).t). The usefulness of the method therefore depends on the fact that some paths are more probable than others. Here and in the following we need one or the other of the following integrals (cf.6a) . the Fresnel integral (13. of which one is illustrated in Fig. 21. The multiple integral involves individual integrals like f dxk exp Xk)2 + {xk Zfci) 2 } with Xk contained in the step up to this point and in the step away from it.28)): f dr\ exp ia a) ' (21.1 (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.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. 21. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. as will be shown to be possible below.ti) to its final position at (xf.1 A path in (x. We now consider intervals in position coordinates and in particular the following integral over intermediate position coordinates Xk which corresponds to the sum of exp(iS/h) over the uncountable number of possible paths from Xi — XQ to Xf = xn+i.
. tf > U. for example.xk+i) exp .t ~Xk+l> + xk+i 2 ) +{xk .506 CHAPTER 21. We then integrate and finally let /x go to zero..Zfci) 2 } I +{xk+i x*.5)): G(xf.. ie f me\ 2a\a~) 1/2 (21.6b) —c ia o drjr] e x p—V € (21.mQf Xk L ~2e^ ^ . one has j ^ dr1eifa+i^ /£ = . J—oo dxneiS/h — • 0.Xi. 2 + 2a.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.0 ^7 (x f c + i .tf. dxneiSn/n. One therefore removes the troublesome factor by defining (Sn being the sum in Eq.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.xk_i) dx exp i — { 2 a .z f c _i) 2 } } . dx exp z—{a. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.37)) dxk exp d(xk .2 + (x + x f c + 1 .ti) '•= f g m ) (N(€\]n+i dxL. (21.8) Note t h a t in spite of the n integrations.m0 . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.s f c _i) + (xk+i .xk+x oo —oo oo . n > 0.x f c _ i ) 5 \m0 J (21.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7. 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.(x fc+ i . (21.6c) In the evaluation of these integrals we replace a by a + ifi.
g.Xi. 183. (21.2). A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. See the square root in front of the exponential in Eq.tf. However. 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. in fact.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure.2 The Path Integral and Green's Functions 507 Eq.1. summation over all possible paths. we arrive at a path integral without the measure factor N(e) above. These integrations are indicated by the horizontal arrows in Fig. Different constructions of these families of paths usually lead to different measures. nor the limit of integration of some variable Xi. which all vary from —oo to oo. (21. § See e. We shall see later (with Eq.t In the following we demonstrate that the function constructed in this way is.24). and so on.21.e.ti) For tf > ti we now write* G(xf.tf.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration.Xi. i.U) = / JXi=x(ti) = K(xf. however prefixed with a new factor. 21.g. 183. the Green's function K of Eq. (21. It is neither a variable of integration in T){x} —> Ylndxn.5).Xi. 11 For discussions see e. (21. . it may be noted that with a different philosophy concerning the summation over paths.tf.5). p.7) for n — 1) to give a time interval. and will therefore not be attempted here. Felsager [91]. B. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. B.10) can be looked at as the product of a succession of free particle Green's functions (or propagators). Felsager [91].e.ti) for tf > U. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. (21. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. p. (21. (21.26)) that — ignoring the potential — the multiple integral (21. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. G(xf. i. as in our discretization in Eq.
188.ti).tf + e. Then we have on the right the factor G{x+rj). however.Xi.xi.Then we can construct a recurrence relation as follows. the other with that of the action of the free particle SQ. Each of these linear pieces is completely characterized by its endpoints. Felsager [91]. 21. pp. Consider a time t = tf + e for position Xf = xn+2. Expanding this in a Taylor series around x. It is for these reasons that — unless required — the measure is frequently ignored. and one is calculating the Feynman amplitude or kernel relative to that of the free theory.ti) = 1 j^p: f°° cte n +iexp } • fm0.ti. We have G(xn+2. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.U) imo 2 G(xn+i W)Idr.tf + e.Xi. one with the full action S. Then the measure drops out.1 and allowing n to go to infinity. 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. so that the piecewise linear paths fill the entire space of paths. (23. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. (21.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf. . The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. the discussion in B. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. I We employ this method later in Example 23.74)).tf. we have See.3 in the evaluation of I a specific path integral (see Eqs.exp 2e n ieV(xn+2) G(xn+2 + V. we have G(xn+2.xi.508 CHAPTER 21. we obtain the same as before except that there is no factor N(e).72) to (23. 187 .
t Thus + 0(e2 (21.tf + e.Xi.6a). It follows that ip(xf.13) We integrate with the help of the integrals (21.U).xi.16) .Xi.U) +••• \ 509 G(x.ieV(x) + ri< —G(x.tf.tf.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x. tf.ti) +V(x)G(x. ti) satisfies the timedependent Schrodinger equation.tf.Xi.2 The Path Integral and Green's Functions so that G(x.xi.Xi.(21. tf.tf + e.xi.Xi.ti) = 1 d2 + V(x) G(x. U) = e^G + 0(e 2 ) dtf 1 d2G(x.Xi. tf. we obtain: G(x.ti) +\if(^G{x.xi. d and this means d i—G(x.ie 1 d2 2nV0dx^G{X. Xi.U + 0(e ).ti) = G(x.Xi. 2mo dx2 (21.— p G ( j .ti) N(e) or G(x.tf. U) .ti)ip(x.6c).ti) = j^rr dnexp I ^rj1 .xi.tf.14) = G(x.xi. tf + e.ti)\ .x.xi.ti) for tf > ti (21.tf.15) We see that G(x.tf + e.tf.ti) .G(x.Xi.ti) 2 —l€ — ^2 + V(x)G(x.tf)= dxG(xf. tf.6b) and (21. Retaining only terms up to those linear in e.tf''Xi.tf.tf.xi. . U) + . U) + Q(e2 = (lieV(x) + 0(e2)) G{x.tf.21. (21.tf.Xi.
510 CHAPTER 21.Xi. The result. denoted by KQ. Hence our original function K is the same as G.ti) = = 9{tfti)—G + G5(tfti) . (21. or K(xf.10) is indeed to be identified with the timedependent Green's function. is the Fourier transform of the nonrelativistic propagator.17) Differentiating K with respect to tf we obtain —K(xf.3.x. 21.x.ti) = 5(xj — x). The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. and the Green's function KQ\ S0= Here (cf. K0 = 9(tf .x. Eq.U)6(tf . (21.15) is a wave function (cf. Eq. now called So.tf.20) . as may be verified by inserting this into Eq. We have therefore established that the Feynman integral (21.tf.U) = / T){x} eiS\ [fdt \mx2.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. as this is also called. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function. (21.U. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.U) = G(xf.U).x. (21. In a problem where V is small it is convenient to calculate K with a perturbation method.1)).tf.tf.8)) Go(xf. 21. (21.16). It follows therefore that {{W ~V+2m~0~^) K{Xh t/.U)G0. The function G which satisfies the timedependent Schrodinger equation (21.
7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.Xi.i ) 2 .ti)9(t).oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.Xi.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.tf. (7. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.37) — by explicit solution of the differential equation which the Green's function satisfies.24) ( s / . Then Go(xf.ti)= iX (o. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.21) fe=l and t = (n + l)e = tf — U and h = 1.Xi.tf. we have Go(xf.21.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2./ .22) • \ n/2 .z / c . We obtained the same expression previously — see Eq.. exp 2 (21.t) K0(x.Xi.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.tf.8). In Example 21. (21. (21.Xi. n + i .x 0 )" n + 1 (in agreement with Eq.tf.tf. But t = tf — ti = (n + l)e. exp where Xj = xo and Xf = xn+\ and there k = 1).Xi.ti) Ko(xf.23) for Go and obtain Go(xf.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.1 we show t h a t oo /. f c .tf. (21..ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.ti) = = G0(x. (21.23) We now insert our expression for In into Eq.25) T h e quantity Ko(x. . 2t Go(xf.3 The Green's Function for Potential and hence Go{xf.
Then oo 2 / (21. as / JXi f r D{x{t)}eiS°lh . Solution: The result can be obtained by cumbersome integration. we have instead of 5*0 and KQ expressions S and K.an) Then n I 1/2 (n + 1)A" "I 1/2 fX r ioo ^n + dxn+1eiX{^+v2y2Hx"+2x"+1 l ) 2 } J —OO (n + l)A n «ix„+1ea(Sl»22»lI«^I. (21. The Feynman Path Integral Formalism We can now see the group property of the Green's function or propagator by observing that with the result (21.5) is carried along.22).25) we can rewrite the Feynman path integral or kernel (21.)!} J — OO .54). i. Example 21.26) with n integrations and n + l factors KQ.xi)2 = 21 xi ^2o(„. the latter being a Green's function with interaction like that for the harmonic oscillator.1: Evaluation of a path integral Verify Eq. (21. In our proof of induction we assume the result is correct for n.x „ + i ) 2 = (x„ + 2 .t0) (21.10) in a product form. Eq. or else by induction.f_^y e. (7.x ^ 2 + y2 . n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .)+(1"+2.xo)2 + (x2 ._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.512 CHAPTER 21.x0.22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. (21. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 . For n = 1 (see below) h This result can be verified like Eq..tn)••• K0(xi. (21. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .. ^ (xi . If the potential V(x) of Eq. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.tn+l•xn.ti.2y(xn+2 .x „ + i ) 2 = ( x n + 2 Xi + Xi .6a).e.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.7) by direct integration with ^ .
3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj).2.a . (21.29) Hence with KQ from Eq. We pass over to momentum space with the help of the following two integrals.x2rriQ 1/2 toij_. a dp exp [ipx + iap2] = exp — i(21. The first. ) H — (xn+2Xi) n+2 n+l n+2 y2 . (n+l)A n n j(a.25): Ko{xJ) = = (S) e{t)(^X'2 y J .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. it .e. Using Eq.2 y ( a . and + 2 2 l n + 2 ( x n + 2 .21.3. A(n + 2) 21. 21. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . i.Q .2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ.Xi) + ( x n + 2 Xi)2. This quantity is also described as the nonrelativistic free particle propagator.i) . (21.30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .i) H —3/ . (21.i j ) H n +— z 1 .2y(xn+2 .27) we have (with a = t/2m0) .28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21. (21.n+2a. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. n + 2 . iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .6a). The second integral is an integral representation of the step function. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. e > 0 .
The first order correction contains one factor of the potential V. The generalization to three space dimensions (required below in Eq.2 Contours of integration.28) this becomes K0(x. ( 3 b) ° The expression E+ V — it 2m.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. The Feynman Path Integral Formalism Im t \t>0 Rex Fig. This is the type of expansion most people describe as perturbation theory.Q (21. (21.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion.514 CHAPTER 21. Thus in first . i. Replacing 9(t) by its integral representation (21.51)) is fairly selfevident.*' where 2mo — r. 21. the second order two factors of V and so on.e. an expansion in rising powers of the coupling constant (contained in the potential).31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. 21.
(21.4 Including V to First Order: Perturbation Expansion h=l. \2meJ n+l J.U). li f f 515 order (cf.xi.Xi..^.32) Since — we recall this for convenience G(xf. (21.)} (21.. . ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2. dxr.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.xfc_ OO J— OO J—OO L J.5) a n d (21. .t) JSo dtV(x...i._i By regrouping factors this becomes Gl n+l lim ^ e n—>oo V{xhle) dxn / i ci) n+l dx z+i / * * / d / _ i .21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/. .i V e / D { a .^ J_00 V(xk. x (n+l)/2 \ lim .21.t)] n+l . } e °° f l f •/ oo /"oo F ( a .ti. Eqs.i.tf. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.t)\.(xfc .tf.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.t/.^){iV(a. . (21.33) xG0(xi.20)) and with elS = exp i S i o / hi dtV(x.Xi.t) the first order correction is (G\ ~ G — Go) Gi(xf.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x. dxn exp i ^ —.
ti).ti)K0(x. OO /"OO / oo dt / J—oo dxKo(xf. Jti Consider the term of second order.35) We also observe here that the problem of timeordering arises only in perturbation theory.x. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.xi. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.ti) t2)K0(x2. (21. (21.t)K0(x.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).516 CHAPTER 21. .t. in writing down the expansion exp i dtV(x. t2.t2.t2)V(x2.e.tf.xi. i. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.I dtV(x. They must therefore be deleted.U) xV(xi.i / dtV(x.t2)V(x2.ti). The factors of "V" originate from the series expansion of the exponential exp ftf . (21. i.tf.x2. This is cancelled by n! different time orderings. = i dt2 / dx2Ki(xf.t) (21.t)V(x. dx2Ko(xf.xi. Xi.36) {if •j.t) =li dtV(x.tf.e.t2) xKo(x2.t) ! Jti r*f dt'V(x. / Vdt+ JIfVdt)( fti = Jti Jti ftf ) \Jti pti Jti I' Vdt' + f f Vdt' Jti rtf Vdt' J t\ j Hi Vdt rt\ Vdt' Jti Vdt Jt\ Vdt'+ Vdt+ Jti + \ S Vdt I Jti f Vdt'.x2. t) Ju Thus the term containing n times V contains a factor 1/n!. t') + • • • .37) Jti The last two contributions do not describe the time evolution from ti to tf.
xi. The diagrams represent mathematical quantities and hence are designed on the basis of rules.t.ti) (21. x"^ X x"2. U) (21.t) ) K K K0 'V X K o Fig.tf. or in momentum space as mostly in field theory.tf.t) is V(p.21.i ] P I dt J dxKn(xf.t2 K o (x 2'l2 .t)Ko(x.3 along with the diagramatic representation of Eq.Xi. called Feynman rules.Xi. These may be formulated in configuration space as below. The Fourier transform of V(x. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. Such diagrams in the context of field theory are called Feynman diagrams. tf.E). we are concerned with quantum mechanics.t)= f dpdEeipxiEtV(p.x.3 Feynman rules and representation of the Green's function. of course.ti) i dt = K0(xf. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. Here.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.Xi. E) which is given by (in onetime plus onespace dimensions) V(x.t)V(x.39).t)V(x.Xi. (21.x.t)K0(x.tf.ti). (21.40) The rules for the present case are given in Fig.Xi.ti) = K0(xf.39) dxK(xf. x r l i 'V(x.tf. For various reasons — such as illustration.38) . 21. 21.ti).tf.t. .
= l. dn = (L/27r) 3 dk.Xi. to one particle in all of space.e. (21.k = 2wn/L. with p = Kk.t) = . i.7 = e i k ' x _ i : l i with f i/}*i/jdx.tj)K{xf.ti).e.518 CHAPTER 21.41) We have to pass over to three space dimensions. and dn = dnxdnydnz. for the allowed values of k: kx = 2imx/L. * k 1 • . (3) The transition probability per second is = \A\2 . . 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. (21. and this means the transition of a nonrelativistic particle from an initial state 'i' to a final state ' / ' with wave functions ipi (or ipin) and ipf (or if>out). There the wave functions ipi.ipf have to be properly normalized.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. We recall for convenience from **Note that from the wave function we obtain. T = duration of time for which potential is switched on. The plane wave function normalized to one over a box of volume V = L3 is. i. mo V since ip is normalized to one particle in volume V. here with E — k 2 /2mo. The Feynman Path Integral Formalism 21.ti)il)i(xi.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit threemomentum interval** = V/(2irh)3. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. ^(x.tf. (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.
— . Proceeding in this manner one obtains Fermi's "golden rule".T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.e.47) .21. (21. (21. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. (10.44) = u(Ek. P(*) = j r (2145) (cf. H = H0 + \H'. i.t) = e~^lT\ r where r = Ixl. u = ^\H'kfp(k). Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. «= — .r)eiE^\ H'ki(t) = J dv u*k(r)H' Ui(r) Thus the amplitude ak is effectively the spacetime Fourier transform of the interaction or potential XH' contained in the Hamiltonian. in which T is a large but finite time.k. 47T 1 3 7 (21.= 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. Then H'ki can be taken out of the integrand in Eq..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.. V(x.117) and Example 10. Eq.44) and f_oo can be replaced by fQ.4). and with these the following perturbation ansatz: ih^* with an(t) = HV. # = ^an(i)VnM) n = a^(t) + \a^(t) + . The transition probability per unit time is .
also called Born approximation.39) OO /"00 / dt / OO J —OO dxK0(xf. (21.tf). T a large but finite time.tf)K0(xf. (21. t)K0(x.f.t) (21.30a) K0(xf. by threedimensional dxf and so on. so that 0(t) can be dropped) K0(xf.44) (here in the scattering problem with initial and final energies equal).520 CHAPTER 21. (21.e..tf)Ki(x. i. t. Xj.* / . x. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.f.. This now has the form of the amplitude (21. Thus with the ansatz Ko(x. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x. (21.49) together with the following wave functions (of asymptotically free particles) 1>i(*i. dt dxip*f(xf.ti).t. (21. we obtain A1 = clS(ti tf) / dxf(i)V(xf.tf. (21.>/>. In our previous onedimensional considerations we had with Eq.50) Here x — xj — Xi and t = tf — ti.48) where from Eq.51) Inserting this into Eq.yLi. (21. The threedimensional generalization is (taking tf » T and ti <C —T.52) .^ / .x. The lowest order approximation of A.).49) i.tfx.t) = ^ J*ie*<*'*><<*'*>.Xi.ti). Ai = dxf / dx. (21.ti)'ipi(xi.x.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .44) by contracting the propagators to instantaneous point interactions.tf)ip*(xf. t.t)K0(x. £. we have to replace the onedimensional quantities dxf etc.tf.tf)ipi(xf.tf. (x*.i x (i)V(x.tf. ti) = co5(t — U)5(x — Xj) etc. We can relate this amplitude to the amplitude ark of Eq.t)V(x.xuti) = ^ J°° d p e i p x _ i ^ * .
5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. £). Then integrating over q and q' we obtain . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.p ) .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 . 2 „ x v~ m o (21.21.k.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .T) = ^JdXJdt%*t2/T2exp i ( k . which is evaluated in Example 21. Then we require the following integral.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 .54) Now inserting the expression for the potential V(x.).2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.k. (21.k. k2 ipx + i—t + ikx — i 1 2mo 2mo (21. this becomes A(p. P2 .k.T) = JdtJdx(^\V(x.56) (21.p 2. (21. (21.57) we have A(p. .T) = x exp L z/ q/2 k2 521 f dx.~ .t)exp A(p.59) .iaT2/8: L With 0 0 • .r) = = y .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.i ( k 2 .
mo m (21 .„." + M ] p) ' "•"• ^2 . <rtot. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson). integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.64) Inserting the result now into our expression for the total cross section. dE = PjV = ^ E ± .63) we have: Probability per unit time = 77777.\ 2?7lo / 1 (21. . ( 21 .62) We want to consider the limit T — oo. / / P 2 _ .P ) 2 + M2]2 4 VvrT2. we obtain .59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . T ) ~ . 66) o .< ^ £ \M VTTT2.k. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . Using the result (21. Using Eq.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. In our case this has not been done. write down the probability per unit time which is ( 8 \1'2 _a2 16^ . (21. e (21.T2f 8 V/ 2 . k . y 2 [ ( k .^ .59).„ 2.a ( — ) ^ .60) Now. { I i _ F T l } . This expression assumes that the integration over time t is normalized to 1.63) We establish this result for T — 1/e in Example 21. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. otot contains the factor \A(p. (21.e. 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . e 8{k) = lim 7=*—^ (21. k2 . i.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.522 CHAPTER 21.3.65) But E=*.51 — 2 /w2 y ~ —.T)2/T.
setting M2 = 0.69) This is the wellknown Rutherford formula. —Mr 4w = . 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 .70) . so t h a t (21.e.j .5 Rederivation so t h a t of the Rutherford Formula 523 M^f) . 2mo / dft.j . i. (21.21. /Yukawa = ) Solution: We choose the 2axis along the vector q.^ .68) Inserting (21.3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a*o 2K Joo for a > 0 and hence the following representation eikx = lim — .67) and (21.* 9 ' d r e . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.cos 9) = 4k 2 sin 2  . Example 21.4.= e " f c a ^ o 2v^fa /4a = lim £^o —. and inserting the expression thus obtained into the final result of Example 10.M £ M * = / m0kdn = m0^2m0E (k . This result may also be obtained from lyukawa of Eq. p 2 = k 2 .59) by replacing there q 2 by ( k — p ) 2 .^ .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21. Then (x = r) foo „2jj. 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.65) with the cutoff M > 0.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .e. (21.e .67) T h e delta function here implies energy conservation.p ) 2 = 2 k 2 ( l . esfn (21. (21.68) into Eq. giving the potential a coupling constant..
'<*> = . 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 ) .x) (21.75) . a.6 P a t h Integrals in Dirac's Notation It is instructive to devote a little more time on the Feynman formalism by rewriting it in terms of Dirac's bra and ket notation.q).. however. We have x> = I J ^ l p X p l x ) . ' i a = lim e —0 pk 2 /e2 e^TT 21. We consider again nonrelativistic quantum mechanics. (21. Va we obtain g(k) = Je~k V a = 0.* .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. a simpler method suffices. (21. Differentiation of (21. In configuration space the state of the particle is written x). We write the state vector of a particle of threemomentum p in the momentum space representation: p). In the present case. (yx) = 6(y .72) co From (21. Since dxe" 1 1 1 = « / .70) and (21.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . 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).524 The function required is CHAPTER 21.^0 a^C 2^/TTO.71) with respect to k yields the same as °° 2 . 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /.oo J / )eikx. Direct integration /4a .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx.
H.t).6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc ..81) Comments on signs in Eq.79) Feynman's principle is expressed in terms of the propagator or transition function (x'.t> =iH\x. i. i. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.160. pp.t) = +iH\x. (21. we must have d/dt{x. (21. ) H = (x</>o)s. M^ip(x. See also L.21. d/dt\x.t\ = iH{x.p).82): The general Schrodinger equation has the state vector to the right.^ l x ) . Ryder [241]. Then ( X  ^ .£) = ( x V ^ ' . 4) is to satisfy the Schrodinger equation.e.We define x. We consider first an infinitesimal transition. (21. x.t) = e* H t x). (px) = e~ipx.e. The time development is given by the Hamilton operator H.e. Thus (qPp) = p(qp) = p(27r)3<5(q . i.t) = —iHip(x. i.t). as had to be shown.We now introduce the Dirac bra and ket formalism. at initial time ti to its final position xy at time tf. Then the states {x. (with ft = 1) d/dt\ip) = —iH\tp). so thattt d — x. t\ipo)s. x' — x infinitesimal. so that \if>t=o)s = H>)H.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p).t).t'x.77) (21.t)} constitute a "moving reference frame" in the sense that ip{x. . The equation 1 = J ^p)(p.t\. 159 . t' — t.82) (21.t).If '/'(x.0) = e""x).. which describes the propagation of the particle from its initial position x. Similarly we have a position operator X with Xx) = xx).x.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. (21.i) = e iH4 x. t) = (x. Hence for the infinitesimal transition (x'.i) = K(x'.e. i. (21.t'.i'x.e.80) (21.83) (21.76) These relations imply the completeness relations 1 = J dxx><x.
P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.x') + V{x)5(x' .* .i p '.V 2 < 5 ( x .89) Using this and previous relations we obtain for the Hamiltonian (21.p').x = V(p . (2187) (21.90) (P'IVIP) = y<&.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).85) the amplitude or expectation value or position space representation (x'#x) = .91) . i.88) Thus P has the operator representation P= /dxx)V(x. (21. 2mo Since (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.y<&(p'ix')(x'ivix)<Xp) = f dxe.e.x y(x)e i p .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). Pq) = J 7^3Pp)(pq) = J dppp>*(p . "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21. (21.526 CHAPTER 21.x).
EQ — Hi .93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.j.tj+i\x.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .fx. We have <x'.p') + V(p .tj). for the amplitude of an infinitesimal transition.82)).94) In the above we have chosen the signs such that p = —iV and H = i— ot (21. tf subdivided into infinitesimal transitions (xj+i. i.84). (21. (21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx . Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj. We now consider the transition of the particle from Xj. t).95) when applied to the bra vector (x.t') \^(27rf5(p 2mo +V(pp' where we used (21. i.92).t) (cf.p') • • • j W < 5 ( p ' . But e'ipx + 0[{t .21. the signs are reversed when these operators are applied to the ket vector \x.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 '. (x'.92) We now return to the expression (21.p'). ti to x j .p) + i{t .e. 527 (21.i'x.
tn respectively.tj) = / ••• / d x „ _ i d x n _ 2 .xic) by fc — Xfej H(pk. .+ V(x f c ). . x n which the particle passes at times tQ.a — —e/2mo and divide b o t h > sides by 2TT.e.xk (21. t n  x n _ i . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.x 0 ) + p 2 • (x 2 .tixo. i..528 CHAPTER 21. x i ..i/xj. d x i . (Xf.ti) rx/=x(i/) d x i dx2 .dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi . d x i ( x n .tf\Xi. x f c ) = ^ . .. x i ) ( t i . (21.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .tfei) fe=i . i.x i ) H h pn • ( x n .2.. We now insert for each of the infinitesimal intermediate transitions the amplitude (21. .t 0 ) + H(p2..e. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. dx. d x n . then corresponds to the sum over all paths.X n _i)] h) (21. ..xn)(tn in_i)].. . x exp [ . d x „ _ i . zmo T h e n t h e integration over all p ^ can be performed by using Eq.i { f l " ( p i . T h e integration over the n—1 intermediate positions.. t n _ i ) t„2><xi. . = 1 X X<jPfc We now replace H(pfc.t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o .98) tk — *fcl # ( p f c .94) and thus obtain a phase space expression.97) H hH(pn.ti.tf\Xi.99) . .x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa .27) in which we make the replacements x —> eifc.
101) Then Eq. i.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. > Thus the phase space representation of the pathintegral* (21. i.H(x.103) is another form of the pathintegral for which the integration with the help of Eq. (21.100) over V{p}= L J ^ rwooll V (27T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O.e.p)]\.. to / m0 \ 3 / 2 las.e. (21. Smolyanov. A. (21. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.98). .fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .ti) = J D{x(t)}D{p(t)}expi [' dt\pi. G.] ex H ^I (. Truman [254].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.tf\xi.21. (21.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. / {xf. Tokarev and A.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.8) as N(e) so t h a t the integral does not vanish for e — 0. G.
I hi dtL(x.8).x) (21. (21. this agrees with the corresponding discrepancy in the onedimensional case of Eq. xfc). Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk.ti) where nl D{x(£)}exp '.iJ(x fc .tf\xi. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. £ /  X J . t.t'}.104) where for the particularly simple Lagrangian L under consideration here m0±l . Thus finally we have Feynman's principle: (Xf.tf\xi.8). Garrod [106].x) . nl (xf. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq.e. operator (21.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation.530 CHAPTER 21. We have now obtained the path integral representation of the matrix element ( X / . Consider the following expectation value with primed and doubledprimed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'.V(xk) = m0±l . p .105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}.H(xfe. i. x. t to x. . (21. pk) = L(xk. pfc) = Pfc • xfc . (21.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). £ J ) with no operator in between. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x.
/ ^ I < « . Thus <xV'xx'.* * f dp = j (2^ p) ^ d .107).76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). With the wave functions (21. (21.*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.t"\x\x'. With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'.5 l^ t 'lp / > e . .* * .t"\x5){x5\x'.i p '• x . . (21.6 Path Integrals in Dirac's Notation For the verification we use Eq.21.109) . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers. x fdxG  ^ P <p"x3)<x3e^V) x (21.77) and thus have 531 *!*') = *M = * 7 ^ I P > « . ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'.t'}. .v ' x ' and thus {x". (px).t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.
0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .tf subdivided into n transitions (xj+i.x') . This is the expression corresponding to that of Eq.0 dp x„ / (27T): 1 . We now proceed from there along parallel lines.110) Recalling Eq.£j to Xf. (21.i ( t " .£'> = x"[«5(x" . ( x " .x ' ) (2^ i^+vM + D «p(x"x') (21.90) we can rewrite this expression as (x". Inserting here the integral representation (21.94) for the case with no operator in between.*/<x"x5)(x5tfx'> + .£ n xx n _i.• dx1 (x n . ] .iit!' 1') 2mo + vtf) + <5(x"x').tf — tn and x$ = XQ. . Thus with x^ = xn.tj) at equal time intervals e.i(t" . tixx 0 .t"\x\x'. . we obtain with (21.£j) = / ••• / dx n _idx n _2 . £ n _ixx n _ 2 . = x"6{x" . (21.113) Inserting for each of the infinitesimal intermediate transitions the amplitude .85) (xVxx'.ti = to as before.i{t" .t'){x"\H\x') = X + •••} (21. (21. (21.£/xxj.112) where x" is a cnumber.tj+i\x\xj.111) 1 .. x / / e .i t " ( x "  #  x 5 ) ( x V ) +.t n _i) x ( x n _ i .f ) i f e i p . t0).t') (xV'xx'.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • . We now consider the transition of the expectation value of operator x from Xj.t') dp .x') .74) of the delta function. tn2) • • • (xi. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.532 CHAPTER 21. we obtain (here and in the following the limit n —> oo at the end being understood) (x/.
.21. U).3 " n d x f c .t"\x'. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 . tj  xj1 x». We demonstrate first that the operator representation of P given by Eq.« • > We reconsider therefore the transition amplitude (x".y (27F (27F •" (2^(2^)3 exp[.iyxxj.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'.7 Canonical Quantization from Path Integrals (21. dyi l xi x 2 . the string of factors xi X2 .ti) 533 rxf=x(tf) = / • • • / dXidx.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq..{^( P i. (21.112).\Xi.t') (21.2 .x n _ i ) ] . Xn _i X„ xi dx. n—>oo *•»• 21.t"\x'.t') . dxnn _i Xi X2 .. tj) = xf (x/.x 1 )(t 1 Z & ) +H(p2.t"\x'.115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] . Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta. we obtain* (xf.x 0 ) + p 2 • (x 2 .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{. For simplicity we consider the case of one spatial dimension.. .( x „ .t') sider the variation S(x". (21.114) x exp [i{pi • (xi ..t') = (x" + 5x".2 . ..xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/.89) is consistent with the conjugate expression dL (2L116 * .x2)(t2 ~ h) + • • • + H(p ••• + p „ .t"\x'.105).10) or (21. (21. x n _ i x n reduces to the single factor x^.and hence may be put in front of the integrals to yield (xf.. x (21.t') = and con 5x"Vx„{x".x ..114). tj \ xt.{x". .tf\x.
Thus in this case of classical mechanics dL_ d fdV.534 CHAPTER 21.118) JI(x)/h . " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t.t').t i_(dL_ x't' h\dx J t„ (21. Then. t".120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). (21.118) — = (21 5x"(x".t'): 1 fx" 6(x".(x". (21.10) and set D{x} = D°{x}/JV(i/ . and we demand the validity of the equation of motion.122a) = 17) 5x"Vx/.tt). The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. we separate the metric factors from T){x} in Eq.i/(x)/fi ) j ^ ) dtL(x. (21. dx dt \ dx _ (21.t"\x'. <9L' dt —dx + — 5 x ox ox f . dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. For the purpose of enabling some formulations below. with N = N(t" . jt/ 37 ( £T. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. x) Jt Jt dL.t') =  6— / 1 V°{x}t .l&cdi.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. dL r /•*" . . " (t).t"\x'. Hence w « = (t).x).
125c) Corresponding to our relation (21.Xf = x(tf). Streater and A.124) where with A = tf — ti. The following properties are assumed to hold: (1) / Jti V°{x} = ^ 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.7 Canonical Quantization from Path Integrals or JL(x»y\x. (21. F.1 (A)iV.125b) and the measure property (3) N~\A + A') = A^.t'). Wightman [264]. (21.t')(x'.t'\ = l.122b) By comparison with Eq. . 6 (21.125a) (2) the completeness relation holds. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. 535 (21.tf\xi.126) R. (21. x' (21.21. ^2\x'.1>) = ^{x". Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). I(x) = / JA dtL(x. S..t"\p(t")\x'.115) in three space dimensions. / Jxi V°{x} / Jx' V°{x}.1 (A / ).x).U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. § The evolution of the system is expressed by the Feynman functional integral (xf.
(21.127) We can now write down an expression for a timeordered product of operators. Then the relation follows with the help of the properties (1). Eq.. (21.t"\x'. (21.t"\x'.ti).129) = ^ y With Eq. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.x](x.t"\X'.t').127)) (x".t"\x'.536 CHAPTER 21. But for t = ti we have (and equivalently for t = tf) as we can see from Eq. We consider the case n = 2. We can rewrite this expression as the commutator relation (ITU*" x"^](x".128) where tf is later than ti and t\. (21.. (21.t\x'.133) = ih{x"..ti) f ' V0{x}xlX2eu^lh. we have the relation (xf.t') + x"^J^{x". Thus if T denotes such ordering.tf\xi. (21.t"\x'.ti) = Xi(xf.tf\x(ti)\xi.tf\xi\x'.t'){x\t'\x2\xi.t').t') (21.132) The generalization to higher spatial dimensions proceeds along parallel lines.ti) = —!— p V0{x}xx • • • xneu^h..tf\T(xix2)\xi. Thus (xf.114) (xf.t\x'.122b) we can deduce the nontrivial canonical commutation relation arising here.t>).130) ~i5x7jX"{x".t"\x"\x'.t').tf\T(Xl • • • xn)\xi. (21. The Feynman Path Integral Formalism The time t is in general a time between ti and tf. (2) and (3) above. We have (cf. .t') \ i ox" i ox"J or in commutator form [P.t"\x'. tn lie correspondingly in consecutive order between U and tf.ti) = x' 'Y^{xf.t') = ih(x.131) = ih(x".t')=x"(x". (21.
1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. such as the fine structure constant of quantum electrodynamics. T. Besides. i. p. 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. An essential singularity is really an ambiguity rather than an infinity. in the sense of rigorous convergence tests. Consider e. If the expansion is mathematically well defined. Expansions of this type arise. although the first few terms indicate a convergent behaviour.e. the series converges.e. divergent. Such expansions are strictly speaking. 102. The vast majority of perturbation expansions discussed e. for sufficiently large r. it has some finite.g. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. see e. Whittaker and G. as we have seen in Chapter 8. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless.g. integrated over) beyond its radius of convergence. In physical applications such cases are rare. z = a + i/3 with a — oo or f3 —> oo or both. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. Watson [283]. N. so that successive contributions to the first approximation do not invalidate this approximation. > 537 . E. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation.Chapter 22 Classical Field Configurations 22. i. Such an expansion parameter is usually known to be or assumed to be small in some sense.g. but asymptotic. It was therefore natural to search for alternative methods of expansion. in the context of quantum mechanics are not convergent.e. nonzero radius of convergence in the domain of the coupling parameter. i. W = e x p ( l / z ) .e.g.
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). A method which achieves this is in particular Feynman's path integral procedure.h2.e. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential.e. higher order corrections are of orders ±. the latter's topological properties if any. nature of the dominant approximation does not imply that this describes classical motion. and has led to insights which previously seemed unimaginable. There.h3. i. 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.. Many of these aspects are interesting by themselves .e. and in general this change is accompanied by certain constraints.h°. and the corresponding expansion is called the "loop expansion". This type of expansion is decsribed as "semiclassicar.. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". A further challenging task was the development of methods of quantization of theories which incorporate classical. such as soliton theory.538 CHAPTER 22.g. 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.. cnumber first approximations. On the contrary..h. The classical. 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. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). In the following we consider various typical examples. although this terminology is not precise. such > a series begins with a contribution proportional to (e. In searching for other means of expansion. However. cnumber. One therefore faces the problem of quantizing a system which is subject to constraints. These methods are often described as methods of collective coordinates. The fundamental extension of the method of canonical quantization (i.) l/h2. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. 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. i.
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. modified by going to imaginary time). . to keep this case in mind.g. or even a solution to a modified field equation (e. d^] = d^d^ . 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. vortices and kinks. a static (time independent but spacedependent) solution of the classical equations of motion. it is important for a better understanding of the considerations which follow. We shall not be concerned so much with the case of a constant first approximation. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". 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). may like to consult the article of D. V{</>) = m§0V + A0(<^)2. 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.22.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. 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>. Tong [272].V(4>). It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration.1) ' T h e reader who wants a highly advanced overview of instantons. 22. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. Thus references to field theory will be exceptional and hopefully do not irritate the reader. This means theories with field densities which are therefore infinitedimensional.2 The Constant Classical Field 539 and in any case deserve detailed study. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). (22. However. monopoles.
. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. 433.540 CHAPTER 22. 22. cnumber) solutions <j)(x) = 4>c = const.e. 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). V ) . Classical Field Configurations where d^ — (—d/d(ct). 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.g. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. as illustrated in Fig.£).e.1. V[<t>] m 2>o o Fig. p. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. £)) is bounded from below we must have Ao > 0. Felsager [91]. i. and hence here [d^ + m20}(l) = Xo(f (22. of the equation of motion? For these we must have all derivatives of <j) zero. m l + \04>*4> )$ = <). § (22. 22.2) We ask: Are there classical (i.1 Different potentials for different signs of m§. B.3) See e.S£/<90(x.
We can convince ourselves that <fic is the field associated with a state of minimized energy.5) .</>2)plane. mg < 0. i. The Hamiltonian density H is defined by the Legendre transform H[<f>.e. we have 541 0. TTIQ > 0. aq> For (b = const. H = —C = V. with 0 = 0. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution.ir]=7r<fiC. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion.2 The spontaneously chosen phase. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. the equation has this invariance.2 The Constant Classical Field For Ao > 0. 22. 22.e. not the solution x(t)).4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. 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. w A iP 3ml 5* "XT" (22. the only such solution is the case we wish to consider. But for Ao > 0. and so ft[<M=m6:(^* 1 + Ao(0*0)2. (22.22. Fig. TT = —J.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.
We examine this in more detail by setting <f>(x) equal to the classical cnumber configuration plus a fluctuation field n(x) which is again complex like <f>(x). e.3) obtained above. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. along the trough of V (which we can call a longitudinal direction).3). we set cf>(x) = <f)c + ri(x). where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. . This is the condition (22.7b) In Eq. (22.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.e. i. &H/d(fi = 0) if [m20 + ^>]</> = 0. Classical Field Configurations the density Ti is minimized (i.2. 22.2) and separating real and imaginary parts. (22.tp2. in a transverse direction. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. >° for m 0 < °> . Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.7b) the constant on the left and the coefficient of A on the right of Eq. Suppose we choose one such configuration.3ml = 2ml = ml + TA 0 A 2 = 0. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .7a) vanish on account of Eq.Then m\ m\ = ml + A0A2 = m2.e. the one for (3 = 0. i.6) Inserting this into Eq.e. V?) (22. (22.g. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . and partly by climbing up the parabolic wall on either side.7a) (22. as indicated in Fig. and we wish to investigate the behaviour of the field cf> in its neighbourhood. i.542 CHAPTER 22. (22.
2 The Constant Classical Field 543 We observe: The field ipi(x) has acquired a real and positive mass whereas the field ^2 0*0 is massless! We observe that the field ^2(2) is that component of ip which is directed along the trough of the potential. 22. has to be removed in order to allow a well defined perturbation procedure (i. Goldstone's theorem applies to fully relativistic field theories. then a massless boson exists. irrespective of the question whether the perturbation series as such converges or not. (3=0 Fig.e. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. 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.22. (22. t') = d(r . r'. 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 £. tangential to the classical path.3 The spontaneously chosen phase seen from above.r')S(t . of course. We can see the problem here by looking at Eq. i. 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. whereas ipi(x) is y ' " " " " " ^ . which.7b). t.e.t'). ^ = h eip / \ ^2 l . There the wave function with an associated vanishing eigenvalue is called a "zero mode". though not identical phenomenon. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. The elimination of the . Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. 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. the existence of individual perturbation contributions). Thus the Green's function is similar to that in electrodynamics.
£) in one spatial dimension. (22.* Potentials of higher polynomial order in $ than $ 4 can also be considered. Khare [17]. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. however. spinless field $(x. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. . and we choose the metric ?7oo = + 1 . F. N. We make one more assumption concerning V which. Zimmerschied and H. Later we wish to develop a perturbation series about a classical configuration of $. 7711 = — 1 ( 1 6 Minkowski manifold). W. We see therefore that the classical cnumber field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties.d^} = ~d^d^V[^) = ^2^(^j V[$]. 22. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). MullerKirsten [291]. Classical Field Configurations other components results from the vanishing mass and the constraint called the gaugefixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution.544 CHAPTER 22. J. Behera and A. Rebbi and G. C. To insure that the energy (see below) is positive definite we require V[$] > 0. A. instead we shall consider static and timevarying solutions in important models of one spatial dimension.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories.g. i. 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[^. *S. Lohe [179]. see e. Soliani [237]. equates to zero the coefficient of this wouldbedivergence. t Potentials of this type have been discussed by M.* We consider the theory (later: quantum theory) of a real.t In particular the sextic potential can be considered along parallel lines.e. we shall need only at a later stage.
13b) We are not interested in just any solution of this equation. (22. In a quantum theory the field $ is an operator defined on a space of states.12) where the prime denotes differentiation with respect to $ . . Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.12) we see that they satisfy the Newtonianlike equation <j. Since $(x. The other condition which we impose (and study in detail) is that of stability. the socalled ^theory = ^V[g$. such that J > — — — = 0. ^ ^ * + V'[*]=0. i. (22. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.§ Unstable classical configurations are also important and will be considered in later chapters. i.22.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.t) has an explicit timedependence it is a Heisenbergpicture operator. From Eq.13a) We write such fields <j>(x).l}.e. 1]. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. — m$ 1 — cos Im 1 I ^ $ (22. at least in the present case.11) (22.g] = \ v m For instance.«{x) = V'(<p). The field $(x. (22.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.e.9) with quartic potential is defined by ^V\g$\ = ^V\g*. (22. The first such condition to be imposed is that the energy of the solution of interest must be finite. 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). In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. (22.l].
. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.2\dx 2 ' 1 fd& + V[$].17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>.15) The zerozero component is equivalent to the Hamiltonian density H.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. dC • Too = r?oo£ + =r$ = C + vr$ = H. <9$ i.17) Since we have to integrate over all space.T = <I>.e. We observe here already that the integrand of the integral in Eq. (22. 22. (22.T^ = 0.14) where the overdot implies differentiation with respect to time t. n = & . The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. (22.c = <£z + . i.7r) where IT is the momentum conjugate to $ defined by dC ? r = .546 CHAPTER 22.e. HI tanh m(x>k) 9 Fig. the classical solution will therefore have to be such that this integral is finite. (22. d.4 The wellknown soliton of $4theory.
10) and (22.5. it is not difficult to find classical cnumber solutions 4>c to Eq.1 . m ( x .20) (22.21) This sineGordon configuration is depicted in Fig. In fact.5 The soliton of sineGordon theory. 4m . < = — tan [e >  ° ] 9 2 rrm/g — Fig. 22. Again XQ is an integration constant. 22. . i. (22.11).18) is depicted in Fig. in the case of the first. O±m(xxo) (22.4.x n ). > . one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. "The name sineGordon is derived from the analogous KleinGordon equation or theory. 22.18) Here XQ is an integration constant.22. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n".3 Soliton Theories in One Spatial Dimension 547 As we shall see below. the $*theory. Rubinstein [240]. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. — XQ).e. (22. See J. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> .777.13b) for the specific potentials (22. (22.
548 CHAPTER 22. 4m In tan tan 5 l/e±m(*xo)tan^£^)U.13b) which we can write d \W? = !"(•».9 2 4>{x) = i. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). and hence 4/71 9<t>{x. With we obtain g(/>(a. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! .11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I. or \(<P'? = V(4>)+ const. Classical Field Configurations The solutions (22. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. In the case of the sineGordon theory defined by the potential (22.o) =m7r.1 "»JtKxo) 9 1 . . <^>(x) = ± dx x=g4>/m \ / 2 ( l — COS X I Am dx lx x=g4>/m A / 4 s i n 2 ±1 In tan . (22.18) and (22. ± 1[tanh.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .* ( m' m 2 r)] ^2" .e.20) are obtained from Eq. Inserting the potential (22.
in the onedimensional Ising model. Then we have to demand that the functional derivative of the energy E(<p) be zero. SE((f>) 5cb(y) 0. for instance.e.23) See for instance R. (22.4 Stability of Classical Configurations The concept of stability can be complex. instead of doing this now. we obtain these expressions later (cf. (22. this does not require the action to be minimized. The expressions (22. The relation (22. (22. however. 22.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name 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. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian.21) for the energies can also be obtained from Eq. Suppose now that we demand stability in the sense of minimized energy.22.17). However.19). the others as antikinks.22) and that its second variational derivative be positive semidefinite. the curves rising monotonically from negative to positive values are known as kinks. A state of stability is therefore associated with minimized action and/or energy. Jackiw [141]. 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. the discussion of Bogomol'nyi equations) in a simpler way. i. .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) .
Thus the vanishing of the first variational derivative yields again Eq. (22.e.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. written down in Lorentzinvariant form. namely (p'c. Since it arises from the violation of translational invariance by 4>c.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration.28) . i. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. The original Lagrangian was. The Newtonlike equation (22. 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. with eigenvalue w2. if we apply the generator of spatial translations d/dx to the classical equation. a very general phenomenon which can be established as follows. (22.13b) retains the invariance although the solution (f>c does not. of course. the function 4>c(x + a). (22. (22. in fact. the integrated contribution vanishes. dx (22. in fact ip0(x) const. if 4>c{x) is a solution. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). it is also called a "translational mode". This is. = 0.550 CHAPTER 22. The eigenfunction 4>'c{x) is for this reason called a "zero mode". and the EulerLagrange equation retains this property. In fact.25). (22. we see that the latter has one eigenfunction. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22.13b).24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. its eigenvalues w\ > 0. i.e. dx2 + V"{4>) (22. i. Eq.e. in particular the invariance under spatial translations. Classical Field Configurations Since y ^ ±00. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space.a / 0 is not the same.13b) with classical solution 4>c.26) Comparing this equation with Eq.A.
32) = V'(</>). (22.e. (22. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero.20) are such configurations in two particular models. (22. lim [V(<f>c) + const.27). this integral has to be finite.31) Inserting this into Eq. (22. (22. (22.18) and (22.33) x—>±oo . i.] = 0 or x—>±oo lim V{(/)c) = 0 for const. We can choose the potential V such that the constant is zero. / oo Thus for E(4>c) to be finite. = 0.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0.22.17) we obtain oo dx[V(4>c) + const. We shall see later that these zero modes lead to undefined Green's functions for the semiclassical perturbation expansion unless one or more suitable constraints are imposed. in fact.e. and that S[UC] is finite (or equivalently the energy). i.]. it is an expression like (22. We return to our considerations of stability.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. In the onedimensional case under discussion we can rewrite Eq.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. (22. Then the vanishing of the first functional variation implies that 5S = 0. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. and we have seen that Eqs. Thus TUC is a zero mode. As stated above.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). 0 = S'[UC] = S'[gUc}. We see therefore that the occurrence of these zero modes is a very general phenomenon. we are interested in classical cnumber field configurations of finite energy.
.34) jfc" = e^dv(l since d^k^ = 0.6 The minima of V in $4theory at x — ±oo. 22. Thus the classical energy or vacuum configuration is not unique. For a configuration to have finite energy.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. oo J—oo y The topological quantum number q is defined by 0.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. 22. (22. In the sineGordon theory > • defined by Eq. ± l . V{4>) = 0 at different values of </>). This number is defined like a charge in field theory by the spatial integral of the time component of a current. (22. the potentials have degenerate minima (i. in view of the antisymmetry of eM^.35) and it is seen that in this case q can be ± 1 or 0.552 CHAPTER 22. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. n = 0 . Classical Field Configurations V Fig. (22. .1). . the classical field (bc must approach one minimum for x — . . In the case of the 3>4theory we can define a conserved current k^ by (22.6.c o and another for x — oo. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory.e. » In the <fr4theory defined by Eq. ± 2 . The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22. .
11) are given by ( so that for x — ±oo: > — J. .7T (22. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. . in general. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . A property is. n = 0 .36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. zero is not a meaningful topological number. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . Tfl \ (22. and we call this "topological stability.37) for the onedimensional case here. We can see this as follows.22. continuous values. . ± l ± 2 . one may ask how topology comes into the picture.A2)2 + 0 except for A2 = 1. First.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. Again we see that E(\<f>c) is finite only for A = ± 1 .cos(27rnA)].38) 4 V(\<t>c) >m ~2[l . In the case of the sineGordon theory the minima of the potential (22. . 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. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. termed topological.
00 (22. i. its integral is nonzero only on account of nontrivial boundary conditions. MiillerKirsten [215]. The inequality is saturated. pp. (22.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. 22. We observe (a) the topological current k^ of Eq. the energy is minimized.g. Eq.13b) or ^ ' ( * ) ] 2 = V(</>). 00 (22. (22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). (c) the time component k° is a spatial divergence.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (22. (22.e.5 Bogomol'nyi Equations and Bounds We have seen previously (cf.554 CHAPTER 22.43) **See e. J. 425 . H. Eqs. (For a broader view we consider briefly in Sec.31) to (22.17)) f 00 > 0. (d) the topological quantum number iV arises in a completely classical context.9 the foregoing arguments in a higher dimensional context. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. Thus with one space dimension higher there. if oo / dxy/2V{<l>)<f>'{x). Classical Field Configurations involves the difference An = rioo — n.^ = N. the factor A = exp[ix(#)] varies over a complete circle). . W. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. 22.41) dxy/2V{$)4>'{x).42) Thus the energy has a lower bound given by the right hand side of this inequality.40) (22.428.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents).
^ Clearly it solves the second order equation or (22.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi. Bogomol'nyi [27]. (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.39). (22. (22. 555 (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.22. n E .n > 1. Eq. for instance. 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. We discuss this in the following Example.5 Bogomol'nyi Equations and Bounds We observe that (cf.21).44) This first order equation which saturates the inequality (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. B. Here (V(c/)) > 0.19). 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— .
Solution: Consider a string along the xaxis as described.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .e. Goldstein [114].2 V dt + k[(f>(xn+i.7. 22. 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. all being identical.t) 4>{xn. We can now write down an expression for the energy of such a system of (say) n pendulums.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis. Initially the pendulums are suspended freely in the gravitational field. } n= — oo roc —> / L J .ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. t)).g.t) <j>(xn. Fig. rotate) only in the (y.e.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i.7 The pendulum analogy. Assume that the pendulums can move (i.g. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22.556 CHAPTER 22.cos</>(x„. Construct an expression for the energy of such a classical system and its Lagrangian.46) **See e. z)plane and only such that their elastic strings remain taut. from the yaxis as shown in Fig. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. with mass m and connected with short strings to neighbouring bobs. so that oo > u. ka —> K. 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) measured e.t)\2 +mgr{\ . Assuming that the pendulums can move (i. rotate) only in the (y. H. Classical Field Configurations Example 22. . 22. Chapter XI. xn+\ — xn = a) 00 a —> dx.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. At equidistant points xn = na along this string we attach strings with pendulum bobs. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .
t) is the fluctuation (field). the topological quantum number. wk would be imaginary and so the factor exp(—iwkt) in rj(x. Hence we set ${x. is therefore also called "winding numbed. (22.49) would imply an exponential growth in the future t > 0 or in the past t < 0.49) exp(iwkt)ipk(x). 22. (22. If some w\ were negative.7 linking the pendulum bobs). (22.50) This equation is identical with Eq.e.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. t) of Eq.t): {w With the ansatz r)(x. Eq. (22. and so would invalidate the procedure which assumes that rj(x. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. We can understand what stability means in the present case.13b) we have for small fluctuations rj(x. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. of the continuous curve in Fig.22.t) is a small .25) obtained earlier. which represents an expansion in terms of normal modes. It is instructive to obtain this equation by yet another method. The equation is called stability equation or equation of small fluctuations. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). ' Equating coefficients of exp(—iw k t).6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22.13a) and (22.t) = ° (22'48) (22.47) into (22.t) = <l>c(x) + r}(x. (22. (22.47) where n(x.t).48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). 22. Inserting (22.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. This number.
Inserting (j)c of (22. we obtain V"(<t>c) . If that is necessary.50) has the general form of a timeindependent onedimensional Schrodinger equation. For x — ±oo the quan> tity acting as "potential".52) tan 6 := e±m^xo). The equation of small fluctuations (22.e.A \ Inserting <f)c of Eq. This means that the stability is not universal but only local in a certain sense. we shall see later that the fluctuations n(x.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ .558 CHAPTER 22. In fact. (22.e. V"((/)c(x)). and need 9 = tan" 1 j e±m^Xo) 1.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. Classical Field Configurations fluctuation. In fact. 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. i. the zero mode. cos46» = 2cos2(2(9) .18).20). we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. since we shall need the explicit form of V"{<f>c(x)) we calculate this for the two cases: (a) ^theory: Here "<*> = ^  i n*) = 2m 2 (l£^)*. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. .1. Thus for stability w\ must not be negative. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. V"(<b) = 2m2 + 6g2(f>2. i.2 m 2 + 6m 2 tanh 2 m(x .
. ^+(0) d 0.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. (22. Those of the other case are complex and periodic at infinity. (22.e.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 . which satisfy the boundary conditions V>(o) = o. — XQ)] cosh[±m(x — XQ)]: 1 . Consider the equation fd2 dz2 . We can therefore construct solutions of definite parity. even and odd solutions. A  *(* + !) cosh 2 z tp(z) = 0.ijj^ respectively.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .* For such a potential the spectrum can be b o t h discrete and continuous. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.55) The differential operator is even in z. ip+. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.22. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.53) = m(x — (22. 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. i.2 (22.
(22.4 n ( n .57) Then the equation for x{z) 1S (22. A * An = .60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.Z ) + ^(A + Z2) = 0.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. .e. 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.( 2 n .61) Discrete eigenvalues are obtained when the series (22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . i.I) .58) = (1 + sinh2 z)~l/2x(z).59) Setting (22.60) terminate after a finite number of terms.e.e.Z ) 2 . when in the even case n ( n . i. + l2) = 0. A ^ An = . i.( 2 n + l .l2 = . and in the odd case n+\)U+\l)+\(\ i. 2n < Z.560 CHAPTER 22.Z)2. (22.
N = 0. We can see the continuous spectrum of Eq. (22.62) where even (odd) iV are associated with even (odd) eigenfunctions. From (22.51) into (22. w2 = Am2.10) and (22. 3m 2 .1: w2 = 0. 2 .50) we obtain with z = m(x — xo) tp(z) = 0.2. the continuum is w > Am2.11).62) we see that the equation possesses discrete eigenvalues w —2 .54) by setting X = k2 Then 1 . i. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .e.N<1.e. m i.2Z = .63) we see that the continuum starts at A= 2 2^ = 0.. (22.Ny.( 2 .u2)—^V 2 dio and ip(z) = elkzV{to).64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.\=Kl2 From (22. (22. d w = tanhz. (22.... m N<2.1. °2 i. for N = 0.22.is normalizable provided 2(n + .e.65a) _dz2 cosh 2 z l=2.j < I . We can now derive the spectra of the stability equation for the potentials (22. (22.6 The Small Fluctuation Equation 561 and the function ip. (a) ^theory: the equation Inserting (22.
65b) dz2 cosh^J. This property can. We know that 4>c is a monotonic function. .=1A=^_. In the $ 4 theory we saw that (cf. also be deduced from d(j)c/dx. 22. . d .8 and we see it has the shape of a typical ground state wave function. The zero mode is depicted in Fig. m d .53) into (22. and — as expected — this eigenfunction is even. so that its derivative is nowhere zero (except at ±oo).2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. .27)) . . of course. Thus 4>'c(x) has no node. It is now particularly interesting to look at their wave functions. (22.66) We see that this is a nonvanishing function for any finite value of x..e. i. Eq. Thus either spectrum contains the expected eigenvalue zero.. .50) we obtain with z = m(x — XQ) the equation (22. the zero modes. for N = 0: to2 = 0 and the continuum starts at w2 = m2.8 The zero mode as typical ground state. %(x) Fig. Classical Field Configurations (b) SineGordon theory: Inserting (22. 22. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22.562 CHAPTER 22. N<1. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue.
d . 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 . 1 o rrr tanh x — .z 0 )]oo = 8 . without quantum corrections). i. Finally we consider the normalization of the zero modes.= E(<f>c).e.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ). Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.70) . (22. .t a n h x . . (22.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. .69) The quantity MQ here is called the mass of the kink (it is the mass of the soliton solution only in the classical approximation.3 . ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. (22.6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to .oc (22. — XQ) OC Hence the conclusion is similar to that in the previous case.22.e. .
we consider E(<P) which we also write as (22.g.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. *The considerations of this section extend in part beyond quantum mechanics and into field theory.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . ddx {V4>f + v{4>) (22. H.T(4>). to underline the fundamentally different nature of topological and nontopological finite energy configurations. They may help. for if such configurations exist in a realistic theory with three spatial dimensions. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. One can have different types of scale transformations — see e. Affleck [5] who uses in his Eq. I.72) with T{$) = I' ddx]{V(j))2 > 0.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . H. I. ^For basic aspects of scale transformations see e.> 4>a{x) = 4>(ax).564 CHAPTER 22. f G . i. P. however. P(</>) = f ddxV{(f)) > 0. . Olive [113].41) for the energy of the classical configuration but now with respect to a ddimensional position space. J As a first example we consider the expression (22. (2. This question has important consequences.2d. Goddard and D. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces.4) <j>(x) —• a<f>(ax). Kastrup [145]. it is important to understand their physical implications.g. A. Derrick [68]. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0).g. Classical Field Configurations 22. (22.e. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. *See e.
.x2. the second derivative is positive) only for d = 1. (22.76) we see that E(<f>) can be minimized (i.V(<P) (22.75) Since both T and P are > 0 and d is integral.73) we see that this implies (2 .xd). Consider first the Lagrangian density £[&M = \{d^)(d^) . In fact. x = (x1. we set 0 dE(<pa) da J a = i (22.79) In order to emphasize that (f> — (p^ in any direction. i.e. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>). (22.22.71). d = 1 minimizes the energy. (22. i.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. a=l (22.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. to be finite the field <P ~^ 4>oo £ Mo for x — oo. 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.d)T(<f>) = dP{4>).78) For the energy of a static configuration. since d2E(<pa) da2 2(2d)T(<f>). (22.. we write • = lim J r . Next we consider more complicated cases. 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}..74) From (22.e..80) .\ eMo(22.e.
therefore. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00.and sineGordon theories we assumed $ to be real. . one set can be mapped into the other.e. Thus in this case S^1 is a disconnected set consisting of two points. i. i. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. and. —00 or equivalently S = + 1 .1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . These observations are summarized in Table 22. 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).e. —1.38). x = 00.1. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity.566 CHAPTER 22. If we go to two spatial dimensions (d = 2). that we have a set of disconnected classical vacua (</>oo><^oo. (ptx = <f>oo. of • course. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . —00.e. — 1). is a connected region. as is the case for the constant solutions. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. S1. 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 . Then each point at infinity is characterized by a different value of the polar angle 9. We see. i. are the same.81) and (since d = 1) S = ± 1 . (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. Table 22.
if cj> were real and position space twodimensional. * n reach any other <^(oo) by simple phase shifts.e. We see. phase transformations 6Q ^> 9Q + 59. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.• M0. VMx)) with (0(oo) = linvxx) (p(r.22. (22. This will be "nontrivial" in the above sense only if 0oo is independent of 9.g.e. i. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. Similarly. assuming invariance of C under the transformations of the group SO(3) in internal space. therefore.2. to have topologically distinct classical configurations. —00 or S = 1 . i.7 Existence of Finite Energy Solutions Suppose. On the other hand. .83) . i. Thus.3 (^real) v{4>) = / (22. 9) would depend only on r = x and hence the problem would be effectively onedimensional. This can be achieved e. i. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. for instance. 00. by allowing <fi to have three components in some "internal" (isospin) space. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. map into. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. the field must map position space into field space at infinity.e. cj)(r. we have w e ca {Mx)}> and 1 3 1. if <p is real.e. we would have only two points at infinity. i.e. that if the problem is to be topologically nontrivial. If position space were onedimensional.82) Now Mo := \ 4>(oc) ^24>2 = a2\. the one point S = + 1 . Thus. i.1 .e. $(r) : Sl .
circle. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. A) + T{A) + P{<j>). going once around S%~1 (i. The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du. $ — > > $ exp{i Xlo=i Pa.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. the subsequent discussion requires only the general form of the Lagrangians. + i ( D " f t W ) a .2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. the appropriate rotation. Example 22.V{4>a]. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space.2. Georgi and S. Olesen [220].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) £(</>.a M J . $ — $ + 27 > • T in sineGordon theory. Nielsen and P. $ — $exp(ia) in a U(l) symmetric theory. A) = T(4>. i. sphere. through each point at infinity). Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. .e.e. we cover <S^_1 n times. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. i. invariance under replacements $ — —$ in $ 4 theory. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. Glashow [108]. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. it is effectively a surface (and so a continuous set) S2. so that details of these models are irrelevant. B.</>] = ~G^Gai.„ where G r = 9 M £ .568 CHAPTER 22. 11 H. L..3. If. a = 1.. . the winding number is n. % .e. The symmetry of the potential (and Lagrangian) — i.e.
it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.Ax) dX Since T((f>.d)T(4>. A.e. J 4 ) + A 4 .d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).87a) C?M„(x) — \2G^{\x). We now consider scale transformations with A £ R. we have {^/»)A^ (22. ^ A ) = A 2 .84) is positive semidefinite and so must be finite by itself. (22. Thus we set dE(<fix.e.<^[x) transforms like a vector. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.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'>. > Since T>IJ. A A ) = A2dT(^. (22.d)T(A) .i T(AX) = X4~dT(A). (22. (22. some contribution must be . negative.84) in this way one finds that the quantities involved transform T ( ^ . A). i.T{A). (i)^ 2d T Considering the contributions of (22.dP(4>) = 0.d T ( A ) + AdP(</>). i. = 0.22.A).. Solution: Each contribution in (22. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . l j ddxGijGai] > 0.84) and investigate the existence of finite energy classical field configurations.86) (22.7 Existence of Finite Energy Solutions where T{<j>.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0.d T ( ^ . >• A„ x ( x ) = AAM(Ax). Mx') = E with a scale transformation x — x' = Ax. A) + (4 .87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. and so B ( ^ A .89) P{4>) are positive semidefinite and not all zero. <0 >• <£A(X) = <KAx). (22. (2 . T(A) = f ddxV(4>) > 0.
H. (22. is temperature dependent and changes sign at the critical temperature Tc. Thus such a contribution could counterbalance T{4>. ft V .t) such that  ^  2 describes the density of Cooper pairs. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. A) or P{4>) for d = 3. 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. In the following we consider a related theory.92) The expression ^ ( x ) ! is known as the "order parameter.e.89). of course. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. . 22. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. a = a(T) = a T —T c .90) and we would have to add to Eq.** e.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory.89) the contribution (4 — d)T(4>). J. Ginzburg and L. Landau [111]. D. This case is. (22. One can thus define a scalar field or wave function ty(x.89). (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. We then have a situation of the potential like that described in the context **T. Tchrakian and H. W. a > 0. 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). R. T^{<t>x) = \4dT(</>). (22. (e) For d > 5: In this case it is not possible to stabilize Eq. L. (22. We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. Skyrme [253]. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory.570 CHAPTER 22. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). MiillerKirsten [268]. Classical Field Configurations (a) Thus for d = 1: P(4>) must be present but T[A] and so the vector field would not be required to satisfy Eq. 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.g. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. exemplified by soliton and sine—Gordon theories. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D.
i.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. 22. Sec.2): For T > Tc the overall potential has a single well (minimum). (22. setting S£.e. i. Example 22. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material.2) £(^. Variation as before. ^(00) = a (22.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . (22.^ A n (this is called "minimal coupling of the electromagnetic field^ with vector potential A to the Cooper pair field ip in a nonrelativistic treatment) where g — 2e~ is the charge of the Cooper pair.94) to check the dimension) 1 = I—a.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. 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. The parameter 7 serves to put the minima of the double well at V = 0. from x < 0) means — as is familiar from electrodynamics — replacing V by V .A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4.93) J' (22. Then the static energy density becomes (cf. Clearly the length (see Eq. but for T < Tc it has a double well shape. 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. is a measure of the distance from the surface of the superconductor to where .(tp) = 0.22. assuming there is no superconducting material in this domain).e.
yJ—a/[3.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B . The density of the Cooper or super current is given by the typical expression of a current. In the Maxwell equation See e. h [^DV'V'DV]. 22. i. W. MiillerKirsten [215].572 no superconductor CHAPTER 22.M.g. exhibits a completely different behaviour — in fact.9. as indicated in Fig. 450. We can see this as follows. .9 Behaviour of ip and B in the superconductor.e. the order parameter \ip\ attains (approximately) its asymptotic value. (22. 2mi Setting < / > \P\e vp B = ViA. The magnetic field on the other hand.e.96) ipH. H.98) where M is the magnetization resulting from atomic currents. i. Classical Field Configurations superconductor (x>0) Fig. V x M = j s .e. J. 22. p. (22. (22. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ). i.
M i. There are two types of superconductors. Also dD/dt = 0 in a static situation like the one here.div grad" and V • B = 0. These vortices therefore carry magnetic flux which is necessarily quantized . Hence 0 = V x H = V x ( — B ..100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. (22. 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.e. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. In the onedimensional case we have B{x) = B(0)ex/XL. Using the relation "curl curl = grad div . This is zero in the present case.e.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. m xi A B . those of the socalled type I and those of type II.5 to 10~6 cm). K i ™*. V x B = [i0V x M = /lois. i. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value.99) Inserting (22.L B = O. (22. Approximating p by its equilibrium value.22. this becomes A B = no V x j 5 . m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . and V x (V x B) = / J 0 V x j s .
e.36).. Thus the asymptotic field </>(r. in fact a periodic function of the polar angle with period 27r. We can think of (e.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. 'See the remark in H. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. De Gennes [67]. *A.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. 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. de Vega and F. Classical Field Configurations (see below). Abrikosov [2].574 CHAPTER 22.101) (p(r. x—>oo Thus at spatial infinity </» has the value of a zero of the potential.102) In the onedimensional soliton case of Eqs. (22. In the two models of Example 22. i. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). In the NielsenOlesen model defined in Example 22.9). (22.e. . where exp(ix(0)) is an arbitrary phase factor. G. A. we must have that lim V ( H ) = 0. (22. 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). of two points on a line. J. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI.g) twodimensional space as being bounded by a circle at r = oo. 22. in other words. Schaposnik [69] after their equation (3.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. The existence of these vortices has been confirmed experimentally. the space consists of two diametrically opposite points on the circle.9) ^e^ '>\<poo\. Finiteness of E requires \<t>\ = \<fioo\ on this circle. See also P. i. A.37).
22. Naturally an integer cannot change continuously. We may define this as « = ^[X(0 = 2 T T ) . X is called the domain of / and y its range.e.118)). 0 < 0 < 2TT. W. (22.10 States of winding number 0. topological group with the usual multiplication as group operation. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. Thus smooth deformations *See D. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. Sl := {zeC\\z\ = 1}. .1.X ( 0 = O)]. commutative). Fig. The 1sphere is the unit circle in the space of all complex numbers.* 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 22.102) is called the exponential map of S1 onto S1. (22. Misner [96]. i.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n.10.e.22. abelian (i.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. defines the orientation of this line element. In mathematical language S1 is therefore a compact. Finkelstein and C. The two theories we considered above in Example 22. Then a continuous function of (j> on [0.1). Suppose <j>. 2n] remains continuous if the function assumes the same value at the two endpoints. The map A : S1 > S1 defined on S1 C R2 by (22.9 Introduction to Homotopy Classes 575 By a map or mapping / : X —> Y of a space X into a space y we mean a singlevalued continuous function from X to 1". The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. 22.
576 CHAPTER 22. The gauge field required in the asymptotic limit (22. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below).e. For that reason n is called a "topological invariant?..g.104) so that d2x hJ Since A has direction e#. r^oo 1 dx(9) n r o0 (22. We observe that since time evolution is continuous. Thus two homeomorphic spaces have equivalent topological structures.. one hole.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). we required that lim r—>oo w o. The winding numbers n are such topological invariants. i.e. if it has one twist as above or a hole. i. elements in the same class have the same topological characteristic. we can write 1 d h J ee rde w At h r^oo .g..) remain the same under homeomorphic transformations and are therefore called topological invariants. no hole. i.i'ldx(0) 9A Thus for (22. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr.e. The different kinds of connectivity of these topological spaces (e. connectivity. the winding number must be a constant of the motion.105) is described as "pure gauge" since it is determined entirely by the phase x(#).105) the corrections falling off faster than 1/r. finite. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. In other . (22. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e.
A field configuration with one loop (or twist).105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. (22.106) J g j r d d g Thus the magnetic flux is quantized. the group space of (the phase transformations of) U(l) continuously. cannot be deformed continuously into one with two loops (n = 2). Ffiu. if A = V x . is also zero at r = oo. for B / 0 . i. However.e. (22. cannot be pure gauge everywhere. An example is the mapping H(t.108) map the spatial circle S^.9 Introduction to Homotopy Classes words. it is determined by a gradient (l/r)d/dd. into S i .107) The fields <f>(r.l]. the higher order contributions in (22.109) A W(0) = e^ W. we can consider transformations which describe continuous transformations of a configuration with fixed n. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . going once around S^.e. Therefore. with n = 1. we have B = V x A = curlgradx = 0. It follows that the field Fij. i. However. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. (22. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.<b rd9—4^ = n.105) insure that B ^ O . the gauge field. i.9):=t(f>£\9) + (lt)(f)W(9) with te[0. 9){^ and hence the functions (22. gauge transformation of a static abelian field is A > A' = A + Vx 577 and we know that the The field which is determined entirely by V x is therefore termed "pure gauge". the number of flux quanta 2irh/g being the winding number n. Of course.e. we can go n times around S^. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).110) .22. the vector potential A.
111) as a map from the space described by 9. (b) symmetry and (c) transitivity.113) . H. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).g.1]. 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. the unit interval I = [0. The map H(t.110) could be written down for configurations with different winding numbers. But this type of combination is excluded sind H(t.578 CHAPTER 22. Since the passage of time is continuous. We can consider (22. it is clear that this is also a homotopy.e.T.lQSl^S^ with H(0. Hu [134]. 9) is an equivalence relation. Dittrich and M. time cannot make a field jump from one homotopy class (see below) to another.9) = <fiW(9) and H(l. f T h e standard texts on homotopy theory are N. Classical Field Configurations which is continuous in the parameter t and is such that H(0.e. The relation (22. Steenrod [261] and S. these two maps (fin from S^. (22.e.e. i. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. Thomas [270]. A superposition of classical configurations with different winding numbers does not minimize the energy.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). Solution: As stated. the phase factors of (f)(9). 8) = cfinl\9). we have to show that the map satisfies the three properties of (a) identity. i. Reuter [77]. i. (b) symmetry and (c) transitivity.8) itself must be a configuration with definite winding number. like <pn (0). e.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). A brief introduction can be found in the very readable book of W.9) = ^°\0). n. i. (22. S%.111) In principle an expression like (22. i. Example 22. to the internal group space of £7(1). In fact. Two curves which can be related to each other in this way. We verify these in the following example.3: Homotopy an equivalence relation Verify that the homotopy map H(t.0) = $\6). — S\ > (22. H. H(1.9) is called a homotopy (emphasis on the second syllable).e. and the space of t. n' ^ n. (fin (9) in the above example.
these groups are then called homotopy groups (cf. i. ' . Returning to the maps \(x) : Si ^ S1^ (i. (b) Next we check that if 0<°) ~ <p^. 9) = <f>W (0). then </>W ~ <t>{0)• We define an H(t. since and tf (1.9) = . .t. being the continuous map connecting </>(°) with <j>(1'>. 0).9) H(0.0) = </>(0) is continuous there.* 22. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). 9) = H(0.e.0).10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t.9) = <f>(0\6) and H(l. which therefore demonstrates that the homotopy relationship is transitive.22. (c) We can demonstrate the property of transitivity as follows.e) tf2(2il. 0) = ^ (0) and 6(1. Hence <f>(0^ ~ <p 2 ).0) for for 0<t<. below). Then H(t. In the above case of 7ri[[/(l)] we know from our earlier considerations that every equivalence class is characterized by a specific value of the winding number. so that H(t. Homotopy classes are sometimes also called Chern—Pontryagin classes.10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the onedimensional sphere S1. U(l)). <^(2). as they are called. at t = 1/2 both i?i and H 2 give c/^1'. 9) connects <pn (0) with itself continuously.e. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. 0) = 4>W (0). On the other hand H(0. . (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. negative or zero). is such that H(0. <f>W . 0) = <£{2). i<t<l. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. We now define the function H ( f 0 ff .e. we see that of the set of these maps some are equivalent and others are not. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. Moreover. \ M)\ H!(2t.ffi(O. a class being a set of equivalent maps. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. 0). 9) := H(l . i? is a homotopy connecting </>(°) with <^>(2'. is isomorphic to) the set of integers Z (positive. 9) = H(l. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). Hence the set of all such maps breaks up into equivalence classes or homotopy classes. 0) = tf2(l.
T (22.g. Appropriate maps.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. In order to really appreciate how topology comes into the picture here one should understand (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. . p. 22. 52.) why 7i"i[S2] = 0.11 the circle drawn with the continuous line is the circle S%. In the diagrams^ of Fig. Rajaraman [235]. n=0: t=1: t=1/2: i. S 2 the surface of a sphere in a threedimensional Euclidean space. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. which is to be mapped into another circle Si indicated by dashed lines. 22.580 CHAPTER 22.115) Varying t € [0. Sketch of idea demonstrating that nilS1] = Z. 22. T < 9 < 2TT. and then introduce some of the related mathematics. How can we see this? We first sketch the idea.e. Fig. to make the result plausible.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.11 The trivial n = 0 map allowing shrinkage to a point. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. but TTI[S2] = 0: S 1 is a circle.
For winding number n we have the mapping (i.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. d (22. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. (22. One can prove that the expression (22.117) in ±. 22. We make another observation at this point.117).I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. the transformations^ (the result may be looked at as 8n of expression (22.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.h. 22. 2 the maps are wound around the dashed circle.e.s. of (22. i.e. 0 < X(P) < 2vr.h. It is seen t h a t in these cases the continuous curve can be distorted into t h a t of case (a) only by cutting the dashed circle — thus these cases do not allow a continuous shrinkage to a point. (22.12.e. Coleman [55].22. i. and hence the winding number.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. In the cases n = 1.117) 2^ d6X(x)\1(x).s. The maps for these cases are illustrated in Fig.116) In fact from this we obtain (see below) 2  n and this means r. to the case of winding number zero. . is invariant under infinitesimal (continuous) deformations.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).
the cyclic group with two elements (since 50(3) is doubly connected).e. 1 1 See N. 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)=0.8. Thus there is only the trivial map into a point.e. Steenrod [261]. only one homotopy class. the identity class (the group space S2 is simply connected. . Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. there is a theorem which says thatH TTg(Sn) = 0 for q < n. 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. In fact. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. an integer modulo 2) i. whereas 7ri(50(3)) = Z2 (i. Sec. 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. i.582 CHAPTER 22. 15. Classical Field Configurations Under this transformation n changes by 5n Now.e. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W).
different from literature. our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. 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!). We consider such a theory now with action S((p) depending on the real scalar field 583 . This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. and in particular the exponentially small level splittings of the double well potential.e. in such a way that the factors appearing in the result are exhibited in a transparent way. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.Chapter 23 P a t h Integrals and Instantons 23.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. Since we have not yet introduced periodic (i. 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. 23. finite Euclidean time) classical configurations (which are nontopological).
! " ^ j = ^ 2 . (23. note that here the mass of the equivalent classical particle is taken to be mo = 1. Fig.e.2. 2 2 4 ™ = ~. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. Gildener and A.1 The double well'potential. Patrascioiu [110] and Chapter 18.1.1)) and hence E0 = mh/y/2.v{4>) <?VV m4 __ 1 . 23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O .584 <)>{t). and ^2(0) : m* .2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. in particular E. (23. Chapter 6) 1 2 . "For comparison with literature. i. 23. n = 0.2) Fig. The Hamiltonian of the simple harmonic oscillator (cf. Path Integrals and Instantons S(<f>) / dt \i>2 .* /•tf CHAPTER 23.
23.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. as we did in Chapter 18 — but as we saw. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). we have the Lagrangian L(x. but there is tunneling between the two minima if the central hump is not infinitely high. must have definite parity as in a physical > situation. </. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. Fig. e. W(0) = 0.g.2 The wave functions of the lowest states. 23. 23. i.2 — and we can deduce from this the boundary conditions. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator .e. We can develop a perturbation theory around either of the minima. This tunneling affects the eigenvalues. this alone does not yield the level splitting. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. and in the above 4> corresponds to x._(0) = 0. The states of the system here must be even or odd under x — ±x.In mechanics of a particle of mass 1 with ~ position coordinate x(t).x) = x2 V(x).
e.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).586 CHAPTER 23. (23. dr We observe that Eq. The transition amplitude for this. classically forbidden domains then become "classically" accessible.tf\ . i. Then (0o.  . d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. i. (23.e.<h. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g. Equation (23. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o.U) = J'D{<f>}exp[iS(<l>)/h] (23.2. But quantum mechanically it can tunnel through the hump from one well to the other. SE. r . dr m+™ (23. (23.2 m 2 ( l m z (23.2 V'(<l>) = . i.e. For small h we can attempt a stationary phase method.4) J <bi =—(An rTf (ITLE = I JT.6) d*l dr 2 . n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.7) so that (23. Euclidean time T = it.8) V(4>) + const.7. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. is finite if we go to imaginary.0o.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. In the semiclassical path integral procedure that we want to use here. (23.3) for tf — oo.5) and iS = dLE d dr[d(d(f)/dT)} i.£'(0) = 0.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.9) .e. dLE = 0. that given by Eq.3).
In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".e.3.10) where TO is an integration constant. i. Thus we know the solution from there which is the kink solution (22. i. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. this means the total energy of the particle in Euclidean motion is zero. 23. and hence kinetic energy gained is equal to potential energy lost. we see that the equation is the same as the classical equation for a static soliton (i. from fa — —rn/g. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p).18).e.e. The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description. is depicted in Fig.2 Instantons and AntiInstantons 587 In fact. . Fig. — (23. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. 23. the classical path. in the present case (T — To) = m tanh m(r — TO). If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below).23.3 The instanton path. Its trajectory.
'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. (23.7) admits another configuration which is the negative of (23. 23. Using Eq.].12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. we obtain lim f+ OO.13).Tj) oc exp 4 m3 m^2 (23.^ .7"/! ). (23.4).' .13). We also observe that the expression is singular for g2 —> 0. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. This antiinstanton configuration rises monotonically from right to left as shown in Fig.TJ T fro. the constant must be zero.8) we obtain T f SE dr[2V((j)c) + const.11) For this quantity to be finite for TJ = —oo.14) But this is not yet enough. thus indicating the nonperturbative nature of the expression. (23.10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23. Inserting the result (23.13) We observe that this expression is identical with the energy of the static soliton of the 1 + 1 dimensional soliton theory which we considered in Chapter 22.10) is given by (23. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .r 0 ) ' (23.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .13) into Eq. The configuration (23. Tf = oo. Path Integrals and Instantons The Euclidean action of the instanton solution (23.4.5). The classical equation (23. but with the same Euclidean action (23.588 CHAPTER 23.
(23. e.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.TO far apart and T'Q 3> TQ. and an antiinstanton localized at (say) TQ.15b) is not exactly a solution. / > Looking at the instanton (23.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). 23.5. The terminology is motivated by analogy with the phenomenology of particles: Thus like a proton has electric charge + 1 (in units of the fundamental charge e) and the antiproton has electric charge —1. from the sketch or from the variation of SE. in the case of (23.4 The antiinstanton.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig.TO) + ^ c ( r . The sum (23.15a) we have a simple superposition.g.5.g. Clearly we are here interested in configurations of type (23. 23. .Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . #(r) = </>C(T . b u t if TQ.15b) since we require configurations from vacuum to vacuum. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g. 23.23. 0c(r) = &(T).r0).15b) a solution of the classical equation? Is it a topological configuration — if so. (23. as we can see e.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.15b) has the shape shown in Fig. 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. 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.15b) In the case of (23.
Ignoring exponentially small contributions the Euclidean action of the configuration (23.T 0 ) } ] .15a) is not exactly a classical solution. we see that in the domain [—oo. 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 .r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. T\] with LECIT = 0. Path Integrals and Instantons '. 23. Finally. g cosh m(r — TQ) (23. vanishes as T\ —> ±oo (of course. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. of course. #(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 .16) . at oo the variation is <<> = 0). In fact. it also does not exactly stationarize the Euclidean action.2 m ( T o .e x p { .15a) is a configuration which starts from 4> = —i^/g and ends there.CHAPTER 23. since (23.5 Superposition of an instanton and a widely separated antiinstanton. 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.15a) can be argued to be zero. 5^ In order to answer the second question we observe that (23.
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).
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.T')dT' (23.63b).64) as above and consider the determinant of the expression. ^This is the method explained by B. are summed over all values from 1 to n.N{T) Finally we add the following remarks concerning the discretization (23. 192.67) i=l v ) = lim exp V In ( 1 + ' L K(T. .66) •Dr) .T2) I exp I K(n. r ' ) is to be evaluated at r = T ' .e. In the limit S —* 0 .63a). = 1 52 K(n. (23.TI) K(T2.lnJV(T)}] N(T) N(T) (23. Felsager [91]. r i ) + .j. we obtain exactly the nonvanishing terms on the right of Eq.. (23.n) i=l N lim exp £*(T. H. so that (with T *T) 1/2 •Dz N(T) (23..68) D above.n) o K(TJ. [44]. the kernel K(r. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. Martin [43]. .64). 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.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: . T. n )d? (23.64) Inserting the explicit expression for the kernel given in Eq.n) K(Tj. (23. X(TI.T1)dT1 + J dn J dr2.7 Y. (23. Thus in our case one of the kernels is zero and we have to decide what we do when.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits..exp • /_:T JV(T')" dr' = e x p [ . as in Eq.n) z=l x (23. where a f 1 See R. whereas the determinant arises only once in the original determinant Dn.64) D (23. the value of the determinant is half (or arithmetic mean value) along the diagonal.a { l n i V ( r ) . The Fredholm kernel is in general — as in our illustration above with a Green's function — a quantity with a discontinuity between the two constant integration limits.Ti) i=l : exp \ £ K(T. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.23. Thus D„ = l + 5 ^ K ( r i . This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). 2! . p. We observe that if we discretize Eq.TI) 0 K(T2.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. i. Cameron and W. Thus the factor a inserted above has to be taken as 1/2.4 Field Fluctuations 607 This equation represents a set of n linear equations which has a unique solution provided the determinant Dn of the coefficient matrix on the right does not vanish. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) ..
represents a constant translation.59) the endpoint conditions »)(±T) = 0 and evaluate the integral.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this. Eq. N(T') . I2 and using induction one obtains (cf.— .70) and this now requires also integration over the endpoint coordinate z(T).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. (23. Eq. With partial integration (in the second step) we obtain (since (cf.59)) + f(T)] = — f dz(T) f°° dae^W+f™.l + £7t) «i (23.608 CHAPTER 23. (23. (23.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. .( .e—*0 In.^ .Z i . Evaluating Ii.L dT{ \ (i(r) exp T>z 1 N(T) + 1>n exp N(T)\2 ' N(T) J a2 N2(T) 2 JV 2 (r) We perform the functional integration as in Chapter 21. T + N(T) / T' = T JT dr' N(T') . rn+i = T. Path Integrals and Instantons Example 23. n J=. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T).57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T .69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf.73) .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. (23. In every integration / dzi the contribution €7.Z ( T ).N(T) .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . In = dzi dz2 • • • dzn e x p .71) Hence (with h = 1) l = j . Briefly. Solution: From Eq. (21. f(T) = N{T) J^ dr'^^z(r') (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).2o) 2 rc^O. and hence may be ignored. (23.(T) + /_ dr — .
£Q) — aTo]dTod£o. 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. This result verifies Eq. Eq. £n = ~2 dTT](T)7]n(l (23. (21. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. N. ^E.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.74) where we reinserted h. Ignoring the remaining metric factor with the reasoning explained > between Eqs. 1/2 T W 2 (r)J (23. 11 L.10) and (21.12). . / d£o <9r0 5[F(TQ. i.^ 23. (23. J co § dr Recall that / f ^ doce^ = yfVfp. Gildener and A. a = const. (3.4.11).23.59) and agrees with that in the literature.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode. D.e. / — / / We started off with V(T) = ^CnVniT).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.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. Patrascioiu [110].. Faddeev and V." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. j / dTT]n(T)r]m(l r . (23. Popov [89].
(23.e. We now impose (i. (23. i /*oo (23.e. at r = ro.78) <t>ij) = Y.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). (23. Then cn = ~2 / dr(pc(T)rjn(T).e.79) (23.^n{r). / dTm{r)n{T).g. (23.80) i r or •i /*oo dr7?o(r)0(r) = 0. . e.£o)/a. Thus.75) enforces the collective coordinate ro to be given by F(ro.J drm(r)Mr + T0) (23.^O) with dependence on ro which the replacement (23. In analogy to (23. this constraint implies that we go to the subspace which is orthogonal to the zero mode.75) requires.610 this means that CHAPTER 23. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. The number £o is the component of the fluctuation rj(r). The delta function 8[F(TQ. i. Path Integrals and Instantons _^ ^ w . In the integral of Eq. We establish the function ^(To>£o) with the following reasoning. where L = in + cn. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. i. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r).81) the rdependence is integrated out and hence the integral does not provide the function F(TO. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T). We consider the following integral / ( r 0 ) := . demand validity of) the constraint £o = 0. in the direction of the zero mode. (23.83) **The collective coordinate in the present context is also called "instanton time".81) 2 / dr»7o(r)0c(r) = . (23.76) we set 4>c{r) = ^CnVnir).£Q) — aro] contained in Eq.
Zo)y/SETQ We now define A FP '• =0 • (23.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o. so t h a t with F(TQ. . We have AFP = /F(T0.91) " F o r more discussion of this point and related aspects see also B. Felsager [91].£O) OTQ = = £ " .84) (23.75) 1 = AFP f dT06[{F{rQ. . jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.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 ) ./ drri0(T)<pc(T) + PJ (23. We evaluate this quantity as follows in the leading approximation.85) PJ W i t h Eq.(23. (23.23.89) Here the factor App is called the FaddeevPopov determinant. 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) . p.86) = Jdrrjoir^Mr F(TO.<t>c{r)} .88) ro0 Then in Eq.y/S^ro] (23.y/SET0 = 0. (23.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). (23. 158.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).87) l/M) (23.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23..^) .
using the relation (23. the "static energy" invariant. if ^ + d£o = 0. 0 d r (23.93) where AT = TT — 00. 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. if the coefficient of 770(70) vanishes. .93).94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.91) into Eq. P J AT (23. 7 Thus there is no fluctuation along the direction of the zero mode in the sum. i.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . i.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. Inserting (23. /(TO) ^/SE~T0} V~SET0 c d£. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). Consider. 770(7").e. tangential to (J)C(T)) leaves this and hence the square of this quantity.e. in d(j)c(To) + dr/fa).e. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.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. ~ There is a quick way to obtain the result (23.0AFP8 4o H P . (23. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i./ dT0AFP.90). 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.612 CHAPTER 23.e. P the relative sign being of no significance. =~ (23. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To).
This very important intermediate result is the contribution of a single instanton or kink configuration to the p a t h integral (ii has been defined by (23. (23. (23.98) . We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton. We determine I'Q by observing t h a t this differs from IQ of Eq.43)).60a) for I0 and Eq. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.96) lim . (23.93).23. (23.4.60a) in not involving the zero mode contributions. We observe t h a t the result (23.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.55b).4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23. The source of lack of similarity is attributed to the fact t h a t the contribution of a single instanton to the p a t h integral is not enough.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. (23.96) does not yet allow a comparison with (23. Since w\ is for finite T given by WQ of Eq.4 Field 23.97) is a constant still to be evaluated.A T A F P i " o e x p T—>oo p where I'o n^O. 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.
i n s t a n t o n contributions As already mentioned.2 m T e x p 2 irgh ' T»oo (23. (23.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. 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. . Although we choose p2 = 1.r).102) *For comparison with the parameters of E. Gildener and A.4>C exp h (23. (3. ^See E. the approximation is then known as the "dilute gas approximation".20). the result becomes h lim 2 m r ^ ^ 1 e . Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 .32).TiMiy+'Mr «>.101) 2T > oo (cf.5 I n s t a n t o n . 0c M. Gildener and A. Patrascioiu [110]. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. *For the validity of the calculation the instantons or kinks have to be far apart. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .* Finally we rewrite the expression (23.614 CHAPTER 23. (23. they have to introduce the inverse of that factor on the far right of their expression (3.55b)) w0 = v0e VQ 23.100) in such a way that it exhibits the crucial factors of which it is composed.e. i. In view of their normalization of the normal modes. This means we are concerned with configurations like* 2n+l d(r .100) We observe that the dependence on the normalization constant p cancels out. ~ rKo')e{Tx . Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.4.a n t i . their Eq. we drag the factor p2 along for ease of comparison with literature.
(l>i) (23. (3) (23.r / 3 .<j>i) .Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.4>I) ld2SE SE(c/>i.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23. The factor ( . Proceeding as before we now consider first in analogy with Eq.fo) = Id2 St SE(<Pf.l ) i + 1 in the sum of Eq. T3 = T) d(f>l((pf. . .Tf\(p2. are varied. (n = 0 implies the single instanton solution). . r 0 ( 2 ) = 0. ( 2 3 .T2\(f)l. (23. 7 i = .105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations. the total Feynman amplitude is given by (^/.<t>ff + h .<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f.H)2 + .<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .23. here (f>2. (23. Taking these configurations into account.T2}((t)2.Tl\(j)i.6.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).2 T / 3 .104) Considering the case n — 1 of a kink with one additional kinkantikink pair. 1 0 6 a ) and SE{(J>I.4>i. Consider (with T 0 = T.36) (/.<t>2)eSE{<t>2. this means we have to consider the amplitude (23.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.4 Field Fluctuations 615 for n = 1. r0(3) = 2T/3.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. (23.106b) SE{<t>l.<Pi)2 + • • • . 2 .Tl}{4>l.r/l&. T2 = T/3. 23.4>l)eSE{4>l. Since these endpoint configurations.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.r 0 ( 1 ) = .
e.3 9 L 3m 2 and 9SE((/)) ± .2d2SE ~c 4>i d(f)i e x p SE^f.108) where the last expression is obtained as follows.2J. each of which implies a multiplicative factor 7.106a). each of the three intermediate kernels contained in Eq. dT 9 70(oo) V m J sE(4