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. even though the expansions are mostly asymptotic. With the growing importance of models in statistical mechanics and in field theory. 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. basic postulates and standard applications.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. Thus this first part . In the first part. but arise also in recently studied models of large spins. its mathematical foundations. like periodic potentials. With various techniques and deeper studies.e.g. one problem being that there are few nontrivial models which permit a deeper insight into their connection. we recapitulate the origin of quantum mechanics. screened Coulomb potentials and maybe singular potentials. be treated to a considerable degree of satisfaction perturbatively. has not always been understood well. numerous problems could. the aforementioned exactly solvable cases. whereas for the calculation of discrete eigenvalues the Schrodinger equation. with degenerate vacua) were first and more easily derived from the Schrodinger equation. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. again point the way: For scattering problems the path integral seems particularly convenient. but also about complex energies that he encounters in a parallel course on nuclear physics. in Feynman diagrams in quantum electrodynamics). Excluding spin. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. in fact. Chapters 1 to 14. These basic cases will be dealt with in detail by both methods in this text. However. 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. that is the Coulomb potential and the harmonic oscillator. Diverse and more detailed quantum mechanical investigations in the second half of the last century revealed that perturbation theory frequently does permit systematic procedures (as is evident e.
The most prominent examples here are the double well potential and its inverted form. Chapters 15 to 20. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. 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. The potentials with degenerate minima will be seen to reappear throughout the text.e. The solution of this case — however in nonperiodic form — turns out to be a prerequisite for the complete solution of the Schrodinger equation for the singular potential 1/r 4 in Chapter 19. and their eigenvalues.XVII deals mainly with standard quantum mechanics although we do not dwell here on a large number of other aspects which are treated in detail in the longestablished and wellknown textbooks. i. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. In the second part. 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. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). and the elliptic or Lame potential — here introduced earlier as a generaliza . 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). we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. Thereafter the concepts of instantons. periodic instantons. 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. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). Using perturbation theory. which is presumably the only such singular case permitting complete solution and was achieved only recently. The following chapters deal with the derivation of level splitting formulas (including excited states) for periodic potentials and anharmonic oscillators and — in the oneloop approximation considered — are shown to agree with those obtained by perturbation theory with associated boundary conditions. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. the chapter thereafter deals with Yukawa potentials. 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.
this comparison on a transparent level being one of the main aims of this text. it is. anharmonic oscillator potentials on a comparable level with. except that these are no longer of hypergeometric type. from our precise reference to unavoidable elliptic integrals taken from Tables). the Mathieu equation. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model.g. This puts Schrodinger equations with e. however. for instance. All results are compared with those obtained by perturbation theory.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. we 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 application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. 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. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. quartic. for instance. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. and whenever available also with the results of WKB calculations. In fact. 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. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. an approximation whose higher . 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. We then illustrate the relevance of this in the method of collective coordinates. cubic). we can make the following observations. such as limiting procedures. This method is therefore more complicated. With a fully systematic perturbation method and with applied boundary conditions.
Throughout the text some calculations which require special attention. D.g. we also consider at various points of the text comparisons with WKB approximations. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). also for the verification of results.Q. not the least for permitting a more detailed and hopefully comprehensible presentation here. Their deep involvement in the attempt described here is evident from the cited bibliography. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. Park (Masan). Whittaker and Watson [283]. J. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). Instead a glance at a nearby footnote provides the reader immediately the names of authors. In writing this text the author considered it of interest to demonstrate the parallel application of both the Schrodinger equation and the path integral to a selection of basic problems. the references referred to are never cited by mere numbers which have to be identified e. The author is deeply indebted to his onetime supervisor Professor R. thereafter University of St. at the end of a chapter (after troublesome turning of pages). formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. Liang (Taiyuan). e. For ease of reading. The line of thinking underlying this text grew out of the author's association with Professor R. H. like E. W. T. in particular Professors J. B. Since this comparison was the guideline in writing the text. 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.XIX order contributions are difficult to obtain.* H. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. Dingle (then University of Western Australia. Andrews). which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. as well as applications and illustrations. Watson [283]. .g.g. N. MiillerKirsten *In the running text references are cited like e. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. D. As a rule. Nonetheless. B. 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. with the source given in the bibliography at the end. Whittaker and G. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. K. shows him that each such formula here has been properly looked up). are relegated to separate subsections which — lacking a better name — we refer to as Examples.
very few explain in this context what he really did in view of involvement with statistical mechanics.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). Instead. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators.Chapter 1 Introduction 1. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien.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. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Kirchhoff's law in thermodynamics says that in the case of equilibrium. We do not enter here into detailed considerations of Planck. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). The best approximation to such a body is a cavity with a tiny opening (of solid angle d£l) and whose inside walls provide a diffuse distribution of the radiation entering through the hole with the intensity of the incoming ray decreasing rapidly after a few reflections from the walls.1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. the amount of radiation absorbed by a body is equal to the amount the body 1 .
(1.3 4 J s. the formula (to be explained) u(u.626 x 10 . The parameters k and h are the constants of Boltzmann and Planck: k = 1.1 Absorption in a cavity. 1. It was found that one expression agreed well with observations in the region of small A (or AT). u(u.1) containing the constant h by treating the radiation in the cavity as something like a gas? By 1900 two theoreticallymotivated (but from today's point of view incorrectly derived) expressions for u(u.T) = dv3eC2U/T. T)du is the mean energy density (i.38 x 1(T 23 J K'1.e. i.2 CHAPTER 1. energy per unit volume) of the radiation (i.l ) where x = ^ = ^ . h = 6.e. How did Planck arrive at the expression (1. kT kXT (1. and the other in the region of large A (or AT). T) were known and tested experimentally. These expressions are: (1) Wien's law. Introduction emits.1) c is the velocity of light with c = u\.T) = 2*?£(?)kT. Black bodies as good absorbers are therefore also good emitters.2) and the . radiators. v + dv in equilibrium with the black body at temperature T. i. 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. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v.1) ' y Here u(v. y J c 3 \ex . and then measuring the increase in temperature of the heat bath. Fig. (1.e.e. Let us look at the final result of Planck.. In Eq. A being the wavelength of the radiation.
T) u(i/.1) as approximations. where the Fig. 1.T) = av e6"/Ti' where a and b are constants.3) are contained in Eq. e xhv.3) Ci. (1.e. "small" (i. When Planck had found this expression.T) 2^^kT. We see. which are indistinguishable) among the N indistinguishable oscillators at . we obtain: u(i/.T) = 2^C3T. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom.1) in regions of a.2 Distributing quanta (dots) among oscillators (boxes). he searched for a derivation.47TZ/ 2 {x small).2) and (1. Considering Eq. C3 being constants. 3 (1. but — here the discreteness appears properly — only in elements (quanta) e. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons". exp(x) ~ 1+x) and "large" (exp(—x) < 1). C2. and thus over a discrete number of admissible states.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. This is the point. that the formulas (1. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U.1. . in the first place Planck had tried to find an expression linking both. (1. Here W is a number which determines the distribution of the energy among a discrete number of objects. discretization begins to enter. and he had succeeded in finding such an expression of the form u(v. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. (x large). To this end he considered Boltzmann's formula S — klnW for the entropy S. Indeed.
with the quanta represented schematically by dots as indicated in Fig. i. as in most other Tables. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. N * oo. 940. M.6) Fig. 1. Agreement with Eq. v + dv. one obtains (cf.g.e. (1. formula 8. 1. (1. S. p. (1. Ryzhik [122]. e.1)!P! (1.g. . h = const. and the second law of thermodynamics ((dS/dU)v Example 1.2) requires that e ex is. The Stirling formula or approximation will appear frequently in later chapters. Then W is given by w = (N With the help of Stirling's formula* {N + piy.2.T)dv.7) *See e. i. there not called Stirling's formula.e. u(i>. Oberhettinger [181]. Gradshteyn and I. e = his.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. with riydu — 2 x —w—dv. We now obtain the energy density of the radiation. W.4 CHAPTER 1. . Introduction temperature T. p. I. U{T) being the average energy emitted by one oscillator. as for small values of e).343(2).3.3 Comparing the polarization modes with those of a 2dimensional oscillator. We visualize the iV oscillators as boxes separated by N — 1 walls. = 1/T).4) IniV! ~ JVlniViV + O(0). Magnus and F.
3 (as for ideal conductors).. The number with frequency v in the interval v. In terms of A we have u(X.2.3.1 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..3. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.7).T) has a maximum which follows from du/dX = 0 (with c = vX).1).1..2..i = 1. rii = 1. i.1. so that the derivative of u implies (x as in Eq.e. 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. as indicated in Fig. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.UJ2/C2 + K? = 0. 2 2 2 r2 L K — 7T n .. where^ 2 [2KUY . L for i = 1. .965 and xmin = 0.8) This is Planck's formula (1. We observe that u(v. KT = I J . Then L^j = nrii. so that .1)) The solutions of this equation are ^max = 4. (1. '''From the equation I \ JW . We obtain the expression (1. nvdv per unit volume. dj\l dM .2. v + dv.. . as in electrodynamics. (1.7) for instance. is given by .UJ = 2KV.V 2 ) E = 0. We obtain therefore u^T) = Unv = 2^fJ^—i.T)dX = ^ehc/*kT_idX.3. (1.
We therefore examine such contradictions next. but one can expect an analogy. Introduction he 4. (1. that it is easier to consider first the case of T = 0. which had also been known before Planck's discovery. n = 0. ra = 0. 2 . that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1.l. which can not be eliminated without a different approach.10) Thus here the socalled zero point energy appears. Later it was realized by H. One expects. 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 .10). Lorentz and Planck that Eq. However. Thus we have a rather complicated system here. the behaviour of the system at zero absolute temperature. . of course. We are not dealing with the linear harmonic oscillator familiar from mechanics here.1. A. x — oo) the mean energy vanishes (0 < U < * > oo). such a procedure leads to contradictions. 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=—. we then have at T = 0 independent oscillators. which can assume the discrete energies en — nhv.8) could be derived much more easily in the context of statistical mechanics. . . If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.—v <L9) We observe that for T — 0 (i.e. One might suppose now.965K = Const. .2 (1. i. which did not arise in Planck's consideration of 1900. that of an oscillation system at absolute temperature T ^ 0. and from which the constant h can be determined from the known value of k.e.6 The first value yields AmaxT = CHAPTER 1. Since temperature originates through contact with other oscillators. This is Wien's displacement law. .
for 2 degrees of freedom.= .1: Mean energy of an oscillator In Boltzmann's statistical mechanics the entropy S is given by the following expression (which we cite here with no further explanation) S = fcln W. .1)! .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV.1. 1.i n . In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente")..InP!] ~ kN The second law of thermodynamics says 1+ 7 ln 1 + ( 7)~7ln7 \au)v For a single oscillator the entropy is s = S/N.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom. with Heisenberg's discovery of the uncertainty relation.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.1 ' u~ e .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > . 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".1)! . we obtain T V S =fc[ln(TV+ P . around 1926.In .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. so that 1 f ds\ T — \dUjy .2 Contradicting Discretization: Uncertainties 7 Example 1. {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) . kT/2. d .ln(TV . i.k dU 1+U\ f ln(l T + U\  U U k ( e .
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
The total time derivative of u can therefore be rewritten with the help of Eqs.4) as the generalization of Eqs. since . One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq.^ ^ j . (2.Pi) is contained implicitly in the canonical variables q^ and pi. whereas the velocity requires observations of space coordinates at different times.1).. i. Dirac [75].24 CHAPTER 2.3). which can be observed directly at time t. (2. (2. Hamiltonian Mechanics is wellknown.1) as d .n n. A. M. Goldstein [114] remarks at the end of his chapter VIII. H. Goldstein [114].qi(t) qi = hm f . It suggests itself therefore to consider more closely the properties of the symbols (2.e. one obtains the Hamilton equations* • OH .P. S »(«.t). The following properties can be verified: *See e.2) as This equation is. this expression is simply a functional determinant.1). . (2. as mass times velocity. One can verify readily that Eq. . (2. (2. All functions u(qi.) = £ [wm + WiK) = £ [WiWi . the equation of motion of the observable u. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. Rather. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. Pi has to be considered as an independent quantity.g. in analogy with Eqs. Compared with an arbitrary function f(qi.pi) of qi. the entire timedependence of observables u(qi.p% are therefore again observables. We can therefore consider Eq.Pi. A system consisting of several mass points is therefore described by a number of such variables.4) contains as special cases the Hamilton Eqs. chapter VIII. x^fdu. ^As H. du \ x^/dudH du dH\ .2) If we have only one degree of freedom (i = 1). which all together describe the state of the system. the standard reference for the application of Poisson brackets is the book of P. chapter VII. qi(t + 6t) . (2.1). (2. 6t>0 5t Real quantities which are directly observable are called observables.
These are {Qi. in other words without any reference to the original definition (2. the Poisson bracket {A.e. we obtain the fundamental Poisson brackets.B}C or C{A. B2}.B}.B*}. A}} + {C.B} = {B. i. 25 (2.B}* (4) product formation: {A.C}. {C. it is irrelevant whether we write {A. Bx} + a2{A.B}C + B{A. B}} = 0. Property (2. that the very general Eq. {qi. which combines the Hamilton equations. like here. B} is completely evaluated.5d) is useful in calculations.4). {B.6) We can now show. Since Eqs.3) of the Poisson bracket. The original definition of the Poisson bracket will not . If. for example. {A. that of the linear harmonic oscillator.Pk} = 5ik.5b) (3) complex conjugation (note: observables are real. (2. Later we shall consider noncommuting quantities. (2. (2. we wish to evaluate {A. where A and B are arbitrary observables. a i S i + a2B2} = ax{A.5c) The first three properties are readily seen to hold. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. then the ordering is taken as in (2.5a) (2.5d) = {A*.6) give the values of these. (2. If we evaluate the Poisson brackets for qi. 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.2. C}} + {B.Pi.5e) = {A.2 The Hamilton Formalism (1) Antisymmetry: {A.5d) above.BC} (5) Jacobi identity: {A.A}.6). B}. (2.Pk} = 0. {pi.Qk} = 0. As an example we consider a case we shall encounter again and again. but could be multiplied by a complex number): {A. (2) linearity: {A. (2.
yPot3 + •••• (2. These are transformations qi—>Qi = Qi(q. (2.p.. (2.9) and (2. (2.p. q = {q.\(p2 + q2)} = l({q.4) we have for u = q.9) Similarly we obtain from Eq.q2}) = = 2i{q. From Eqs. (2. Pi—> Pi = Pi(q. "4' = q.p}p + p{q.H}. (2.8b) We insert (2. (2.26 CHAPTER 2.H}. from which we infer that q(t) — qocost + posint. and Pi are ordinary real number variables and that H(q.11a) 'q' = q.12) . (2. and so q = q. we consider as Hamiltonian the function H(q..t).t). (2.8a) and use the properties of the Poisson bracket and Eqs. According to Eq.7) into (2.11b) (2. Hamiltonian Mechanics be used at all..p\) P. In the evaluation one should also note that the fact that g.qot2 .. or q(t) = qo+ Pot .6).8b) p=q.7) q = [q. Then we obtain: (2.10) we deduce q = p = q. and P={p.p) is an ordinary function is also irrelevant.10) In classical mechanics one studies also canonical transformations.p.P2} + {q.p) = ±(p2 + q2). Since constants are also irrelevant in this context.8a) (2. q + q = o.
pk} = 5ik.P.2 The Hamilton Formalism 27 for which the new coordinates are also canonical.p <> Q.2 below contains a further illustration of the use of Poisson brackets. i. i. for which Hamilton's equations hold. {Qi. (2. _ax p__dK_ (2.15a) °raS {AiB}Q. that (dropping the subscripts therefore) {Pi. B} of two observables A and B in terms of either set of canonical variables. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.1 we verify only Eq. {qi.e.Qk} = 0. and Example 2. which transforms qi. i. chapter VIII. Example 2. .3 deals with the relativistic extension. {PhPk} = 0. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.P is canonical in the sense defined above. as {A. (2. those at a time t — 0.p = {A. {QhPk} = Sik.15b) The proof of the latter invariance is too long to be reproduced in detail here but can be found in the book of Goldstein. (2. *H. (2. i. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.B}q. Example 2.B}q.3) we can now express the Poisson bracket {^4.P.12) Qi—>qi = qi{Q. Pi^Pi=Pi(Q. Eq.Pi back to their constant initial values.2.qk} = 0.15a). verify that the Poisson bracket of two observables A and B is invariant.e. (2.13) We write the reversal of the transformation (2. . P) exists.t).e.* Hence in Example 2. i.e.P provided the transformation q. {qi. which means that a Hamilton function K(Q. Goldstein [114].B}Q>P.Pk} = 0. Of course.e.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).t).P One can then show that {A.15a).14) With the help of the definition (2.
B}QIP. . Mittelstaedt [197]. we have —Et. we obtain analogously dA dQk' Inserting both of these results into the first equation.p^ "Pk. V V °Qk dPk dPk dQk J E x a m p l e 2.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. their product Et — qp being relativistically invariant.15b) dA {Qk.Pj}q. (2. we obtain as claimed by Eq. p. {A. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u. Replacing here A by Q and B by A.F}r ~di d~E ~ d~E~8t Consider F = H(q.s F 5 7 r ={A. dA {Qk.+ dQ.p = J2 d~Fk Replacing in the above A by P and B by A.p^v dP.q/c) and (E. Solution: Relativistically we have to treat space and time on an equal footing. Qk}q. OQk Pk}q.28 CHAPTER 2.15a) r„ ™ v . Thus whenever q and p are multiplied.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential).Qj}q. 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 „ „•.E(t). 236.3: Relativistic Poisson brackets By extending qi.P^pr.p) .p^— + {A. § See P.A}q.p(0) = poSolution: The solution can be looked up in the literature. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo.( 9A = {Pk.pc). Thus we extend q and p to spacetime vectors (t.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. we obtain dA {Qk. (2.B}q.p = dB dA dB \ rA 1 {A'B^ E b r s r .A}q. § E x a m p l e 2.
i. But now we consider a system whose phase space coordinates are not known precisely.14) and area A of the phase space ellipse given by Eq.to. We consider first the a priori weighting or a priori probability. and E as a function of t). (1. (*.Po. also boundary points of one domain are mapped into boundary domains of the other.p) to G\{q.e. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . 2TTE UJ and hence g oc — . Since Hamilton's equations give a continuous map of one domain onto another.' ±=j1V c \dtj' u^. rrr .2. H(q.p) .3 Liouville Equation. Let us assume that at some initial time to the system may be found in a domain Go(q. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.1 Liouville Equation. it is the equations of motion which lead from Go(q. and the space spanned by the entire set of canonical coordinates is described as its phase space.16) For example. g oc AqAp.p). This probability is evidently proportional to AqAp. of course. but partial derivatives of F do not vanish. c dt 2 2 w^ du du dr .. so that Go(qo.. Probabilities 29 (This is.t0. I .3. Of course.t).. Then . 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.y.17a). . dudH s l u . and at time t > to in a domain G\(q. one obtains Gi with q = q((lo. We distinguish in the following between two kinds of probabilities. i.t). since H is expressed as a function of q and p.)•'=(*)**£.e. . du q\ P = —. numerically zero. in the case of the linear oscillator with energy E given by Eq. to . dE .dt y/l  qZ/c2' 2. (2.p). Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.p) around some point qo. du .po. g. p= p(qo.Pi are the coordinates of some mass point m.p). (1.3 2.po is a point on Go.Po. we have A = (p apaq — J .. if qo.E(t)}r = Hence at Relativistically we really should have clu/dr.._.po) = Gi(q.
Thus g must be independent of t.. dpdq and similarly dt = — Ap.14) t o (1. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. J0=0 Hence g oc 8n2IdE.)curve for constant E and $ is — as may be seen by comparison with Eqs. Show that the total volume of phase space covered for constant E is 8n2IE. the (pg.eded<t> = 8TT2IE. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' .p^. Hence from the difference: — .4. (1. in view of this independence it can be expressed in terms of the conserved energy E.30 CHAPTER 2. — dp A A — In(AqAp)s = — + — = Example 2. dt dt Aq d ( A p ) dt ' Ap Here d(Aq)/dt is the rate at which the qwalls of the phase space element move away from the centre of the element.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. dt dq and with with Hamilton's equations (2. Je=0 . dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. Example 2. and hence g oc %ir2IdE.— = — Aq.4: Liouville's theorem Show that A q A p is independent of time i. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin.15) — an ellipse of area § dpgdp. Pi 2 / \.1): d ./. = 2TrIEsm6.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . dt dq dq dp dp d2H dqdp d2H = 0. Example 2. this has the same value at a time to. as is demonstrated by Liouville's theorem in Example 2. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. which means.
Probabilities 2. dn — p(q.Po.e.3 Liouville Equation.e.p.p.17) F ^GT^^ 0 Fig.*). i.p.t)dqdp = N.18) With a suitable normalization we can write this / W(q.t)dqdp= 1.t). i.2.q + dq.19) Thus W is the probability to find the system at time t at q. d d 1 P = JJdqidpi. Since W has the dimension of a reciprocal action.p. 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. whose positions in phase space are characterized by points. by jr = P(9»P. W=p(q. it is suggestive to introduce a factor 2KK with every pair dpdq without.3. The total number of systems N is obtained by integrating over the whole of phase space. however. Thus dn is that number of systems which at time t are contained in the domain q. (2.p + dp.1 The system moving from domain Go to domain G\. (2.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.Thus we assume a large number of identical sytems.20) .p. (2. leaving the basis of classical mechanics! Hence we set / «w)^ =. (2. 2.
We are now interested in how n ov W changes in time.e. t + dt)dqdp — p(q. i.a t dt + q+dq. In order to derive this equation. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. 2. 2. p(q.e.p.t)dp{ jt Q. t)dqdp p(q. 2.P dt p(q + dq.p. The equation of motion for n or W is the socalled Liouville equation. * q Fig. — if vq(q.e.p. In doing this.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. p + dp G P O q qndq Fig.p. we consider the domain G in Fig.p. The number of points at time t + dt in domain G. i.3 The region G.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.p) and vp(q.32 CHAPTER 2.t)dp( — I . that in our consideration no additional points are created or destroyed.2 The ensemble in phase space.p) denote the velocities in directions q and p — p{q. i. we take into account. how the system moves about in phase space. t + dt)dqdp.
p.2.4) shows that p and u satisfy very similar equations.p(q + dq.p.19). p. p.P *' % = iH>P} ( 2 .p.t) dt p(q.p(q + dq.22) The generalization to n degrees of freedom is evident: The volume element of phase space is .t + dt)dqdp — p(q.t)vq{q.p.p + dp) dp = or  However.p(q. Comparison of Eq. (2.W(q.21) we can also write dW(q.p)dtdp +p(q. t)dqdp = p(q.p(q.p)dtdq .p.t)vp(q.t) dt {H(q. = K(9.P)]. . (2.t)vp(q.p + dp.P. With Eqs.t + dt) . vp(q.p) = q = g^.20) and (2.p.p) dq p(q.p. Probabilities and thus p(q.p.t)vp(q.t)} with JW(q. 33 Dividing both sides by dqdpdt this becomes p(q.t)vp(q.p) . t)vq(q + dq. put differently.p)=p = 3H —.p + dp)dtdq.P)]K(Q.p). so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . .3 Liouville Equation.p) .p + dp.21 ) This is the Liouville equation which describes the motion of the ensemble or.t)^ = 1. . (2.21) with Eq.p)dtdp . dH Mq. (2. (2. .t)vq(q.p.p(q.p. the probable motion of the system under consideration.t)vq(q + dq.
24). for i<«>!/«<*p>"w>(^)".that the density or probability W(q.4 Expectation Values of Observables Let u = u(q. The first and most immediate possibility is — as indicated .4). that dtj\y v if qo.p)W(q.4). 2. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.24) since the total derivative is made up of precisely the partial derivatives contained in Eq.po are the initial values of q. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. and hence that equal phase space volumes contain the same number of systems. t) depends explicitly on time t (if determined at a fixed .t)(^J.p (cf.p) be an observable. Equation (2. Thus in Fig.e. We now inquire about the time variation of the expectation value (it) of u. Example 2. (2.p.p.p). (.e. since no systems are created or destroyed. With Eq. We shall see that we have two possibilities for this. 2. p+dp. We define as expectation value of u(q. i. q+dq.34 CHAPTER 2. (2. (2. p. We deduce from the Liouville equation the important consequence that ^ M = 0.p) the following expression: (u)=Ju(q. i. (2.26) we described the time variation of the observable u(q.1 the area Go is equal to the area G\.24) implies that p is a constant in time.
p) assumes certain values) that depends explicitly on time. We expect. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. (2. Then Eq. at t = 0.t). Sec.t) = = W(g(q0.24): W(q.27) becomes !<»>  /«(*P>!"W>(^)" (2.29).Po. since W oc p is constant in time according to Eq.8.^ Solving the equations of motion for q.Po.po.e.po.p){H(q.i).p0. we have Qo = g(q.t). so that Q = g(qo. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t)W0(q0.p. However. (2.f(q0.29) The distribution of the canonical variables is given by W(q.t).p) = uo{qo.p. and hence u(q. (2. Observables 35 point in phase space).t). and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.33) po = f(q.0) = W0(q0.p0)(^^J.30) p = f(qo.0) = u0{qo.t).34) S e e also H. Goldstein [114]. (2.0) =u(g(qo.28) = Ju(q.p).t) W(q0.W(q. (2.t)}(^y.p.4 Expectation Values.po.t).p. 8.22).t).p. we can also employ a more complicated consideration.p) = u(q. (2. We verify this claim as follows.t).32) In this expression the time t is contained explicitly in the observable u(q. where we used Eq. (2.Po) at time t = 0.po. Reversing Eq.Po.Po.31) i. of course.p. ie.t). (2.t). .f(qo.Po. 1 (2. (2. Thus we can write.p. we can express these in terms of their initial values qo.po.2. that (u) = (u)0.
t) dB__ (du\ °P \°PS P=f(qo.36) and (2.32).\„.0) u(q.p.p. we obtain an expression which is different from that in Eq.po.POit)i p=f(qo.(2.p). Hamiltonian Mechanics so that on the other hand with Eq.t).u{q.t) dp M du\ ° q 'P=f(qo." V " " ™0' > • • » « » 0 .t).t).p0) = W(q. (2.f(q.Po.t). (2.29) q = g(g(q.M 27Th i . (2.36 CHAPTER 2. Eq.39) Substituting this into Eq.p.32).t).31)) W0(q0.P0.f(q.Po.36) = (2.t) = (2 30) (2.f({H(q.t) v{g(<j(q.t).t)J(q. Eq.37) and (2.28).t) (2.t) . (2.p)W{q.p)})q=g(QO. (2.30) Moreover.t). (cf.t).po.Po. d dt {U)0 = dt.) f / u (q0.p.po)(~ .p.f(q.o.30): duo(qQ.p.p). [ du0{q0. (2.p.t). p = f(<j(q.f(g(q.t) 0H_ 0 q =  {{H(q.Po.p. (2.p. Taking now the total time derivative of (it)o.P0. we obtain («)o = / u{q.t)\—^\ = (u). i.t) d t _ (du\ \ d<i dq / P=f(qo.Po)(^^)n.PO.t).t).p.u(q.p).38) we obtain £<«)„ = .P.t).t)W "(q0. Inserting Eqs.f(q.po.P. (2.P)}).35).e. / ^fdq0dpc .(dq0dp0\n We deal with the partial derivative with the help of Eq. (2.25) into Eq.0)=u(q.t).t). (240) . (2.t) M f 9u\ V °P / P=f(q0.=9(. .p. (2.30) u0(g(q.37) as had to be shown.) Wo(qo.35) Inserting these expressions into no we obtain uo(qo.
41) and (2. (2.p)) is attributed to the probability W(q.25) and (2.42) the relation d .t) dqdp\n 2irhJ (2.p.p).p). In the first case.W(q. However. uW} = {H. This timedependence is described by the Liouville equation dW(q. The considerations we just performed demonstrate that we have two ways of treating the timedependence: The explicit timedependence can either be contained in the probability W or in the (transformed) observables.p..28).t) of u assuming certain values q and p.p. we obtain zero after partial integration of I and hence from Eqs.t)}.p){H(q.p)}W(q...28). . so that ~d~t <«>0 = J{H(q.u(q.p. (2. Observables 37 Here we perform the transformation (2.2. . we obtain {H.(2. described as "Schrodinger picture".43) ——uW = 0.W(q. u}W + u{H.4 Expectation Values.28) and (2. dt <«)o v f u(q. The phasespace integral of a Poisson bracket like I (2. ^ „.45) in agreement with Eq.p.t) dt {H(q.41) This expression contains {H.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). Alternatively we could deduce from Eqs. Consider {H.p).42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. u}W instead of u{H. x r „. lim ——uW — 0.t)} .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.W} in Eq. the observable u is treated as a function u(q.34) and use (2.45) the Liouville equation. and the time variation of (u(q. (2.p). from the properties of the Poisson bracket.31). W}. (2. / d q d p \ n 2irh J (2.
i. for which the axioms of a commutative group apply.rt. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. .po) is assumed.38 CHAPTER 2. This formulation deals with observables u representing physical quantities.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. called u Heisenberg picture". as the equation of motion of an ensemble or of a probability distribution on the one hand.p0.p)}. which describe the state of a system. and probabilities W.4) of an observable on the other. Thus we arrive at a more general theory if we define u and W with the following properties: (a) An addition is defined between the quantities." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.po. does not have to be commutative. Koppe [152] at the university of Munich around 1964. " This means: Only in the Heisenberg picture dW/dt = 0.p)}. (2. (b) As usual a muliplication by a complex number is defined. Eq.t). Hamiltonian Mechanics In the other case.po.p).H(q. (c) A multiplication of the quantities among themselves is defined.46) We thus also recognize the connection between the Liouville equation. (cf. the probability of the initial values Wo(qo. (2.po are constant initial values — we have du0(q0. as we observed. but does satisfy the usual associative and distributive laws. which.** 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. and the equation of motion (2.. 2. " T h e author learned this approach from lectures of H. The equation of motion is then that of an observable.F(«.t) dt du(q.p) It = {u(q.e.4)) ^ = {«(. The time dependence of the expectation values can be dealt with in two different ways. the reason being that — since qo.t) dt __ ~ (2J30) ( 4) du0(qo.
and to be able to correlate these with the above classical considerations. also written "TV". Introducing it here assumes.p} = l {q. In Chapter 9 we attempt a corresponding approach for classical systems with . ~[q.q}=0. therefore.2. (2. Moreover. also called "density matrix". (2. by Traceu:=Ju(q. which all apply to (2. so that we consider this next. B] := AB — BA in the following way: 'i {q.26)) (u) = Tr(wW). for a phasespace integral we define the word or symbol "trace". That this is the case. (b) TV(cm + j3u) = oTV u + /3Tr v. (d) Tr(ut. if u ^ 0.6) a socalled "commutator''' [A. (2.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. (c) TV (u*u) > 0.) = TV (vu). ox In view of our later considerations. it is helpful to introduce already at this stage some additional terminology.p} = 0 • — ^ ~[q. y:\p. Eq.48) In matrix theory the symbol "trace" has a welldefined meaning.p] = l. px > in—.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. that its use here implies the essential properties it has in matrix theory. Thus we can write the expectation value of an observable u (cf.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.48): (a) TV u* = (Tr «)* = TV^. can be deduced from the following characteristic properties of a trace. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''.p)(^y.q} = 0 {p.
Hamiltonian Mechanics a wavelike nature. it will be shown that in electrodynamics. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. i. These aspects will be considered in Chapter 27.e. excepting the Poisson brackets. a constraint) has to be taken into account. .40 CHAPTER 2. the Poisson brackets require modification to Dirac brackets.e. and. However. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. electrodynamics. since gauge fixing (i.
of measurable quantities. i. Building upon this. We found that the Poisson algebra permits extensions to noncnumber formulations. 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. 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. if the elements ipi of M satisfy the usual axioms of addition and K 41 . = {ipi} is called a linear vector space on the set of numbers I 6 {C}. 3. which turn out to be those of the theory today known as quantum mechanics.e. i. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. 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.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics.e.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. we can define the Hilbert space as the space of states of a physical system. A set M.
. . . .ip2). . fa.ip2. are linearly dependent.^1) = (01. such that n ^ = YJCi^i.1) aipi + aipj. (ipi + ipj) + ijjk = ipi + (ipj + ipk).M is said to be a metric vector space or a preHilbert space.  . (ipi.aiV>i + a 2 ^ 2 ) = ai(ip. . (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). so that n 5 > ^ = 0.e. (aptyi. . the vectors il)i. .M. if numbers eti. i = 1. (V>. E (3.4) The vector space .i = 1.2) Vectors ip\. (3.^) called inner product. 2 .. —»• IK.5b) . If n + 1 elements ^ € M. (a. (3. i.ipn are said to be linearly dependent. In each case n linearly independent vectors are said to form a basis.42 CHAPTER 3.^i) .5a) (3. A + tpj = tpj + i'i.~ip2) '• M. ( n (3. if any two elements ip\.02)* (hermiticity). with the properties (a* G IK): (^2. ip2 of this space can be associated with a complex number (V'l. .3) If all on = 0.tp2. x M. n. n exist (not all zero). i=l (3. . ^ i = 0 if ^ = 0. 2 .ipn are said to be linearly independent. and n is the smallest such number. • • • . Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. (3 5c) + a2{ip./?eK). f > 0 if V T ^ O . n is called the dimension of . if every vector ip E M can be associated with numbers Q .
(3. ^ ) + A 2 (V2. meaning oneandahalffold linearity). »2 2 . for arbitrary A and ip2 ^ 0: (V>i + AV>2. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.tp2 £ M.V.^ 2 ) =   ^   2 + 2 ^ i . ^ i ) + A ( ^ i .9a) we start from if) = ipi + \if>2 6 M.^1) 2 h 11 0 <%«e + M .9c) HV'i + tp2\\2 + \\1p1 .. The first two properties imply ( a i ^ i + a2ip2. (3.is defined by d(ipi.7) In addition the following relations hold in a metric space M.V') +"2*(V. (3.9a) (3. (3.9d)   V i   = sup  ( ^ i . ^ 2 )  .9b) = 0 (Pythagoras theorem). In order to verify Eq. The distance between two vectors if>i.)> (36) i. for ip\. antilinearity in the first component (also described as sesquilinearity.2 Hilbert Spaces 43 where the asterix * means complex conjugation. linearity in the second component. which we can write 0 < ( ^ i . (310) (^2. WA+Ml2 = ll^il 2 + ll^2 2 .e. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).^) = ai*(V'i.^2)1 < H^ill • 11^211 (Schwarz inequality). ^ i ) + A * ( ^ ..2.3.V>i + A</>2) > 0 .*h)=\\AH\(38) (3.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations. if Wi. A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. tp2 £ M: 1(^1.rh) (3.
(3. (3.44 or CHAPTER 3. Examples of metric vector spaces: (1) Let M.9a). . (3.V2) IIV'l2 + V'22 + 21(^2.9c). (3.^)1 <IIV>il.^1)1 Vi 2 + ^2 2 + 2   ^   . i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . V 2 ) = 0 . a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0.* Two vectors i f i . for which CO 2 := V^l^l 2 < 00.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. hence verification in each case is necessary. (3.ii) thus verifying Eq. HIHI (3. ^ e M are said to be orthogonal if (</>!.  ^ l l = (HlMI +   « 2 . so that   ^ l + ^ 2   <   ^ l   + ^2.13) the vector is said to be normalized. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. ll^ill + llVdl.11). (3. Mathematical Foundations of Quantum Mechanics 1(^2.12) We omit here the verification of the remaining relations (3.\\ih\\.9d). o ^ V i = o. beginning with A = 1 in the second line of Eq. In verifying the triangle inequality we use this result (3.9e).
2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). Js the space C? is not yet a metric vector space although for (cf. which satisfies relations (3. Elements of the classes are then called representatives of these classes.e. / ) = 0. (3. Eq. are combined to an equivalence class [/] (with space L2). with inner product oo 45 (v. (3. etc. = C? be the set of all complexvalued integrable functions / ( x ) on 5 C IR3 (in the sense of Lebesgue) for which / VSCR 3 /(x) 2 d 3 x < oo.5a).g)= [ /(x)^(x)d3x. Then L2 is the space of all these equivalence classes. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. But this applies also in the case of any function which is nonzero only on a set of measure zero.5c). all squareintegrable functions / which are "almost everywhere equal". */ x } = / /o for x = 0. 2 and so on. i. which differ solely on a set of measure zero. (3. and one defines addition and multiplication by complex numbers with respect to these classes. is defined by ([/]. •'^ \ 0 otherwise.5b). . With the scalar product (f.[$]):= / JM /(x)* 5 (x)d 3 *.e.3. In order to avoid this difficulty.5c)) / ( x ) = 0 = > ( / . (2) Let M. i. for which the scalar product.w) :=J2v*Wi.
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.
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. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V.5) 4. (2) (u\u) > 0. if \u) = 0 (see (3. which we write (u\.62 CHAPTER 4. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. (4.3 Linear Operators. or (4.4b) (4.\u) = (u\0*. Two ketvectors \u). if (u\v) — 0. (3) (u\u) = 0. \v) E V are said to be orthogonal. It follows that we can now write the Schwarzian inequality (3.9a): \{u\v)\2 < (u\u)(v\v). This expression is linear with respect to \u) and antilinear with respect to \v).e.and BraFormalism Furthermore we have: ( x  = 0. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. V \u) (in the case of £) with infinite norm!). i.(t. that the vector \u) E V is associated with a vector in V. the norm squared.5c)).4a) \v) = J\(m<% and (4. The dimension of V is the same as the dimension of V.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). if x\u) = 0 V kets \u) E V. Dirac's Ket.
Then (x(^4it)) is a linear functional of \u) £ V (since A is linear).. The product of such vectors is written: \u^u^) = \u)^\u)^. The inverse of A is written A . as we can see as follows. If Vi has the dimension A/i. 1 and \u) = B\v). Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V.or direct product of vector spaces. It follows that (r. BA = 1. defines the unit or identity operator. Let the following bravector be given: (x\ £ V. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. Multiplication by unity. Furthermore. The inverse of a product is (AB)'1 =B1A~1. Then the following rules apply: A + B = B + A.4. denned as a linear operator on V. the space Vi <S> V2 has the dimension . also acts on the dual space V. This does not always exist. AB^BA. 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.. i. on bravectors are similar to those on ketvectors. let (771 6 V be the bravector defined by this functional. Furthermore linearity applies. The product vectors l?^ 1 )^ 2 )) span a new vector space.or Kronecker. the space Vi <S> V2.e. and one writes: (V\ = (X\A. ltt) = \u). In this way the operations of operators A. i. Finally we define the tensor. B.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u).e. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . if \v) — A\u) i. This product is commutative..e. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2.3 Linear Operators. if AB = 1. The operator A.
we obtain the conjugate relation <tAt«> = (u\A\t)* (4.7C) («)(u)t = \v){u\.e. i.7a) (4. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). Since we demanded the scalar products to have the property (t\v) = (v\ty. [A^.e. Next we construct the following scalar products: (t\v) = (t\Ai\u). (v\t) = (u\A\t). the operator A^ is called hermitian conjugate or adjoint operator of A (cf.A^] = 0.6) for all \u). Dirac's Ket. Every operator A^> commutes with every operator A(2\ i. we write \v) = A*\u). Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above.7d) . and (v\ = (u\A. in the product space.and BraFormalism an operator in the Mi A/2. also Sec. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4.e.64 CHAPTER 4. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. i. As a consequence we arrive at the following properties (e. Every such operator is then also associated with an operator.e. i. 3. Let A^> be an operator in the space Vi and A^ space V2. here designated with the same symbol. If (v\ = (u\A.3). AW\uW)\uW) = \vWUW).g. \t) G V.7b) (4. AW\U)W = \V)W. then \v) is a linear function of \u).
if H is hermitian and (u\H\u) > 0 for all ketvectors \u). Analogous considerations apply in the case of bravectors (u.K}. or (u\A\u)* — {u\A^\u) = (u\A\u). so that (a)(a)t = a)(a.K\. it follows that (u\A\u) = a(u\u). for we had (\u){v\)i = \v){u\. The commutator of two hermitian operators is antihermitian: [H.K]^ = = [HKKH]* KHHK = = K^H^ [H. It follows that a is real. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. so that (n^4n) is real.e. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. The operator H is said to be positive definite. i. [H. hermitian The product of two hermitian operators is not necessarily hermitian. i. K] = 0. Obviously the quantity \a)(a\ is an hermitian operator. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. if H — H' and antihermitian. An hermitian operator A has the properties (a) that all its eigenvalues are real. 65 The linear operator H is called hermitian. .e. but only if these commutators commute. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH.4.3 Linear Operators. An important theorem is the following. Verification: Since A — A^ and A\u) = a\u). as is also (u\u).
) = 0. so that (v\u) = 0. (v\A = b(v\. (u\Ps = {us\. (4. i.Au) = a(v\u) — b(v\u) — (a — b)(v\u). Next we introduce the very useful concept of projection operators. Ps\ua. ( n  z / ) = 0 .e. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. Ps2 = Ps(4. which includes vectors of infinite norm. If these form a complete system. i. Let S be a subspace of the Hilbert space !K. (u\u') = 5{v .66 CHAPTER 4. since a is real. \u8*) E S*. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round. Dirac's Ket. we have 0 = (uArt) — (u. so that (u\Ps = (us\. The entire set of ketvectors spans the extended Hilbert space.and BraFormalism Moreover. Ps is idempotent. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. i. Assuming A\u)=a\u).e.e. a ^ b. then the hermitian operator defined on this space is an observable. but the continuum vectors are normalized to a delta function.8) The projection operator has the following important properties: Ps is hermitian. as we can see as follows. and let S* be its complement. i.9b) (4. The above theorem remains unchanged. {us\us*) = 0.u'). (n\n')=5nnl. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). since a ^ b.e.9a) . A continuous spectrum can immediately be included by passing to the extended Hilbert space. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us).
\ua) = c\a). . (4.9d) Thus the vectors P\u). Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. as can be seen as follows. i.3 Linear Operators.e.4. i.. so that (Ps2 ~ Ps)\u) = 0. = (p2p)\p). Let p be an eigenvalue of the projection operator P. . If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. so that \ua) = (a\u)\a) = \a)(a\u).e. Set \u) = \ua) + \ua*) with (a\ua*) = 0. 67 (4. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u).e.. \u) arbitrary. Very similarly we can construct operators which project onto the subspace spanned by the continuum states. 2). iV) is a set of orthonormalized states. Let a continuum state be written £). where £ is a continuous index.. Ps is projector onto S.e. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. c = (a\u). i. The quantity \a)(a\ is called elementary projector.9c) This property can be seen as follows. (1 — Ps) is projector onto S*. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. Then 0 = (P2P)\p) and hence p = 0. p 2 . i. (1 — P)\u) are orthogonal. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw).1.p = 0. if the set of vectors 1). P\p) =p\p). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN.
If there are r such vectors. which are orthogonal to each other (i. see Eq. and Eq.. (11.. that one and the same eigenvalue Aj is associated with several eigenfunctions. Examples 8.6. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. Observables are representatives of measurable quantities. Coulomb potential. Degeneracy will be discussed at various points in this text. the operator P is correspondingly P2\u)= f2\0m\u).e.g. 4. i. . are linearly independent). u2).1 and 11.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. See e. one says the degree of degeneracy is r . P = £n)(n.114b).114c). Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). 8. be a system of basis vectors in this space. In general it is possible. which we introduced earlier. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. Let \u\).e. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. Sec.6.4 Observables Operators which play a particular role in quantum mechanics are those called observables. in this case the spectrum also has a continuous part.68 CHAPTER 4. 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. However. Dime's Ket..r)(ui. (11.r\ "An example is provided by the case of the hydrogen atom.
i. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.10) follows from the fact that i Applying the operator A to the projector (4.11) i. (4.e.r'} = Sii>6rr>. Applied to an arbitrary vector \u) G "K Eq. ^ A i P i = A = ^Aiui. and APi = XiPi.r)(ui. Together with the orthogonality condition. is always implied).r \ui. 69 i If A is an observable and if the spectrum is purely discrete.4. PA = y£pi i = Yt\ui.r (4.r =^ i.r \(ui.e.12) .r) (the convergence. then PiPi' = 0.e.r.r\u)\2. we obtain i i i i i. The uniqueness of the expression (4.10) gives the linear combination it) = ^ i. in the case of degeneracy with (ui.r\u) i.10) This expression is known as completeness relation or closure relation. (4. If Aj ^ Aj'.r\ = l. the relation (4.4 Observables The dimension of this subspace is that of the degree of degeneracy.10). and the eigenvectors \ui. the operator A is completely determined by specification of the eigenvalues A.r\ui/.r>(ui. The operators Pi are linearly independent.e.\i)Pi = 0.r)(ui. i i (4. which we do not enter into here. Let us set {A . i. or also as subdivision of unity or of the unit operator. i.r\u). the projection of this operator is the entire space.10) expresses the completeness of the orthonormal system.r)(ui.
e. Let v be the continuous parameter which characterizes the continuum. (3. 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.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 we denote by \uv) the eigenvector of A whose eigenvalue is \{v). (4.17) previously. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). now however. (4. the eigenvector \ui): f(A)\ui) = f{\i)\ui).14) . In a similar way we can handle functions f(A) of an observable A. i. One defines as action of f(A) on. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. written in Dirac's notation. for instance. (u„zv) = 8{v .e. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. The ketvectors \uv) are orthonormalized to a delta function. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv.70 CHAPTER 4.v').and BraFormalism This is the expression called Parseval equation which we encountered with Eq. we have the corresponding generalizations. If the spectrum of an observable A consists of discrete as well as continuous parts. for instance exponential functions. Dime's Ket. as explained earlier. i.
59). i. (4... if and only if they possess at least one common system of basis vectors. 4. Eq. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.16) represent subdi . for which the completeness relation is f dk\k)(k\=l. Extending this we can say: A sequence of observables A.e. We shall return to this theorem.and bravectors. (4. with energy E. Since both expressions (4.58) in terms of ket. B].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. In the following we consider more generally ketvectors \ip) as representatives of the physical states. here presented without proof. as a formal orthonormality condition. Therefore we want to reexpress the Fourier transform (3.C. \k)€Fk. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. if (a) they all commute pairwise.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. First of all we can rewrite the integral representation of the delta function. In the onedimensional case we write {x\x') = S(xx'). [A. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). and (b) if they possess a uniquely determined system of basis vectors. For many practical purposes the use of the Fourier transform is unavoidable.15) and (4.e... (4. form a complete set of commuting observables.B. (3. again later. i.17) The Fourier transform provides the transition from one representation and basis to the other.4. where the symbol ip is already indicative of the Schrodinger wave function ij. / dx\x){x\ = 1. their commutator vanishes. which all commute with one another.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. x) € Fx.
which are representatives of the states of our physical system. called wave function.and BraFormalism visions of the unit operator.15): 5(x .1a). il>{x) := (x\i/}) : "K * C. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. i.e.space representation. V 27T (k\x') = .x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). Obviously we obtain this by inserting a complete system of basis vectors of Fk. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). (4.19) Comparison with the orthonormalized system of trigonometric functions (4. (4.e. <a#) = [ dk{x\k)(k\ip). 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).72 CHAPTER 4. (4.18) According to Eq. i.59) or Eq. we can rewrite Eq.2a) etc.L e " * * ' = {x'\k)\ V27T (4. . (x\k) = ^=eikx. (3. this expression has to be identified with — / dkeikxeikx\ i.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. Rather \x) and \k) serve as basis vectors in the representation spaces Fx. ${k) := (k\if>). (4.e. Dime's Ket.21) provides the ketvector \ij)) in the A. (4.Fk.
Schrodinger [244]. Thus we distinguish between socalled pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator.e. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). which is the quantum mechanical analogue of the classical probability density p(q. 73 . with the complementary system or *E. i.t).5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. if we do not take into account the complementary part of the universe.2 The Density Matrix We shall establish a matrix p.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.Chapter 5 Schrodinger Equation and Liouville Equation 5. 1. the calculation of which for specific cases is the subject of Chapter 7. 5.p. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. We mentioned earlier (in Sec.
Schrodinger Equation and Liouville Equation the measuring instruments.. which in Dirac's notation we can write as = J]C a »i) a. i.b. i. \a). we have (a\b) = 5ab. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a.i.. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. For an exact treatment we would have to express the actual state \ip) of the system as a supersposition not only of the states \i). These are the quantum mechanical states which are also referred to as microstates. that is. T .e.e. we have to deal with a socalled "mixed state". The actual and real state of a system is then a superposition of two types of states. but also of the states of the complementary set of the universe.. \a). (a\i)=0 = {j\b). are called "pure states". as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. p" ( C C * ) t = CTC* = p. The states \i) £ "K.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. a (5. called the density matrix.. of an operator that acts only on the states of the system we are interested in. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. where i represents collectively all quantum numbers of the state of the system under consideration. where T means transpose: p = CTC*.i. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows..74 CHAPTER 5. (52) (5.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a.3a) and N J  a ) ( o  = l in the complementary subspace.b. \i).^ Since p is hermitian.3b) The quantum mechanical expectation value of an observable A...
The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). i. a projection operator.j(x'i)(ia. which are all subjected to the same macroscopic boundary and subsidiary conditions. This ensemble is interpreted as a large set of identical systems..x) := (x'\p\x) = yju. The ensemble can be represented as a set of points in phase space.e. in the position state representation or xrepresentation.2) and (5.e.e.10) and is therefore. This transition to a new set of basis vectors in which the matrix becomes diagonal. as we discussed previously. P ^ ) = i)(#) = (#)i) (5. e. In this sense we have p(x'. which today is a fundamental term in statistical mechanics. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities.g. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann. (5. (5. If the ensemble corresponding to the system under consideration consists of Z elements (i. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.5.6) (i'\p\i") = Wi'Si'vi = real.2 The Density Matrix 75 \i') 6 "K. seen macroscopically of Z copies of the system).. i i (511) In Eq. but occupy in general different microscopic. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. a complete orthonormal system with properties (5. i. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).4) we had the expression (A) = Y/Pij(j\A\i) i. . In the following we write simply \i) instead of \i').. as we saw earlier. where x represents collectively the entire set of position coordinates of the (particle) system. (see below) eM.9) is diagonal. (5. Gibbs introduced the concept of ensemble.ji(x / )i*(x).3a).7) p = ^2\ii)(i.e.\p\i")(i"\ = Y. in a specific representation. or (5.) = YJc t .3 with Pi^YtCaiCaj*. quantum states.* We can specify the state of our system.
We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) .4)) (5.16) (5. as claimed by Eq. Hence Wj/ > 0. On the other hand. (5. (5. 3 (5. (A. i t (5.76 CHAPTER 5.18).17) Hence the sum of the classical probabilities is 1. Eq. (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. (5. Thus Tr(j>ii)<i)=l. Then the expectation value of the observable A is (cf.14) which we can write (A) = Tr(pA). Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. In particular for A = 1. (5. Eq.18) Substituting this into Eq.3 = pvv = uv. (5.5) above) (i\p\j)=Pij. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). (5.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l.4).13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).) := (mm = (#') W> = l(#')2 > 0. according to Eq. we have Tr(p) = l. We can also show that Ui > 0.
The effect of the interaction of the system under consideration with the surroundings. ^2ui\i){i\.t) = {x\i).23) Instead of i(x.t) and instead of (x\Pi\x) we now write p(x. but we see here. We now consider the latter expression.3 The Probability Density 77 We see therefore that p defines the mixed states. (Question: Is the state of the universe a pure state?) 5. (5. in all other cases the state of the system is a mixed state. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. and in the other states with probability zero.t)\2. so that p{x. by representing the vector \i) by the wave function i(x. Thus we have a system in a pure state. that — different from u>i — P. the system is in a pure state. Equation (5.t) we now write tp(x. If all ui but one are nought. since they do not take into account the (interaction with the) rest of the universe.x').21) p=\i){i\=Pi.24) s (x\x') = 5{x .8). one says. that these are actually not sufficiently general.20) expresses that the system can be found with probability 1 in the state \i). In the following we are mostly concerned with considerations of systems in pure states. i. t) p= i The expression (5. defines the statistical operator. thus enters these states. We see this more clearly by going to the position space representation. the only remaining one is 1.22) In the following (x\i) is always to be understood as (x\i(t)}. For the case n we have TJ=J (520) (5.e.3 The Probability Density p(x. We observed above. which is contained in the coefficients Cai.t).5.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l.e. not in a mixture of states. i. i. represents a quantum mechanical probability.t) = \^{x.e. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. (5. (5. .
Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).^ The equation which is the conjugate of Eq.t) = \rp(x. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal). with respect to time. Schrodinger Equation and Liouville Equation The relation (5. (5.27) is iHJ\ jt= m .t) = / dx{i\x)(x\i) = (i\i) = 1. (5. This differentiation requires the time derivative of a state vector \i) E "K. or the wave function tp(x. the Schrodinger equation is always postulated in some way. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. t).28) Differentiating Eq.26) 5. We shall convince ourselves later that the postulated equation is sensible (in fact. 2. which is chosen in such a way that the desired analogy is obtained. 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. We now want to obtain the quantum mechanical analogue.e. (5.8).25) J Thus the particle whose state is described by \i).27) where H is the Hamilton operator of the system. Equation (5. (5. The generalization to a threedimensional position space is selfevident: JK 1 / p(x.e. it is not "derived"). the expression (5. To this end we have to differentiate the density matrix p.t)\2. (5.4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec.8) and multiplying by ih. .27) is known as the Schrodinger equation. i. and hence the integral over all space dxp(x. but the linear operator states are moving — even if the system is observably stationary).78 CHAPTER 5.3. We therefore postulate first an equation for the time dependence of a state vector. (5.22) can also be written (xi) 2 = (i\x)(x\i). can be found with probability 1 in the space R1.
31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.34) .27).21). i. We can now consider the Schrodinger equation (5.37)) (5. both uji and i)(i). (2.e. Eq. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5. (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi. (5. (5.e.28) the states are considered as timedependent. such a formulation is described as Schrodinger picture. (5. (5. With Eqs. p also contains "quantum mechanical probabilities" (i. The reason for this is that in addition to the "classical probabilities". The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.e.p} + ihJ2^\J)(J\. We thus obtain the operator form of the Liouville equation.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.p}.32) Comparing Eq. the relation ih^ = [H. not the partial derivative. (cf. i.27) and (5.28) we obtain (see also Eq. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. dp/dt ^ 0.5. we observe that here on the left hand side we have the total time derivative.24)).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).32) with its classical counterpart (2. (5. (3.Xi)\j). Eq.
35) Here we can look at the Hamilton operator as a "time generator3''. i. i.x 1 ^ ( x .35) can be written'! ip(x. t ) .4. we write V>(x. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. 0) or \il))t = \il>)t=o exp (5.4. the state \i) is still timedependent. „ / ith—. we can replace these states by corresponding timeindependent states. . Schrodinger Equation and Liouville Equation or with (x\j) — ip(x..38) (x\j)t=o = Yt(x\E^(En\J)t=o. With the help of Eq. which we shall need later. but with respect to the parameter /3 = 1/kT.Xi){x\En) = En(x\En). operator" the exponentiated oper .** In view of the close connection between time.80 CHAPTER 5. n (537) This equation is described as the timeindependent Schrodinger equation. in Sec. however..and temperaturedependent Green's functions. to) = exp[—iH(t — to)/H\.t): > ih—</>(x. 5. appearing in the Boltzmann distribution. since the solution of Eq..t) = Hld .e. (5. Feynman [94]. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. we shall describe as "time development ator given by U(t. it is plausible to refer to this equation at this stage.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. (5. with respect to time t. or H\En) = En\En).e. T meaning temperature. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed. 10. (5. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later.t) = exp ip(x. P.36) in in The initial wave function or timeindependent wave function ^(x.36) however. In expression (5. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. a (5. in p = ^2i^i\i}(i\.8) for p.0): where H(ih—. **R.
(5.44) r (TrePH)' (5.x'(3) = Si e ~^W*V). Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. (5. Eq. (5.42) Inserting here Eq.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. i (5. (543) Since H(f>i(x) = Ei<f>i(x). i. we obtain p(x.45) since on the one hand •PEi i i i and on the other hand . the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.e.4 Schrodinger Equation and Liouville Equation 81 Example 5.1).5. we can rewrite p.40) 1.41 ) or with 4>i(x) := (x\Ei) this is p(x.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). or ^ = ^ e 0Ei _m. (1. i ( 5 .^ n tne position space representation Eq.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. (5. x') = ^ i LUiMxWix').x')^p(x.40).
x'. [A.48) = 1.82 CHAPTER 5.4> = ^ g ^ .51) is seen to be very similar to the Schrodinger equation (5. [B. (5. Equation (5. With Eqs.52) Hence this expectation value can now be evaluated with a knowledge of pN. (5. B}} .x. We now rewrite the factor exp(—(3H) in Eq. Example 5.47) with respect to (3. B] + L [A. This solution is the Green's function or kernel K(x.X'.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. (5.47) (5. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5. . A}} + • Solution: The relation can be verified by expansion of the left hand side.51) where the subscript x indicates that H acts on x of PN(X.(3) := (x\pN(P)\x') = (x\e~0H\x').46) is pN(x. (5. whose calculation for specific cases is a topic of Chapter 7 (see Eq. (5./3).46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.x'. Schrodinger Equation and Liouville Equation so that p is given by Eq.46) we can write the expectation value of an observable <^Tr<„. (5. (5. (3).x'.45). M.45) and (5./?). (7.tt tt For a more extensive discussion see R.51).35).L [B.45) without normalization as PN(J3) := e~pH.(3). (5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).49) In the position or configuration space representation Eq.50) Differentiating this equation with respect to j3. we obtain = HxPN{x. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A. (5.55)). (5. Tr(pN) = / dxpN(x. Wilcox [284]. Differentiating Eq.
83 .. 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. Noz [149]. 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).Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. 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. * Comment of Y. (8. Kim and M. This is a fundamental topic since the harmonic oscillator.53)). we shall see on the way. 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. However. like the Coulomb potential. E.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator." Quite apart from this basic significance of the harmonic oscillator.. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ". S. can — in fact — be quantized in quite a few different ways.the present form of quantum mechanics is largely a physics of harmonic oscillators. Since the spectrum of the harmonic oscillator is entirely discrete. that the oscillator.51) to (8.
which is — in the first place — representationindependent. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . This equation would then have to be solved as a second order differential equation.' ^mgLOh by setting and *~. in the present case + [EmooJ2x2)ip = 0. hence we do not use a distinguishing notation. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. . the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. of which the only nontrivial one here is [q. When such a term arises one has to resort to some definition.^ The first problem is to determine the eigenvalues and eigenfunctions of H. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). To this end we introduce first of all the dimensionless quantities rriQUi fl q.p]=ih (6. p^ih—. we can also proceed in a different way. so that there is no ambiguity due to commutation.p. 2mo dx \ 2 2 (6. h = l. However. Quantum Mechanics of the Harmonic Oscillator 6. For instance we can go to the special configuration or position space representation given by q^x.1) for the hermitian operators q. like taking half and half. This can be done in a number of ways. Also observe that H does not contain terms like pq or qp.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers.84 CHAPTER 6.2 The OneDimensional Linear Oscillator In the case of the onedimensional.to = 1 reduces to the parabolic cylinder equation.2) which for mo = 1/2. which is called Weyl ordering.
A*} = A^A. (6. (6.C].e.e. C] = A[B. We now use the relation (3.BC] = [A.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. + 1). (6. Prom these relations we obtain ( ^ = ^ . i. . With the help of Eq.1 ) .7) (6.14a) we deduce for an eigenvector \a) of A^A. C}B. q are hermitian). for ^ A  a ) = a\a).B}C + B[A.A^] = 1.6) and p = y/motoh AAi =^. (6. i.8) The eigenstates of H are therefore essentially those of N := ^ f A (6.10) then a = (a\A^A\a) = \\A\a)\\2>0. A^] = A\ (6. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . if A*A\a) = a\a).2 The OneDimensional Linear Oscillator 85 (p.5) Reexpressing q and p in terms of A.34d): [A. and [AB. A] = [A\ A]A = A.1) we obtain immediately the commutation relation [A.12) in order to obtain the following expressions: [tfA. (6. (6. A\ we obtain h A + A^ rriQLU y^ (6.11) Thus the eigenvalues are real and nonnegative. C] + [A.6.13) ( ^ A ) i t = A\A^A From Eq.14b) [A* A. (6.14a) (6. (6.
since this equation implies that the eigenvalues cannot be negative. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .l)A\a). unless A 2 a) = 0. but (b) An\n>\ llA An+1a)=0.20) . (6. (6.2)A a). Next we consider the vector A 2 a).l)A\a) = A(A^AA . or Ata) = v ^ T T ..16) This means.A)\a) A{AAU . However.e. in view of Eq. Thus for a certain value of n > 0.11) this is not possible.86 the relation CHAPTER 6. unless A^\a) = 0.2A)\a) 2 (a . (6.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). unless A\a) = 0. If we continue like this and consider the vectors An\a) ^ 0 for all n.21) .19) i. we must have (a) Let \an):= Ana)^0.  a) is eigenvector of A^A with eigenvalue (a + 1). (6. (6. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). A 2 a) is eigenvector of A^A with eigenvalue (a — 2). (6. (6.l)\a) = (a . In this case we have (A*A)A2\a) (6 (6.17) (6. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1.l)\a) = A(a .18) = '= a ) A(A^A . or \\A\a)\\ = Va. Similarly we obtain from Eq. This would mean that for sufficiently large values of n the eigenvalue would be negative.2A)\a) = A(Aa . we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).
n ) = " A»a>=» ' = A«a)F = 1 . so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1. From relation (6.22) With relation (b) of Eq (6.. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a .' " /. that an = 0. (6.I n U .27) According to (b) of (6.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers. (6.17) a by (a — n). For a — n = 0 we deduce from (b) of (6.18) we obtain for a = n: p t  n )   2 = n + l.e. (6. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. v.. \ 9 \ I / II 4 T ) .20) that A0) = 0. The state 0) is called ground state or vacuum state.25) (6. Moreover.20) we obtain from this that the right hand side vanishes. (6. \ I 2 2 x 87 Replacing in Eq. (6.5) to shift operators A to the right which then acting on 0) give zero.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).20) we have for a = n = 1: A2\1) = 0. I „ . But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. i.6. or a = n > 0.n  a . In the following manipulations we always use the commutator (6.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.24) This is a very important relation which we can use as definition of the state vector 0).28) .
5) and again (6. l>ocAt0).26) the equality l) = A t 0). But using Eqs. In general we have \n) = ^(A^n\0) (6.33) . (0^ 2 (At) 2 0) = 2. as we can see as follows. (6.11) we have [A. and in general [A. (A*)2} = [A. Similarly we find 2)oc^Ut0).e.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.^ ( O l A ^ n o ) . and in view of Eq. (6. (A^)3] = [A. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0.32) (6.29) Hence we obtain and in view of Eq.29). we have 2> = ^=dAi\0). The states \n) thus defined are orthonormal. According to Eq. (6. (6. i.11).32) we have (nm) = . A^} = 2A\ and [A. (At)2]A+ + (A*)2[A.34) (6. (6. A^A* + A*[A.(A*)n] = n(Ai)n\ (6.31) (6.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).(6.88 and CHAPTER 6.
1 ) . .41) The contribution hu/2 is called zero point energy. .1 0). (6.37) n^)711}^) = v n  n .33). l .39) A^A\n) = y/nA^\n .1) (6. ldnm = n\5nm. i. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.e. (6.38) (6. (6.1) = n\n). Here.35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. . . whose number is given by the integer eigenvalue of the number operator N = A^A. With Eq. .40) (6. . n = 0 .1 (A + ) m .36) + 1) (6.J \n) = hw (n + i J \n). •(» (6.32): rf\n) = .6. (6. we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). (6.34) A\n) = ±=A{Al)n\0) .8) we obtain therefore H\n) = hu (A^A + ].34) and (6.'n\ and = v n ![ ( A t ) M i= + (6. we obtain (n\m) = 5nm We also deduce from Eq. however.2 The OneDimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An1A^mA\0) + (0\An1m{A^)ml\0} 89 0 + m(0A n . chosen in close analogy to the postulated "second quantization relations" of field theory. .35) Inserting this result into Eq. In view of the properties (6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq. 2 . The terminology is. 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 . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . in quantum mechanics.
39): {n\A.90 CHAPTER 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 (6.46) v^ . Quantum Mechanics of the Harmonic Oscillator 6.3 The Energy Representation of the Oscillator The energy representation. Eq. is defined by projection of operators on eigenfunctions of energy. (6. A^ from the relations (6. Vn + 1..43) = = y/n(n'\n — 1) = y/n5n^ni.45) Correspondingly we obtain with Eqs.. ( 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. other elements zero. i.38) and (6.39)). (6.g.42) We can deduce the energy representation of the operators A. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn.7) the energy representation of the operators q and p.'\n) i'\A*\n) (n + l\A'\n) and similarly (n'An) (n — l\A\n) In matrix form this means 0 0 0 0 0 0 0 0 0 A/4 = = Vn + l(n'\n + 1) = \/n + 15 n / iTl+1 . also called Heisenberg representation.6) and (6.2.1. (5. (6. and — in fact — we have used this already previously (see e.e. n = 0. En = hu(n + )...44) yfn. (6.. other elements zero.
6.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. (6. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n).47) LP 0) = 0 .32).24). (6. (6. by (cf.5) are also satisfied as matrix equations.6. Eq. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .1) and (6. The ground state wave function (/>o(x) is defined by Eq. (6.4 The Configuration Space Representation We saw that the eigenstates are given by Eq.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. A\Q) = 0. i. mow / (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. 2 . ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.e.3)) ^ (q 2n \ + (6.
 ) e"^2. Vn! Now {XlA (6. (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 .51) \l 2h {X q \ m0uP)\l 2h \X m0u.0 = ( .l ) > n ( x ) . (6. (6.L ( x  ( A t ) »  0 ) . Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator.2 n n! The functions Hn(£) are obviously polynomials of the nth degree. In order to obtain the wave functions of higher states. they are called Hermite polynomials.e. c/>n(x) := (x\n) = .54) (6 55) i/4 rzri7r 1 / 4 vo. it suffices according to Eq.92 CHAPTER 6. i.56) < ^ = We have as an operator identity the relation . These polynomials are real for real arguments and have the symmetry property # n ( . so that <t>n(x) = ( .
58a) ' T h i s definition — apart from the usual one — is also cited e. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.4 The Position Space Representation i. p. (6. 93 . Oberhettinger [181]. 81.12. (2036(20.en_ dt. e. (202"2. Magnus and F.g.57b) • Generalizing this result and inserting it into Eq.12(20 . (6. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f.e.58b) by W. . ( 2 0 4 .2)m = (±t)m for some function VKO.6. d£" (6. 2£.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.
£ 4 .He'n{£).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. (6.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6. . = ffen+itO+ntfeniCO. (6. (6. Magnus and F. (6. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. £33£.(e^/2) or Hn(0 = 2n/2Hen(V2ti).59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 .94 CHAPTER 6. = nHe^iO. 81. 80. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. £2l. (6. Oberhettinger [181].6 £ 2 + 3.59b) = (l)^2/2^.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm.^ + n) He n(0 = 0 (6.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . pp. £.60b) J—< "See for instance W.
61) e?l*2nl2Hn{Z).6.0) we obtain + s fdFs 1! V ds s=0 + OF ds 2 dF ds2 = s=0 2(f d S )^«)a s=0 = 2£ = ffi(0.4 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£) = ( . Oberhettinger [181].62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S. W.s) = J2 n=0 (6.s)=F(£. s) as a Taylor series.g. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0. Thus the generating function can be written F(Z. F(S. . and so on.l ) ^ e^HeniO 2 / 4  . 93. Their normalization is therefore given by OO /"OO / OO J —OO z (6.64) n\ See e... Magnus and F.63) The meaning is that if we expand F(£. p.^ = e^/42~n/2Hn(aV2).s) = eM?(Zs)2}(6. (6.
96 CHAPTER 6. mow' Apx = ^m0Luh/2. i. We also see from the eigenvalue t h a t the energy of the lowest (i. 6. We see t h a t the lowest state is symmetric.e. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . for x(t) = 0 and p(t) = mox(t) — 0. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. in order to estimate the lowest energy of the harmonic oscillator. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. In classical mechanics the energy of the oscillator can be zero.1. has parity + 1 . ground) state is fko/2. 6.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. i. and we see t h a t even and odd states alternate. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. 2 . In fact we can use the uncertainty relation AxApx ~ H/2. Quantum Mechanics of the Harmonic Oscillator Fig.e.e.
mass mo. 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)).: 27 T mov2a2. This observability can be taken as direct proof of the uncertainty relation. classical orbit i. Sec.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. Example 6.1/a O l/a Rez all at infinity Fig. setting q = 1/z.6. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. like twoatomic molecules. /classical orbit Here q goes from 0 to a.plane pole at z=0 ^ .1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". and is directly observable.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. n*h. 14.e. In the plane of complex q the integral has no poles for finite values of q. Thus. E . which is the complete orbit. in corrected version instead with n* = n + 1/2. 27rmoi/ <f> V'a 2 — q2dq = n*h. 6..2 The original path and the transformed path at infinity. The analogous Example 6. we obtain the integral . and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). pdq. In Example 6.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. According to "old quantum theory" with integer n.1.4) to obtain by contour integration the eigenvalues of the harmonic oscillator. original path q . to —a.plane Imz z . to 0. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. The kinetic energy is always positive or zero. back to 0. But it has a pole at infinity.
6. z = 0 Vi 2Kmov(Ka2) = n*h = E/v.V62 a2). We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . and obtain with this all properties of the solution.) hv. 2 . we start with a Laplace transform ansatz for the remaining part of the solution. .1.e. After removal of a Gaussian or W K B exponential. a J \d Solution: The solution proceeds as in Example 6. 6.3.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. E = Z2e4/4(l + n+ l ) 2 (cf. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points).5.( 6 . i. « qdq + I2 = f dq . Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. 1 .It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . £ = n*/w = ( n + .65) f ^Va2z2 .2b = Ze2.114a)).2: Contour integration along a classical orbit Setting q = 1/z. harmonic oscillator and Coulomb interaction. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.. Example 6. for instance — without deriving all properties of the solutions.1)! (6. With a2 = E. .2TTI a ^ = .e. in q at infinity). hence we consider the following differential equation in which n = 0 . but here we leave this . 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). (m .1.a) m = 2 .1 D e r i v a t i o n of t h e g e n e r a t i n g function From the above we know t h a t the energy is quantized. i.98 CHAPTER 6. since these two cases.5 . Eq. (11.c2 = (I + 1/2) 2 (natural units.e. are the most basic and exactly solvable cases in quantum mechanics. In the following we investigate in more detail the important differential equation of the harmonic oscillator. / " ^ . . verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .
(6. E. (6. so that the integrand of the remaining integral vanishes. (6.J ±(ps(p))e*dp. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) . Eq. and / dpf(p)pn the Mellin transform of / ( p ) . (11.g. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. (this determines the limits).** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. Oberhettinger [181].e.p t dp = \ps(p)ePt} .66) First we have to remove the £2term. (6.68) § = J p2s(p)e*dp. . Magnus and F.69) **See e. 80. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. Then (with partial integration) *dt = ~*/Ps(p)e. This has to be true for all values of t.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. Substituting these expressions into Eq. .t*ij& + ny{t) = O. W.i ^  (ps(p)) P n P Integrating this expression. (cf.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] = 0.6.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.g. fp{ps{p)) = {v2 + n)s{p). i. p.111)) the Mellin transform of e~p is n!. / dpf(p)e~px the Laplace transform of f(p).
= / • Im p Rep n not an integer Fig. Eq.. This exponential assumes a maximum value when pt \—p2 = minimal. Hence a possible path of integration > is from — oo to oo. we can choose a path from zero to infinity (the condition will vanish at p = 0). To see this.71) . Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. n must then be an integer. (2) If n < 0. We can show that if \p\ is allowed to be large.100 and hence CHAPTER 6.70) = 0 P. The quantum mechanical wave function is then (cf.3. when p = —t (which is possible. as indicated in Fig. The value of the exponential at this point is ~ exp(£ 2 /2). of course). This can be attained if p is allowed to be large (t can be of either sign).e. (3) We can choose a contour once around p = 0 in the complex pplane. (6.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. (1) This will be satisfied if \p\ — ±oo. We now investigate possibilities to satisfy the boundary condition C(p) = 0. 6. i.3 The path when n is not an integer. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. so that y(t) ~ exp(i 2 /2). 6. Therefore we take ptp2/2 y(t) . so that p cannot go to infinity. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. For singlevaluedness. then for large t we have y(t) ~ exp(i 2 /2). consider the exponential in the integrand.n+l between the two limits of integration.
6.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. . .l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . 4 . this becomes jfj. .„y2 v l { ' ^ r (^) \ 2 ' where v = 0. 2 . . The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . In y(t) above.if n is odd.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.3. (676) ..5. if n is even and v = 1. The exponential function here is described as the generating function of the Hermite polynomials Hen(t). We have 27T! Setting q — p + t.. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. pt p2/2 (6.. We can also obtain the derivative form of this Hermite polynomial.P 2 / 2 00 C—tY ( . n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .0) p p»+i v ' Hen(t) = (l)^ e * 2 /2 / 2vri /(9=t) dg " ^(gi)"+i e ?2/2 = e i 2 / 2 ()ne"i2/2. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt).
n + l ) f f e n ^ i ( £ ) + (n. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials.4.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. n — l)Heni(t). (6.1 '2dt using Hen(t) = (l)n27 r .75) as Hermite polynomials. as in the normalization of asymptotic expansions of Mathieu functions). 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). (6.59a).102 CHAPTER 6. (6. (n. and elsewhere. 8.n— 1) = n. ( n .it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). Pptp 2 / 00 /2 dp.3 and 6. for the evaluation of expectation values.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Example 6. Hei(t) Henj.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. Reinterpreting the remaining integrals with Eq. Solution: We have to evaluate °° 2 . in radiation problems.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.g.77) = t. one can construct the tower of polynomials Example 6. we obtain tHe„{t) (l)"n! ~ i. n + l) = l. Example 6.70). We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. Hen(t).e. 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+<. Starting from Heo(t) = 1.
/(n __ 2r — /i)!. 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.66) is \Zn\V2w Example 6.n)\ i s an+2rij. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6.5 nm . (6. g) is n! Next.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.q) = = e .80) f(p.82) ^ . the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.jQ) n n!I ^—' ^Q fi\(n .(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).q).6. and 0 otherwise \/2n(—l)n+mn! if n = TO. zero otherwise (6. Hence the normalized wave function of Eq.78) \/27rn!. (6. so that /2 Q2(n+rrf in / ( 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 .
A o?(. n + l)(n + 1.' .v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .n + 2) + (n.— — ^ .° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . n + l)(n + l . Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).104 CHAPTER 6.e.78).^ ) ! ( n + 2 r .^ / 2 * .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . n .A i ) ! (6.ra + 3)(n + 3. i / ! ( n . n + 3)(n + 3.v)\(s . i. . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . K is the coefficient of a / ( 2 s ) ! in the expression J . n + 2)(n + 2.l . n + 2)(n + 2. n + 2) + (n. The result then follows with half the normalization and orthogonality integrals (6. /i = (n — v) _ / r o o e .u)\ 2s<—" v\(n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h .n+ l)(n + 1. n + 2) + ( n .e. n + l ) ( n + 1.i / ) ! ( 2 r + i/)! > . n + 2){n + 2. so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.v)\ ^ Multiplying both sides by the value of the normalization integral. n.2s _oo ( a 2 / 2 ) i . As an example we obtain for the integral from 0 to oo (also reexpressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e .r .79). n + 4)_ffe n +4(t) + [(ra. n + l ) ( n + 1. the relation (n.l ) ( n ." ) a2^) r coefficient of (2s)! in f^ (s . 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.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.r . n ) ( n .e. i. n — 1) are given by the first index. 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.e '' ^ o^M ! ( n .(t)e«2/2dt Now. by n!(7r/2) 1 / 2 .. Prom Example 6.83) = (n + 2r)!e a / 2 V „ „.3 we know that upgoing coefficients are 1 and downgoing coefficients (n. we obtain the result (6. i.77).( n + Jr). n)(n. n + 2)]ife„ + 2 (t) + ••• . Hence t4 He„ (t) = = Hen+i Hen+4{t) (t) + [(n + 3) + (n + 2) + (n + 1) + n] H e n + 2 (t) + • • • + 2(2n + 3)Hen+2(t) + . n + l ) ( n + 1.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . n + l ) ( n + 1. . the coefficient of T 2s »{sr u)\ As we observed.
The first of these cases will reappear later.x'. which is a point of unstable equilibrium for a classical particle. which is normally not discussed. more generally.Chapter 7 Green's Functions 7.1) be the timeindependent Schrodinger equation which has to be solved. With reference to this case.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. 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. We consider first timeindependent Green's functions. in Chapter 21 (with a different calculation).t). This example also offers a convenient context to introduce the inverted oscillator potential. we consider the sojourn time of a quantum mechanical particle at a point.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. and is thus an important point to be noted here. Let H^ = E^> (7. We 105 . which we write G{x. and those which are timeindependent. 7. in Feynman's path integral as the kernel or free particle propagator. of a differential equation of second order. These are the Green's functions of the timeindependent Schrodinger equation or. which we write here K(x.x').
g. Green's Functions H = H0 + Hi.x'). square integrability. x') has to satisfy correspond to those of * (to insure e. (7. where £ ( 0 ) = lim E.EW)xG{x.x') or ^2^(x)^*(x') i = S(x . i.13). a perturbation part like (3q4. of the kinetic part p2/2mo and some part of the potential. aq2. for instance.e. Eq. e. (7. cf.g.x') by recalling the completeness relation. since the generalization to higher dimensions is selfevident. (7.x') can be written t G (7. The boundary conditions which G(x. and Hi is some other contribution.e. We can readily see that G(x. the relation X') = 5{x .4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). although the explicit calculations can be much more involved.x').5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. appropriately decreasing behaviour at infinity).3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . The timeindependent Green's function G(x. . We can obtain an expression for G(x. e.106 assume a Hamiltonian of the form CHAPTER 7.x') is defined by the equation* (Ho .g. i.2) where Ho consists. 'The case of £(°) equal to some Ei is considered later. (7.
Hi{x'))^(x').x'.x'){E .x'.) = f dx'G(x.#. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).E^ We can check this by considering (H0 . (7.t) (K for "kernel") is defined as solution of the equation ih—K(x.x'. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a..£(0) . We see that the initial condition (7. x'){E . We can see the significance of the timeindependent Green's function for the complete problem as follows.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').t) since ih^K(x.x'.9) = i H0YieE^i%(x)^(x') = H0K(x.11) = = 3) = fdx'(H0 . The timedependent Green's function K(x.7) with the initial condition K(x.7. (7.E<®)xG(x.x.E^ of the eigenvalue. x'\ t) (7.(*'))*fr') Hi{x'))^(x') f dx'5{x .EM)X*(X) ( . (JL x'\ t) = H0K(x. . (7.x')(E .10) The contribution on the right hand side is the socalled perturbation contribution.0) = S(xx'). .E® (EE{0) Hi{x))m{x).t). which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei. i (7.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.
In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. 2nd ed. Green's Functions Can we add on the right hand side of Eq. see B. however.. Wiedemann [183]. Bleecker [33]. D. i. (7. Merzbacher [194]. (7.15) ''This problem and its circumvention can be formulated as a theorem. see E. (7. Instead of Eq.t) for the complete problem as follows. In general. Booss and D. Maharana.11) with a system of linear equations of the form Vi = Mijyj.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. .12) Equations of the form of Eq.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). All such methods of solution are based on the analogy of Eq.6) we would have (7.108 CHAPTER 7. (7. This will be different. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. that (E . Sec. J. if we assume that E^ = Ej. Examples in various contexts have been investigated by L. (7.e. then at best such a contribution would be something like cipi(x). But then the Green's function (7. integral equations are more difficult to solve than the corresponding differential equations. 17. But if we assume that E^ ^ Ei for all i.2 (not contained in the first edition!). H. The theorem is also known as Fredholm alternative. MiillerKirsten and A.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). In cases where an inhomogeneous contribution is given. W. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7.6) is not defined.x'.14) We can now see the significance of the timedependent Green's function K(x. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation).
t) = J2 fdx/^n(x)^*n(x')^(x'. H\En) = En\En).t) = where if(x^'.20) the Green's function K(x. (5. t)\2. (7. Comparison of Eq. (7. 1 See e. (7. 7. 109 (7.16) As usual.17) as initial condition is obviously W)t = Y.x'. Eq. x'. in different formulations. for the oscillator potential.x'.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. We use this in Sec.Q) = 6(xx').t) describes the evolution of the wave function from its initial value ^(a^O). E. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. i. Merzbacher [194].7) with Eq.2 for the computation of the sojourn time. We can write this expression also as§ ip(x. p.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.The solution 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. t) with the Note that when K(x.t) = ^ e £ " f / « V n W < ( ^ ) . (7. t > 0.17) is the initial condition of \4>)t. we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.0). (7.5.158.x'. 2nd ed. *>0. e. (7.19) i/>(x.18) or.15) which contains (7. \En)(En\ip)t=oeE^\ n (7. x'.e.0)eEnt/in..g. this relation provides the probability density \ip(x.g.20) is the timedependent Green's function.' which obviously satisfies the initial condition K(x.7.22) According to Eq.t) is known. (7.21) dx'K(x.t)il>(x'. (6. .
7.23) along the contour C in the plane of complex E as shown in Fig. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. i. we obtain then also the corresponding density matrix.t). we obtain / W : = .E .x'.x') = GE(x. Green's Functions difference that (5 = 1/kT plays the role of it. x')6(t) (7. 24 )  ReE Fig. t) = J2 e £ " f M i ( x ) < ( x ' ) . 7. In the following we shall derive the Green's function for the case of the harmonic oscillator.e.1. between (with E^0' = E) G(x.x') and K(x. t > 0. x'.110 CHAPTER 7.i /f e «/.«^«2 W .x'. (7.t) K(x. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.1 The contour of integration.25) in agreement with the timedependent Green's function . Jc 27T „ hn . Inserting for GE+ie the expression above.ie (7 . With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.
B m0 n _ 2ih 2iK (7. Thus we try the ansatz.— d2 —2K(x.t). A and B being constants.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz. It is clear that it is nontrivial to solve an equation like Eq.e. (7. which is moving in one space dimension.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.26) This is the case of a free particle with mass mo.x>. and identifying coefficients of the same powers of t on both sides.30) Inserting Eqs. / dxK(x. (7. (7.30) into Eq.0) = 1.29) tJJ 2 2B(x .X>.31) The constant A has to be chosen such that K(x.28) B(xx')2 t2 DB(xx') 2 /t (7.e.29) and (7.x'.3 The Green's Function of a Free Particle 111 7. 2mQK h h2 ih(AB) = —{4B2A).0) = 5(xx').x'. v ' 2m 0 v h i. i. . In this case the Green's function is the solution of the equation d hK(x.7).t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.x'. (7.7.3 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. K(x. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7. we obtain *I£I = h2 —(2AB).t) t h2 = .27).
21).ihMxWn{x').t) into Eq. 1st ed. (7. j3 = i(xx').0)oce"Qx2+ifcox. n For a free particle moving in the onedimensional domain \x\ < L — oo.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. *>0.t) = — / dkeak2+Pk 2vr J. (7. (7. we have This is 1 provided A=JB= ^ (7 .oo 27T (7.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. (7. p. (7.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—.35a) K(xy. 158. 32) It follows that x'.112 CHAPTER 7.(7.H We can insert the expression for K(x..33).36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS .zEnt.x't) = Y. i.e. Green's Functions For parameter values such that the following integral exists.t) — for instance for a wave packet given at time t = 0 of the form ^(x. v27r En^^. we > have to make the replacements V ^ fdk. from K{x.x'."737) \27rii/i in agreement with Eq. "See the excercise in E. Merzbacher [194].38) . t) = /jJo_ e mo(««') 2 /««. 2m0 (7. Mx)^M^) „ n so that J =^ .34) K(x.20) and can then obtain ip(x. (7.x'.
x'. —KK(X.37) will later be obtained by a different method — see Eq. I (7. 1 2mo u>n ox n with the initial condition (7. HQ = ^—p2 ZrriQ + \m0uj2q2. .t).0) = 6(xx') at / = 0.X .x'./ ) + fK(x. mQLu .25) — in the context of Feynman's path integral method.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. . . z .7.x usn ot i.t) 1 + m0u2x2K(x.40) becomes 2 . (7. the relation* 6(x) = o<y(os).7) is the solution of d ih—K(x.33).x'..8).t) We now set h2 d2 = . i.40) Then Eq. We rewrite this initial condition in terms of £ and use for a = const.39) In this case the timedependent Green's function K of Eq.t) = h2 2 d2 TW .t) + ~x K(x.. . .——K(x. (7.t). x'. x'.42) .e.3 The Green's Function of the Harmonic Oscillator 113 The result (7..e.d r. i. (7. 5{ax) = ± J eikaxdk = ArS(x). (7. / ) (7. —ih—K(x. §jK(x. 7. K(x. (21.e.e.x'. / ) = ~^K(x. x'. i. . We consider the onedimensional harmonic oscillator with Hamilton operator HQ. (7.x'.4 Green's Function of the Harmonic Oscillator The next most obvious case to consider is that of a particle subjected to the harmonic oscillator potential. in this domain *Recall that S(x) = ± f eikxdk.
b' = Aab.50c) (7.50b) (7. so that for / — 0: > a^—. we obttain a' = l . and we expect K(x. P.b2.f) with a(0)»y.t) become similar to expression (7.33). (7. Green's Functions to the particle behaves almost like a free particle.e. (7.47) We insert this ansatz for K into Eq. Integrating Eq.45) in accordance with (7.4a&£ + 2a . 50. as the limiting case / — 0. (7.^ .49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7. . W ) . that is.x'. Feynman [94].42) and obtain the equation (with a1 = da/df etc. it • is suggestive to attempt for K the following ansatz^ m. (7.45) in this sense. (7.114 CHAPTER 7. Identifying coefficients on both sides.4 a 2 .) a'£2 + b'i + c' = (1 .47) becomes (7.c o t h 2 / .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.e. f i. (7. c(0)^. i. 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.50a) we obtain (7.b2.. (/ = 2a.Z'.48). we must have /o = 0. a = .4a 2 )£ 2 .50a) (7.46) If we interpret Eq.51) See R. p.
(7.7. in particular for the derivation of the sojourn time in Sec.51) for the density matrix PN(X.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. if we return to x. (7.i) (7.40) of the timedependent Green's function with Eq. 174.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . as one can verify.50b) Kf) = —r^77) smh 2 / A independent of / . (7.x'.* With this result we have another important quantity at our disposal. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x.e. . i.27).In B. 2TT^ Inserting a(f).47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. as we shall see in the following.55) 'For an alternative derivation and further discussion see also B. (7. we must have A = ('.48).t) = . / —————rexp y 27rmsin(u. (7.c(f) K = 5 =exp Vsinh 2 / into Eq.48) 6(0) = — £'/2f. (5. i. with A.0). we can use K(x.X'. In order to satisfy Eq. K(x. x .e.53) or. we must have (besides A = —£') = /m 0 u.b(f).45).54) For t —• 0 this expression goes over into the expression (7. t.t) to obtain this element (x. Comparing the Eqs.x'. c(0) = £' / 4 / . Felsager [91]. B independent of / . (7./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. (7.2. and to ensure that we obtain the prefactor of Eq. Finally Eq.33) for the Green's > function of a free particle.50c) yields for c(/). 7. p. x'.5. (7. 115 To ensure that in accordance with Eq.
p)dx W ) = —^r = —p . (5.. p. Feynman [94].x. (5. = jdxp(x. . Thus the system is in a mixed state and the expectation value "Cf.x. R.116 CHAPTER 7.e. For instance we have with Eq. Eq.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. <2 (7. Green's Functions With this expression we can evaluate (cf. ^—• Irpiv J pN(x. eP E i = i + y. 11 For T ~+ 0: u>0 .56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.» l. P.52)) the expectation value of an observable A in the canonical distribution (i.x.55). (7. 52. Eq.uij>0 >• 0.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.P) 2.52): Tr (PNq2) Jx2pN(x.. [ [ 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. we obtain § < Z> = ^ c o t h ^ ^ ° — . 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"}.56a) {q2) = = Y.56a) the expression (7. (5. at temperature T): {A) = Tl{pA) = ^ > . )x Inserting into Eq.e. we skip the algebra here.40)) 1 epEi 1 Ui = y.
QUJ (7.3 The Green's Function of the Harmonic Oscillator 117 (7.58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states.54).59b) n+ l \ ftwt here 2hf exp _ i ( n + . We return to the Green's function (7. In this case the Green's function K(x.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . (7. the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m. 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. (6. x'. This is the first and hence dominant term of the expression (7. which means in the oscillator state \i) with eigenenergy fku(i + 1/2).50) we obtain.7.59a) E 0 = ^fiw. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . and we replace to by LO — ie. setting a = mooj/h.J (w — ie)< e (n+l/2)Et e i(n+l/2)u. If we consider the system in the pure state \i). We assume t > 0 and t —> oo. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. of K(x.54) is K(x. t) of Eq.x'.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. e > 0. 1 mo^ exp lmow / 2 2 ft X (7.e.21). x'.56b) is that with respect to this mixed state (whose cause is the finite temperature T). i. 2 + x'2)le^2xx' to—>w—it t—>co 2 .
lie on surfaces. (7.1 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)]. which are planes. the particle will stay there only for a finite length of time T. as in Eq. the velocity of planes of equal phase.. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0.1 we estimate T semiclassically. i. is defined as vv = .x'.5. (7. The wavelength A is given by The phase velocity v<p. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. of dpN as pN(x. Green's Functions For t large (i.59a). 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. to infinity) the contribution with n — 0 dominates. k = k. i.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7.e.62) . (7. Considered classically.e. and with this we estimate the sojourn time T. 7. In Example 7.118 CHAPTER 7.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. However quantum mechanically in view of the uncertainties in position and momentum.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. In a very analogous manner we obtain the solution of the equation for the density matrix.e.
Q ( k .e.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. (7.7. e . i. We now ask: How and under what conditions does the time variation of the function ^(r. the fundamental postulate on matter waves in Chapter 2): E = hu.k ') 2 .e. xt— + — = 0. essentially again a Gauss curve.Jt + a. (7. ^ ( r . If we assume for / ( k ' ) a Gauss distribution.e. for which \ip\ assumes its largest value: OO .)¥'(*:")) +ism(<p(k') . e. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. to the function ip given by oo / fik'y^'x^dk'. then ^(r. i ) = /*/(k')e i(k '" r '' . /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. The wave packet describes a wave of limited extent.63) The relation u = w(k) is known as dispersion or dispersion law. / oo J—oo . i. a > 0 . t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. u = w(k).)ll/(fc")l[cos(¥'(fe.65) Let f(k') = \f(k')\eia and <p := k'x . or dco' da The centre of mass determines the particular phase. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. oo (7. t) is the spatial Fourier transform of this Gauss distribution — as we know. duo' da .64) where / ( k ' ) differs substantially from zero only near k' = k. ' t) dk'.<p(k"))}dk'dk". i.
we obtain the deBroglie relation p — hk. = — (768) It is instructive to consider at this point the following examples. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . . (7.fc w> vv = ~a^ = S r a d P E «i7 Sd u .^ g=r a rad. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. In the present case. i. as claimed for =0. p = hk the group velocity is equal to the particle velocity v. For E = hio.— g(k). a > 0. 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 = . Solution: The function to be calculated is the integral dxeax OO ezkx.120 CHAPTER 7. The threedimensional generalization is evidently du .. we can use a simpler method.69) In general one uses the theory of functions for the evaluation of this integral. . dE v 9 = . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. (7. Example 7.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)].e. however.
MiillerKirsten [215].74) Example 7.a) + 6{x .69) and (7.. (7.73) 27T 7 . We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . 2 f + . Appendix A. (7.75) ' E .73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . Example 7.69) and (7. where c = 9(0) is a constant. (7.72) Solution: From Eqs.r „• With the help of Eq.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.71) (7.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. (7.0 0 7T 7 0 7T e From Eq.Afc = 81n2.7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . W. ... \a — b\ (7.&)] for a + b.7.70) we obtain 6(h) = lim — f°° dxeax2eikx = . J. (7. (7. with partial fraction decomposition.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .2: Representations of the delta distribution Use Eqs. XZ (e > 0). (7.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x . g ..1 verify the uncertainty relation Aa. or see H.
* o ) 2 / 2 .77b) (7. Green's Functions 1/(27i)1/2a — 1/2(2. (7. at time t = 0). i.« 2 ( * .e. A simple calculation yields A i = 2v/2~In2a.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.70) the Fourier transform of f(x) is (7. is Afe = 2 V 2 1 n 2  (7. (7.78) It follows that the product of the uncertainties is AxAk = 8 In 2. In quantum mechanics the square of the modulus of the wave function ip(x. The wave function corresponds to the function f(x) in the above considerations (e. when Ax is very small. and hence implies large values of Ak.e. where  / ( x )  = m a x  / ( x )  / 2 . Solution: In Fig.) leads to a very broad maximum of the Fourier transform \g(k)\.2 The Gaussian curve.77a) small values of a.CHAPTER 7.g. Physically . The breadth A/c of the curve g(k) around k = kg.r)1/2a Re f(x) Ax Fig. the width Ax is a measure of the uncertainty of the probability. Thus a slim maximum of the curve of /(a.77a) g(k) = e . requires according to Eq.t) is a measure of the probability to find the particle at time t at the position x. i.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. 7. 7. According to Eq. where g(k) = max<?(fc)/2. Thus a sharp maximum of the function f{x).
E. at time t = 0. O ) . This means the particle is localized at x = 0.81) *G. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. (7. (7.e.. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.u> = l. The function g(k) is therefore described as momentum space representation of the wave function f(x). 7.54) with the substitution ui — iu.". many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . ^In Eq..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. the less precise is the determination of its associated momentum.2 ) and U)t. (7. The result (7.7.71) > and (7. Barton [15].h = 1. which means all values of the momentum are equally probable. Barton mentions at the beginning of his paper. In the limit a — 0 we obtain from Eqs. For reasons > of simplicity in the following we set in addition t mo = l.x'. 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\ (7.5. <7*» We assume that initially. Barton recalls that W. (7.75) implies therefore: The more precise the coordinate of a microscopic particle is determined.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x.«> = / ^ C ' W . so that Ki0(x.79) According to Eq.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. . nr = S(x). In the same limit the momentum uncertainty Afc grows beyond all bounds. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. Thus G. i. Lamb at Oxford set the following problem in an examination in 1957: "A pencil is to be balanced so as to stand upright on its point on a horizontal surface.2 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.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .
. . = .124 CHAPTER 7.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*. Green's Functions Inserting (7. ' 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) = . 9 (7. 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 . + cosh 2 1.86) Evaluating the Gaussian integral of Eq.2 (7.t) = J oo P .88) .84) we obtain* ip(x.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7. (7. (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh ..82) i/f(x.85) = —r= o2 7 2 sinh 2i.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.83) ip(x.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.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 (7.79) into (7.cosh t sinh t' Wisll2b ^ (7.81) and (7. A .
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. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.t)= where Co(t) / dx\^(x.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. In taking the modulus of ip(x.4 The Inverted Harmonic Oscillator The real part of Eq.94) / dxf(x)=g'(y)f(g(y)). C2(t) (7. •&(t)=//c(t) = / or.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.91) C(t)' (7..t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .t)\' 2 125 (7.89) = exp[x2/C2(t)} 7rV2C(t) (7. (7.t)\2 Ji == 71 / V " Jo dO ? 6' (7.93) where —{dQ/dt)5t is the probability that the particle leaves the vicinity of the origin in the interval St around t (partial integration leads from the second expression back to the first).£ 2 (7. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.95) .
=0 V^ C'(t) I Jt=0 V^ (7. B.99) .97) sinh 2 1 = . p. we obtain dT _ 1 ~dJ~l 'See for instance H.(4 2 .96) °° l Jr. + 0(e'2) (7.2)(p_7?2)(7.1 ) 2 \n J ^"•""ff^^V^K?. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. 136.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.98) We insert this into Eq. C(t) = .From this we obtain C(t) C'(t) Z2 ^(r?2_.126 We set I C(t)' so that dT ~d~l Now. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr..=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.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. Dwight [81]. (7..= \ / l + 2sinh 2 t.
The equation there describes the classical "rolling down" of this tachyon. Zwiebach [294]. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down. also motivated by this paper. Explain the parameters entering the calculation. 238. 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. See e.g.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7.102) 'As a matter of interest we add that in the theory of a free scalar field x. It follows t h a t T~lnZ. (7. this is T ~ ]n(ly/mou>/h).7. U) (7. p. this equation is the equation of motion of this scalar field with negative masssquared. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. 7.100) Reintroducing the dimensional parameters we had set equal to 1.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. Fig. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. B.4. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". T h e following Example 7. .
so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). For h —> 0 the time T —• oo in agreement with our classical expectation. i. (10. A p are defined by the mean square deviations given by (see Eq. (7. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). and solution x = A cosh Lot + B sinh cot. i. ^(x2)(p2).128 with x2 = LO2X2 + const. 2V ' Let A and Bio be the values of x and x at time t = 0. v A x A p = \ (x2)(p2) v = m0u(x2). Green's Functions x = Aco sinh cot + Bco cosh cot.e.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.g. (x 2 ) h 2mou 2mou> and so l^J^ULl). but instead are subject to an uncertainty relation of the form AxAp > h. / 7 ^ 2 =2J(x }. and for simplicity we take (cf. Eq. so that e. so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. Ap= v /(p 2 )<p)2.5)) Ax = <J(x2){x)2. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. . 2 10 x(0) = B u . 2 x(0) ^ where A i . Thus quantum mechanically x(0).e. x(0) = A. (p2):=mW(x2). (7. CHAPTER 7. Quantum mechanically x(0). the smallest macroscopic length.p(0) cannot be determined with arbitrary precision. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I.102)) p(0) = ^/(p 2 >. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p).
but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4. Frequently even the calculation of the next to leading contributions ip^. in closed form) only for very few potentials.E^ is already a bigg problem.e. The perturbation method generally described in textbooks — and frequently called RayleighSchrodinger perturbation theory — consists in assuming power series expansions for the wave function * and the eigenvalue E in terms of a parameter like (3 which is assumed to be small.e.(i)+/^(2) + ".1 Introductory Remarks The Schrodinger equation can be solved exactly (i.Chapter 8 TimeIndependent Perturbation Theory 8. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.2h2 cos 2 a # = 0. 129 . In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. one would set tt E = = ^(°)+/ty. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. i. In general perturbation series do not converge. For mathematical purists the question of convergence of the series is an even bigger challenge.
at the end of Chapter 26) — and by comparing the results of WKB..3) *See e.2 Asymptotic Series versus Convergent Series Before we actually define asymptotic series we illustrate some of their characteristic properties by considering specific examples which demonstrate also how they differ from convergent series. It will be seen later (e.g. Schafke [193]. 581. the expansions in descending powers of h2. Meixner and F . E.= 0 < 1.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . 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. T. (8. which is known as the Mathieu equation. h hl are in the strict mathematical sense divergent. p. 8. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). (8. N. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •.2 However.e.. see e. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. TimeIndependent Perturbation Theory In the case of this equation. .* i..+ . Whittaker and G. 0. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x.g. W. Watson [283]. . ^For further details. This convergence can be shown with D'Alembert's ratio test since''' lim . J. In the following we explain the difference between convergent and asymptotic series.g..g. E = E^ + h2E^+hAE^ 71 72 7 + .130 CHAPTER 8.2) n—»oo n It is interesting to compare the behaviour of the series (8. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . e. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. concerning also the uniform convergence of the exponential series.
000 000 06 + • • • . Thus e. +0. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. and in fact increase indefinitely.. 235.0.22222 . As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!.g. It is evident that the larger the value of \x\. . i.22222 .33333 . ...4938. + 0..3) with the convergent series (8. We observe in the first place that since lim n—>oo co (8. However. T. Whittaker and G.2. / ( I ) = 1 — 1! + 2! — 3! H • However. E.. begin to increase again.222. /(1000) = 1 .8.0. . the series (8. We can see this as follows.0.0.e.29629. p. + •••.3) diverges for every value of x.l n ( l + ^ ) n=l L ^ ' (8. and we see that the moduli of successive terms first decrease and then. in spite of this the series (8.376 .. for x = 3. . . From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ ..0.000 002 00 . the better the approximation obtained.1). 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. 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. e.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 .001 + 0.4) for every value of x. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. N.9876. Watson [283].3) can still be used to obtain almost correct values of f(x) for x sufficiently large.... + 0.5) Cf.. for larger values of x. Comparing the asymptotic series (8..g. 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. .
Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. 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.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. T. In fact. Vol. Whittaker and G. N. Watson [283]. 11 L. I.J In2.41) and (16. (16. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z.132 CHAPTER 8.e. **A. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). This is not appreciated by mathematical purists.5). p. ^E. the leading term of an asymptotic expansion. TimeIndependent Perturbation Theory where 7 = 0. Schiff [243]. it is sensible to make the assumption. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. meaning convergent. 152." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. It is interesting to observe the comments of various authors on this point. 236. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. in fact. 16. 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. In many respects Messiah aims at more rigour in his arguments. 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. that E and \ijj) (i. Sec.1 between Eqs. § However." 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.5) are practically unknown. ip) can be represented by rapidly converging power series in e. 371. whereas applications of the convergent series (8.1. Messiah [195]. E. . Merzbacher [194]. p. This latter observation hints already at the importance of asymptotic series in applications. p. the dominant approximation is. although this has not been investigated except for a few simple problem^.
" Of these three authors t h e first.* and we follow some of t h e considerations given there. Vol. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . (8.. 198 (German edition).10) Vn Changing t h e variable of integration to u = t2x2.8. *R. In fact. (8. p. He says.. there is no need for such bias.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . Replacing t h e exponential in Eq.2.^=e~x2 / e^^dt.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. II.= / e~t2dt. ft Jx du = 2tdt.7) as / e'* dte* dt = 1 . It is therefore suggestive to write iy poo <t>(x) = 1 ..1 T h e error function a n d Stokes discontinuities Another extremely instructive example which illustrates t h e n a t u r e — a n d in addition t h e origin — of asymptotic expansions is t h e error function 4>{x) denned by t h e integral <f>(x) = 7= / V71" Jo 2 fx e _ t dt.7) 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. We know from books on Special Functions t h a t all of these functions. Dingle [70]. 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. (8. (8. we can rewrite Eq. which have been studied in great detail. (8. Messiah [195].11) A . solutions of second order differential equations. B. Merzbacher. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. 2 (8.7) This function has been considered in detail in t h e book of Dingle. .
since (—4)! = y/n.e.' we have 2 nH 7T 7T (8.12).14) sin7T2.^e~x \Ar i.134 we obtain CHAPTER 8.15) < 1 we can insert this expansion into Eq. Using the reflection formula (*!)!(*)! = for z = n + i .16) . we obtain 1+ i.^ X e U ~ ( 1 + * \ 2 . (8. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . 7T (ni)!sin{7r(n+^)} ( „ . « \~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 (8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8. TimeIndependent Perturbation Theory <P(x) = 1 .e.13) presupposes t h a t < 1.i)[(_!)«' Then. (8. du. Then e (n —\)\( 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.
(8.—yS!L^LLZ e x OO / _ 1\ X2 /CO dueu L \ / V n = Q 1 and we write 0(X)~1 V 7 n=0 v 2^.12) ignoring the fact that the latter is valid (i. i. (8. In the foregoing discussion of the error function we assumed that x was real and positive.l)! = / Jo Then 2 cHnxdt. i. We can actually understand the deeper reason for this in practice.r<7r. whereas in a convergent expansion the terms first increase in magnitude. 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. It is fairly clear that the above considerations remain unaffected for argx < 7T. argx = 0.2 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) originates.e. We now want to relax this condition and allow for phases argx ^ 0. and in fact so slowly. We now see how the asymptotic expansion (8.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. convergent) only in the restricted domain u < x2. As in the case of the gamma function. much more useful than the convergent expansion.19) implies either that we ignore the remainder or the correction term given by the integral in Eq. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. Whichever way we look at the result. The expansion (8. /•oo T(n) = (n .18) or that we insert the binomial expansion (8. In an asymptotic expansion the first few terms successively decrease in magnitude. Frequently "=" is written instead of "~".e. .e. in general.19) approximates <f){x) the better the larger x is. ' arg.15) into (8. (8. the asymptotic expansion of the error function is.17) 0(X) = i.8. (8.
=dv.vds. thus with \v/y2\ < 1. VK Jo V^ Jo (8. We therefore consider this case in detail.2 y V .21) = y2s2. when argx = 7r/2.24) We can read off the binomial expansion of the factor in the integrand from Eq. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.^ f V e ^ . (8.e. (8. (8.15). TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 .26) or else return . V7T Jo Changing the variable of integration to v (8.136 CHAPTER 8. The situation becomes critical. and we can write (j>(iy): y^ Jo t^Q n] \y J If we want to proceed with the evaluation of the integral in order to arrive at an expansion of (f>{iy). (8. (8. dv = 2sds = . we set x = iy in Eq.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds. we can try to proceed with Eq.23) Throughout the range of integration 0 < v < y2 the integrand is real.22) we obtain 2 4>{iy) = . i.20) where we set t = is. (8. however.^ = .
28) n=0 We observe that this expansion differs significantly from (8.^] * dv.27) and (8.28) we observe that the expansion (8. (8. (8.8. These integral contributions are of order 1/y since the integral behaves like e~y and is multiplied by a factor e+y in front.28) because then the correction term oc f°£ becomes smaller and smaller. e ^ ( i y ) ^ ^ . and so the integral is real. y y 2 ( n _i)! y arg(* = iy) = i * .2 Comparison of Asymptotic Series with Convergent Series to Eq. Returning to Eqs. i (8. We know that for v < y2 the integrand is real. the better the approximation expressed by Eq.24). — (8. However. . for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part. Eq.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv.27).e.24). which will be studied in more detail below.e. If we proceed with Eq. (8. We have therefore <j>(iy) = ^e^TL fV e~v ( l .28) ignores the integral contribution of Eq. for large values of y.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! .27) Then . (8. (8.19).17): J£° e~vvndv = n\. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . (8. i. (8. (8.2 yir (8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .29) where 5t means "real parf..^ ^ p . i. ey is much larger than something of order 1/y. Suppose now we consider Eq. and the larger y.30) yvn f Jo V y J Cf.
2 OO rz=0 V . This result is identical with that obtained previously. (8. .33)) would have • a maximally increasing exponential e+x .8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. R.28).138 CHAPTER 8.31).30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.31) and (8. Dingle [70]. and • higher order terms of the associated series all have the same sign. (8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i.3!) which is (8. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. (8. *Cf. y.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. since (8. B. e xZ ^ (n — £)! „ ^ ) . we use the expansion outside its circle of convergence and hence obtain a divergent expansion. Eq. (8. (8.e.26).e. Chapter 1." — E ^ ^ forarg.25) and integrate from 0 to oo. We have therefore found that XTT n=0 ^ ' and . 7r = . i. Looking at Eqs. With a similar type of reasoning one can show that the same expansion. Since (cf. . is obtained for <f>(x) with arg x = —ir/2. TimeIndependent Perturbation Theory If we insert here the expansion (8. (8. The corresponding phase lines are called Stokes rays..* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself.
though.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. in the sum of the right hand sides of Eqs. R.2 Stokes discontinuities of oscillator functions Dingle' has formulated rules which permit one to continue an asymptotic series across a Stokes discontinuity without the necessity of a separate calculation of the asymptotic series in the new domain. Dingle [70].37) Cf. W.32) and (8.e.8. See also Tables of Special Functions. Magnus and F. apply predominantly to asymptotic series of the solutions of second order differential equations. and so may not be universally applicable. p. . (8.e. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. Eq.13.36) 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. pp. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . (6. for arg x = 7r/2. —1 cancel out. 9 . 1/1 ipv{x) v1 vl (±ix).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. i. B. Oberhettinger [181].g. Thus if we set x = XR + ixj. '"Cf. These rules. 91. Equation (8.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. i.35) 8.2). e.34) the contributions 1. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. Chapter 1. 5.2. 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. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x).
The continuation of (8. 9) and XXI. 0 < argz <  (8. 8. B. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. (8. ti asymptotic series (8.140 where^ CHAPTER 8. We also observe that the late (large i) terms of the series in (8.39) (as in the example of the error function). Chapters I (in particular p. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a.e. (8. > > The equation (8. i.7. = 7r/2. Dingle [70].40) § Cf. . v — — f — 1. as can most easily be seen in the case of the Mathieu potential. x~l/~1ipu(x).e.38) are of the type (2i)! i\ i\(2x Y 2 (2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series. 1 1 Cf. Eq.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2.x 2 oo «*)  — E^fi=0 v .e. TimeIndependent Perturbation Theory i. Chapter 1.36) is invariant under these replacements. R. (14). Dingle [70].ei<v+VxvHv{x\ argx = £ . Hence Dil){x) = xv(j>v(x).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.37). R. Thus D{J\x) (i. i. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. Chapter 17. B. with a real proportionality factor. We consider Dv (x). 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. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign.
41) the Stokes multiplier a is. In (8.e. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. (8.e. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i. (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv). Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga.44) + ^iap) = 0 .42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).e.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. with \x\ = (—x) for arg a. i. < 7T. i. The value of Du (x) on the other side of the Stokes ray. < 37r/2.41) to arg a.e. as yet an unknown real constant.42) T ^(eiwu i. (8. beyond arg x — 7r/2. = ir. 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. where in the first line on the right hand side the second term contains the real proportionality constant a/2. for xj = 0) and therefore possesses a Stokes discontinuity there. (8. we must have at arg x = 7 in the dominant factor on the right of Eq. On reaching ir the part containing ^ ( x ) . Since Dv (x) is real when x is real. = ir.41) xuct)v{x) + a{x)~u~lipu{x).8.40). 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. 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). another half of the discontinuity appears on leaving the Stokes ray on the other side. We therefore proceed to continue (8.
v . .—^ ON • — ^ > + ^ ^ L .45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.1) = 2sin7ri/.e.1) (8. (z)\(ziy.14).45) that af3 = 2sin(7r^)) D^\x) = (8. Eq. i.42) that a multiplies ipv(x). (8.1)! TT.e.l)0(v . (8.51) .46) For a given function a(y) this equation determines /3(u). or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. i. Hence the left hand side must have the same property. Dl \X) is (cf. (8. i.e.48) We compare this with the reflection formula (8. (8.1).45) is therefore given by a{v)a(v . (8.1) = ±a(y).42)) for argx = ir (in the second line of Eq. i.1) (8.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .47) Equation (8. argx = vr. (3{v . (8. TimeIndependent Perturbation Theory Now a and (3 can still be functions of v.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 .142 CHAPTER 8. 8. We can see that the equation is satisfied by (3{u) = ±a(u .42) we insert from Eq. (8. a{y)P{y) = a(u .e. V i=0 \U0 v 2w . a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .
2..*"ei°> f ( ^ _  _ L _ = (!)««(. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .. i. However.)(.52) (argx = 7r).. ' ^ (i/l)!z! ^ . i.1 ) ! 4~L i\{2x2Y i=0 V2^(x). Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. (8. (2x2)1 = COSTTU(X) v .7).1 ) ( . (8.e. (8.I / . the second contribution vanishes and we obtain £>«(*) .„ _ix2^(2ivl)l e 2 > —.e^ . (8. —f. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.e.8. (8.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator. For details we refer again to the book of Dingle.50a) because this enables us to define normalizable parabolic cylinder functions.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8.36)) in the domain 0 < arg x < IT. Eq. 8.36).e. i.. u 1 y> {2i + v)\ 1 n 0. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di. the error . Chapter I.. It should be noted that in Eq. We observe that for v — n = 0. solutions of Eq.7r). ' (X) (cf.35).3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. which vanish at x = ±oo.A / ± sin^Cx)"M* 2 V ^ ..1.) (8.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.
e.38) can be obtained.60) into (8. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. and demonstrate.36)) 0 where + x2Q(x)y = 0. x. S2.54) we consider again the important equation of the harmonic oscillator. % = {.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8.55) The large x asymptotic expansion can also be obtained from the differential equation. i. Eq. previously we had ^.^ + 1 ) . cf. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. (8. x°. we obtain equations from which the coefficients So.54) (8.36).60) 2 x x l Inserting (8..59) and equating to zero the coefficients of powers of x2. We write Eq. how series of the type (8.. S\. 0. i. (8. Solving these equations we obtain = = 0. ^ + 2 . (8.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a. x—>±oo (8. (8. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . • • • can be determined..e.37).36) in the form (coefficient of — x2 here chosen to be ~. Instead of dealing with Eq (8. Eq. 5„=4 Sl=o. \ + U (8.144 CHAPTER 8. (8. + — + ^ + • • • .
Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). p.b. In the following we are mostly concerned with asymptotic series in some parameter. 2 X 2Wi. Magnus and F. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ . a z a a ( + 1) z2 which has the unit circle as its circle of convergence. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. function.e.38).91.63) **W.19). 4~l~2 ' i i _i.61) (the dominant terms of (8. (8.37). Nonetheless we shall need asymptotic expansions like (8.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. (8. etc.37) with (8. and it is clear t h a t higher order terms follow accordingly. therefore.62) 4 \ 2 XiFi 2'2 .+ X M W = Q. » 1.z) is a confluent hypergeometric T. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains.35) and (8.36) have correspondingly e±x / 2 ) .8. do not enter here into their derivation and simply quote the result: W i t h \x\ . i. T h e expansions (8. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . Oberhettinger [181]. / i N . . n 22e"T JziL (8. are asymptotic series in a variable x. The derivation of the asymptotic series from integral representations (cf. We.38) in a variable for the solutions of approximated differential equations.
66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2.65) 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. whereas e~x '4 increases exponentially.146 CHAPTER 8. Watson [283].67) Cf.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.4x2 (8. Whittaker and G. then the series is said to be an asymptotic expansion of f(x) in t h a t range if for a fixed value of n lim \x\—>oo ft xn\f(x)Sn{x)\=0 3. Thus 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. 349. E.64) (c) for —7r/4 > a r g £ > —57r/4: Dv{x) e 4x !*£=£2 2x i/i + +• (I/1)! + xe 4 x (i/+ !)(„ + 2 ) 2x 2 " ' (8. 4 (8. T. 8.+ — + ••• of a function f(x) for a given range of a r g x .1 a2 an a0 + — + ^ + . If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . top of p. N. .0. Thus there is no contradiction^ between (a) and (b).
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". 8.69) and the consequent nonunique definition of the expansion. The definition is due to Poincare (1866). Chapter I. i.73) Vn + q/#> + e2T/i2) + • • • .68) is (effectively) simply a statement of the observations made at the beginning. (8.68) The definition(8. (8. A critical discussion of the definition can be found in the book by Dingle [70]. which is assumed to be small. we set E^En * > ^ n = En°1+eEnV = + e2EnV + . In fact. lim xneW x—>oo = 0 (8. first published in 1902. the divergent nature of asymptotic expansions together with a vagueness of definition (which is avoidable as explained by Dingle) frequently tempt mathematically prejudiced purists to turn away from asymptotic expansions.8. possesses a whole chapter on asymptotic expansions.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .g. (8. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics.Sn{x))  = oo n—>oo (8.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e.5 RayleighSchrodinger Perturbation Theory RayleighSchrodinger perturbation theory is the usual perturbation theory which one can describe as textbook perturbation theory as distinguished from other less common methods. in particular with regard to possible exponentially small contributions to the expansion for which e.e.70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (871) / dx%l)*m(x)lpn(x).
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. °° i J or 0 = EWJdx{rnVil>n). ^ ) = 0.e. (8. Ho^+V^n = EW^+E^n. i.V)i.n = 40)V>n.As a consequence the state vl/) is a vector in this space.i4 0) )V4 1} = {EM . TimeIndependent Perturbation Theory We insert these expansions into Eq. (879) We multiply this equation from the left by ip^ and integrate over x. (8. so that we choose / • dxrpM^O. The second equation can be rewritten as {Ho . (8. i. We obtain then the following set of equations: Hrj. (8.76) The first equation is that of the unperturbed problem. With the help of the orthonormality condition (8. We insert the ansatz (8. (8.148 CHAPTER 8. i. ( ^ . we write in the position space representation ^ ^ E ^  (878) A contribution to ipn can be combined with the unperturbed part of the wave function \P. J f°° dxrn E(^ 0) .75) HoW + vW = E^+E^+E^.74) (8.e.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.n.4 0 ) M = / ^C(4 1} .78) into Eq. .w » .e.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .77) The states \tpn) of the unperturbed problem span the Hilbert space"K.70) and equate on both sides the coefficients of the same powers of e.
Vil>n) (880) In order to obtain ipn .Viprt e (8. (883) Thus the procedure forbids the equality of any E\ '.76) which we rewrite as (H0 . (8. the coefficients a\ . the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates.82) *> = ^ + E . i.8.83) makes sense only if Ef] ^ EnV for all i + n.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. (8.e. i. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution.^(EW ~ V ) ^ + E& / J—oo dx^miln . (8. We therefore consider Eq. S ' ti^+Q^)• We observe that Eq. We multiply the equation by ipm. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ .84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above). It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.V)^ Setting + 42tyn.m ^ n and integrate: oo /•oo / oo dxrmY.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n. we return to Eq. (il)i. with En . multplying Eq.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi.i ^ 0.4 0 ) M 2 ) = ( 4 1 } .79). (8.e. ^ n ) + 0(e 2 ). (8.
89) • • • where we guessed the term of 0(e 3 ).2 (V> m .Vlpn)(lpn.Vlpn] M _ P (0K 2 ^r .£i°») 2 Hence +^£o (£«? . Inserting the expression (8.^w .V'>pTl (E^Eny V'fc (8. n) £f) (8.V^n) . En0) (8. j. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.87) (i>k. (8.Vlpn) E^ .Vlpn)E.V^j)(^j.^(8. (8.(1) (i&0) .V4>n){lpn.^j)(V .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' . V ^ i ) X (Q) (Vi. ipn ) = 0 i.88) (Q) Inserting Eq.79) •/—oo oo / Using (8.81) we obtain (2) _ _ {lt>m.B i 0 ) ) ( £ f .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.^ ) + $& w £ (^fc.150 CHAPTER 8. ~Ej E E. ^ n ) j&iift + ^n .86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .VV .85) together with (8.e. (8. FVn) + e 2 ^ ( V > n .4 0) ^+ e 2 E (lpk.
g.n(x) = J2crtn+ii (8.lpn+i) = ^ Cj5k. i. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations.89).. We then obtain (lpk.8.91) The coefficients Cj follow from the recurrence relations. We can therefore rewrite tpn ': The perturbation theory described in Sec. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).Vlpn) = y^Ci(lPk. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. Merzbacher[194]). 8.90) is the term with only one summation J2 m the contribution of 0(e 2 ). The functions ipn(x) are eigenfunctions of some secondorder differential equation.7 is based on this procedure. .e. (8.90) can be found in just about any book on quantum mechanics.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution. An ugly feature of the expansion (8. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i.. One would prefer to lump it together with the other contribution.
97b) are .97a) and (8.93b) The lowest order eigenfunctions belonging to En. (8.e.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . (8. is not uncommon. .93a).93a) (8. Em can now be linear combinations of ipni'ipm.e. + c22^m).70). (8.4 0 ) ) ^ } = (Em] .95a) (8. (8.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . For example the magnetic states enumerated by the magnetic quantum number m. Equations (8.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. that of double degeneracy En = Em . (8. I — 1 ..95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. (8. in a central force problem all have the same energy unless the system is placed in a magnetic field. .93b). . Em = 4 ° ) +eE$ + e2E$ + . In order to deal with the problem of degeneracy we consider the simplest case.94a). —I. i.V)(c2l^n + ci 2 ^ TO ).6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. i. are allowed to be equal since otherwise the expansion becomes undefined. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn. (8. Equations (8.97a) (8. so that En = 4 ° ) + eE™ + e2EP + ••• .152 CHAPTER 8.ipm) = 0 = (^m.. (H0 + eV)^n and (HQ + eV)Vm = Emym. Equality of eigenvalues.94b) with (ipn.96) (Ho . (8. (8. .94a) into Eq. however. all n.We now insert Eqs. degeneracy.97b) with ipn y£ ipm. m = I. i. TimeIndependent Perturbation Theory 8. .ipn).96) are effectively the known equations of ipnitpm and so yield no new information.e.
But (8. ^ o ^ ) ) = 0.98b) + Culpm) + cuV^m). cn(ipm.m). (^m.H0xl>$) .98a) Setting I = n.(lPn.ci 2 (V„. (8.(^04°))41)) = EM^CuTpn (ipl. Vi/>n)] . We now multiply these equations from the left by ip* and integrate. .^ ) ( ^ m .y>m.V^n).C 21 (^ m .8. ^ ) = ( V .m. ^ ) ) = CU[EW .cnVipn and tyi.EJ®(lPn.£ < « = ag)^) . (^n. Equations (8.Vil>m)] V^m).(Vn.^ ( ^ . c2iV^n + c22V7pm).99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) . Wm.C22(Vn. Similarly Wn. (8.ipn n) = 0 = (ipm.J#»a&> = 0j since ii4i = En (degeneracy).6 Degenerate Perturbation Theory 153 inhomogeneous equations. V^n)\ .99) and ( ^ ^ O ^ ) " ^ ^ . and ( ^ m . V^n)./>£>) = C2l[E£> . £ aW^°Vi) . Vl/. t f ) = a k 1 } 4 0 ) " Eg>aU = 0.H0^) . Then (^. H0lpV) . H0lP$) = C22[E$ ~ (Vm. Vlf>m)] .ipm ) the equations (iPn.H0il>W) = cn[EW . we obtain with (ipn.
(8.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy.Vlpn) CHAPTER 8. splits the problem into two independent onedimensional systems. in particular in parabolic coordinates (see Sec. If one wants to perform a perturbation calculation in such a case.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. dx i. However. if)' (/>' — = —.^ ' V = — (•>!>'<t><t>'i>).E)tp = 0. d>" E=^.. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. i.^Vm) 0. ip = c<p. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). Example 11.C22.e.6.ci2. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. . A onedimensional system. \nip = ln(j> + c o n s t .154 and V (^m. 4.100a) determine the coefficients Cn. the degenerate perturbation theory must be used.Similarly Eqs. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. 9} ip. (8.E™ (Vn.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. Let ip and < be two eigenfunctions belonging to the same eigenvalue E.1. like that providing the Stark effect (proportional to rcosO).101) This secular equation determines the first order correction En • The equations (8.e. It follows that ip and <j> are linearly dependent. Then / > i>" 2— = V ip Hence 0 = V ' V . i. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r.5). It is then possible that a special perturbation. In this particular case therefore nondegenerate perturbation theory can be used (cf. At x = ±00 the constant is zero.3).e. (V>V</>V>) = const. Example 8.100b) determine C2i.Vrl>n) . TimeIndependent Perturbation Theory (l/>m. as we saw. i> 4> Hence there is no degeneracy. the secular determinant. . 11.
7 DingleMuller Perturbation Method 155 8. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. The subdivisions (8.103) such that if E is written correspondingly E = E0 + e. It is immaterial here whether D does or does not contain a first order derivative — in fact.V)(p = 0. Eq.* In various versions this method is used throughout this book. in particular in applications to periodic potentials. and such that one can even obtain a recurrence relation for the coefficients of the expansion. also possible in areas not of immediate relevance here.e)<t>. Applications of the method are. We assume that V can be expressed as the sum of two terms. Muller [210]. W. a potential. and the solutions then have to be matched in their regions of overlap. Muller [73]. Dingle [72] and H.102) where D is a second order differential operator and E a parameter. cf. with some restrictions on the function V.V0)<f> = 0 (8.8.e. J.106) (8. (8.104) "This method was developed in R. H. the method has to be applied in each domain separately. B. i. J. J.104) are chosen such that the equation Dcf) + (E0 .t Since in general a solution is valid only in a limited region of the variable. We consider an equation of the form Dcj) + (E . It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. Muller [216].V0)<t> = (v. .102) can be written D<f> + (EQ .103). Thus the method is a systematic method of matched asymptotic expansions. W. (8.105) (8. V = V0 + v. (8. anharmonic oscillator potentials and screened Coulomb potentials. B. of course. the considerations given below would be valid even if D contained only the derivative of the first order. (8. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. See also R.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. W. Dingle and H. T For instance spheroidal functions can be studied with this method.
m + s)" in order to relate them to "steps" from m to m + s. v = {D+EmVo)<t>m. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .109) D(j)m+s + (Em . Considering (frm+s we have.V0)<f>m+a = 0. the function 4>(0) = <t>m (8.m + s)4>m+s s (8.e\ < \V0 . TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m. Then if ]e < \E\ and \v .VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .m + . and so v (8.e ) ^ = 2 ( m .) This result is obtained as follows.+s ' (8.E0\ (the latter in a restricted domain). (8.156 CHAPTER 8.110) If we write the next contribution <f>W to <p(°> in the form m+st (8.107) represents a first approximation to the solution (f> of Eq. The next step is to use the recursion formulae of the functions 4>m in order to reexpress (v — e)4>m as a linear combination of functions <fim+s.111) s then (see below) 0(1 = E (m. where m is an additional integral or nearintegral parameter (depending on boundary conditions).V0)<f>^ .Vo)(j)m+s = (Em .102) with E ~ Em. m + s) which are determined by the recursion formulae. m + s)0m+s.Em+s)(/)m+s.V0)4>m = 0. We write the coefficients "(m. since D<f>m + (Em . s an integer: (ve)(pm = 'Y^{fn. the equation Dcf>m+S + (Em+S .113) .108) with coefficients (m. (8.
119) is effectively the secular equation).m) 1^1^ IW .TP \(T? .m + s)(m + s. (8.m) E J^ ( m ~ Em+S) ' • fa.117) „ .m) — 0 (first approximation). We then find altogether ^  l^m  Em+S) + y^y^ SJ.m + s)(m + s.n\ ^119) + .„N This procedure can be repeated indefinitely..m + s)(m + s. . (8.m)4>m in (8.T? S The first expansion determines the solution 4> (apart from an overall normalization constant). Thus (m. (8. V ^ V ^ {rn.8.Em+S) = 0 (second approximation).m) + > ^ .m) — —. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. . We observe that the coeffcients of expansions (8. /01_. .m + s)(m + s.113) must vanish.118). (8. (8.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. .119) follow a definite pattern and can therefore be constructed in a systematic way.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m. Eq. .102) to order (1) of the perturbation (v — e).Q (m.^ (Em .m + r)(rn + r.e. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s . the coefficient of the term (m.7 DingleMiiller Perturbation Method and so (m.116) m ^ v (m. (8.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.110) and are therefore related to the inverse of a . (8.
this is — so to speak — the price paid for obtaining a clear systematics which is essential. 1 . since it will be used extensively in later chapters. i.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. TimeIndependent Perturbation Theory Green's function. in which q = in + 3.2). for the investigation of the behaviour of the late terms of the perturbation series. S. § H. as will be seen later.[q(llq2 + 1) + 2(33g 2 . 2 . for instance. The result is the following expansion. This can in fact be done and will be dealt with later.6q + 1)1 + 24(5? .423) + Z(2720 9 3 . as already mentioned after Eq.I2q + 64) + 256Z 3 (41 g . . A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. as will be seen in later chapters. ^The potential A r 2 / ( 1 + gr2). Kaushal [146]. and is typical of a large number of cases. n = 0 .* Example 8. The eigenvalue expansion obtained is an asymptotic expansion. This means the solution is not normalized.1)( 2 + 64J3] — . An explicit application of the method (not including matching) seems suitable at this point. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. (8.[ 4 ( 8 5 g 4 + 2q2 . W. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . 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 17.2: 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 g2.9) + 4096J4] + 0 ( l / g 3 ) .120) n4 .38). does not occur in any of the later contributions. i. The following Example 8. . The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. . J.e. for instance.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . Miiller [211].e.158 CHAPTER 8. (8. has been treated by R. (f>m in the above. Other applications can be found in the literature. . Imposing normalization later one obtains the normalization constant as an asymptotic expansion. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders.
b. . . 3\ 2 V ' 2. 1 1(1 + 1) 1 . (.1). S) is a confluent hypergeometric function. . Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ .2a2 A. 2 The solution XQ(S) = $ ( a . 1 . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2. a)ip(a) + (a. 11.V ! 4 I z S *! V2 ipq(z). .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. (8.z2 is a normalizable function if a = —n for n = 0 . Sees.2) where 1/ .123) The first approximation (note the last expression as convenient notation) ^ . multiply the equation by onehalf and set h = —a/g. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf.0(z) = zl+1ez2^fa.123) the contribution fl<°> Ah + . 2 .a + l)if>(a + 1) + (a.6 and 16.2 '(l + l) .. (fc2 + <?2) 4ga and b = I \—.122) where A is of order 1/g. Setting q = in + 3 implies k2 + g2 ga(2l + q). „2 J„2JL. Therefore in the general case we may set k2 + g2 = ga(2l + q) . \2 . We now insert this equation into Eq._ „!2 ^ ( " « 2 r r 2 > .8.121). (8. a .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . i/> with Dq = d2 . <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. The solution i.• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.b.l)ip(a . (8. . 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. b. .
a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . Now proceeding as explained in the text one obtains finally the expansion ip = ^(°) + v ( 1 ) +y>(2) H — along with the expansion from which A is determined.a—lli. The coefficients S m ( a .i ^(i)=£V+i i=0 J2 j = (i+2).b = (g + 3) .3). . a + r + l)(a + r + l . TimeIndependent Perturbation Theory (g . „ . (a. where [a. one could now use the RayleighSchrodinger method and rederive the result (8. In fact.120). . If we now evaluate the various coefficients in the last expansion we obtain the result (8. fa. Hence the next order contribution to tp^ becomes oo i+2 r [a a .^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously. a + r) with the boundary conditions Sn(a.a + j}i+1^(a + j).123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. _ J [a .l . a) = b .a — 211 . one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. a) = 1. a + 111 r [a.Z. a ] i + 0(/i3) [a. i. the latter determining to that order the eigenvalue.2a = I + q. This equation shows that a term fiip(a + j) on the right hand side of Eq. a + j)i/>(a + m). a + r — l)(a + r — l . a + i)(a + r.j^0 ' + jU+1 J ^(a + j) with h[a. of course.a]1=0. a + i) = 0 for i ^ 0. when j = 0. 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 + r) satisfy the following recurrence relation: Smi(a. So (a. a]i + h fa. 0 = h[a. (8. fa.2. a ] x + L _ 2 J 1 [ « .S m (a. a ] i + L ' ^ J 1 [ a + l .a]i = .1) = a . Now we make the following important observation that Dqip(a + j) = jip{a + j).120). ( 5 m ( a .160 where (a. a + r) = 0 for r > m.e. (a. As a check on the usefulness of the perturbation method employed here.a]2+L ^ J1[a + 2 .a + 2]i P . a + 1) = a = In general CHAPTER 8. a + r) + S m _ i ( a . o + r) + S m _ i ( a .a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". a .
proceeds along similar lines. light. as we shall see in Chapter 27.e. i. and this means electrodynamics.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. 161 .Chapter 9 The Density Matrix and Polarization Phenomena 9. We therefore begin with the appropriate classical considerations with a view to generalization. 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.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. and we shall then see that these lead to analogous results as in Chapter 2. Regrettably this has to be left out here in view of lack of space. 9. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. 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. the generalization to many particles. although not exactly straightforward.
3) (9. In general the constant Ao is complex. 9.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. t) = A 0J R cos(cji) + A 0 / sin(u. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.162 CHAPTER 9. A 0 = A0R + iA0I. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0. (9. This is the equation of an ellipse as indicated in Fig.4) (9.1 The vector potential ellipse. (9.i). t) given by A(r. It follows therefore that A(r.1. (9. i. The Density Matrix and Polarization Phenomena potential A(r. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.t) = 3ftA.e.2) (9. 9. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.1) Fig.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0.6) .
y)) components A i . Then their connection is given by Eq. which leads to an outgoing wave which we give the label "1". (9. the connection must be linear. the direction of polarization.A«). AW = £ fc=l. p.AW). See for instance A.t in which k and ui remain unchanged. We set therefore A 0 = Ai + A 2 . i.A2 is the sum of the moduli. 17. Observable quantities can only be those which are invariant under translations and rotations. In classical electrodynamics. (9. ie.A« we can construct only the following scalar products with this invariance: (A(°\A(°)).2) the vector potential A is completely determined by its two (say (x.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.e. for instance.k (910) We give the initial wave the label "0" and subject this to an experiment. determined with the help of a polarizer.2 *i*4 0) (98) with coefficients F^. A2. (9.11) 'This means we consider the polarization. (AW. with k .A 2 . the intensity of a wave.9. (AW.e.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. a calcite crystal or a film of nitrocellulose. 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. Rae [234]. represented by a filter "F".A(°)).g.e. as e. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. (9. .8). (A(°). i. From the vectors A(°).7) We consider now measurements with polarized light. however. we recall. i. the fields E and B are observables.A i = 0 = k . The intensity 7 of a light wave with amplitude coefficients A\' . Here we are interested in observable properties of light waves — like. not. the vector potential A. I=<l2x2>:=E40)*l*40)i.
To this end we observe that as a consequence of Eq.164 CHAPTER 9. Then according to Eq. (9. (9. (F)=Tr(Fp).14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. if (F) is an observable quantity. i.16) i. (9. i.12) i.15) in order to rewrite Eq. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.A*kF£A* = E and hence F = Fl A p i tkAk.e. the connection between the initial polarization and the final polarization. (9.13): j2^FikAk = Y^AiF*kAi = Y.A( 1 )) = ^ ^ F i. so that (F) can be looked at as the expectation value of an observable F.k i (9. pki = AkA*.e. i. (A(°). We use definition (9.A*FikAk.13) the quantity F must be an hermitian matrix and thus representative of an observable. Thus (F) = J2FikPki = ^(Fp)a.k l f c 4 ° \ (9.e.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). (9. We write the observable quantity as (F):=Y.17) . Thus. it must be real and hence (F) = <F>*.k (9. (9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment.13).13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.
21a) follows from Eq. i. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip.e. In the present case of a polarized beam of light we have two possible. . <»^> (9. "incoherently. i. For a 45° polarized wave we have 1 . b = 1 we obtain the expressions (9. we have (9 19) ' In this case there is no preferential direction. 165 (9. we can describe the wave function of the state polarized in the direction of x or y respectively by \<P)I =f0 J and \fh = ( i ) A special state is represented by j)=(J)+<!) with a 2 + 6 2 = l.e.* For a lightwave travelling in the direction of z. like that of a bulb.15) with A\ = a. (9.( J o ) .d "»=(o ! ) " In the case of an admixture with equal portions of 50%.21b) This state is still a pure state.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. is in general unpolarized. The socalled pure case.18) The vector A describes a particular ray of light. i. (9.22) For a — 1.(AAt) = ( £ £ ) . But light. b = 0 and o = 0. A2 = b as </*) . that of a pure state. is given by a single polarized wave.9.e. mutually orthogonal polarization directions x and y described by the matrices (see also below) * . This admixture is described by introducing matrices p.19). The density matrix for the pure state (9.
b = —=.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. l l\ " i °\ ) ' ^925^ 9.23) 1 . We set in analogy to Eq. Equation (9.g.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. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus. (9.15) Pik — A i A k . (9.166 and CHAPTER 9.8) A(1)=E^40)> k R = R(<*). Then according to Eq. (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".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. Pi35° = \ \ 2 ? )• (9.™ V 0 n kmAm / ^^jPim^mk' .Z^ 3. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics. in which the intensity I = Tr(p) is not conserved (e. as a consequence of absorption). 1 = . (9. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M .
q. (9. and p(0' remains unchanged.31). on the other hand.e.28) (9.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. (9. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. in the state functional) in analogy to the time dependence of W(p. the dependence on a is contained in F' = R^(a)FR{a). the observable is transformed. In the case of Eq. in other words without creation or annihilation of systems. (9.34) .e. (9.9. here this contains the dependence on a. i. that the initial intensity Jo is equal to the final state intensity I\. the observable F remaining unchanged. that Io = Tr( / 0 (°))=Tr(pW) = / 1 . In the case of Eq.30) Ti{FRp{®I$) = Tr(F.31) (9. i. The corresponding condition here is that no absorption of the wave takes place. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).29) where (i.33) i.4 The Liouville Equation 167 pW = RpMRl (9.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9.e. (9.pW).4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.e.t) in Chapter 2 in what we described as the Schrodinger picture there.32) (9. 9. (9. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.
i. (9.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.F}. (9.41) We see therefore that the 2 x 2 matrices entering the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics in Chapter 2. (9. the matrix M is antiHermitian: Mf = . (9. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. CHAPTER 9.36) (9. . Comparison of these equations with Eqs. (2.39) and (9. Eq.35) Thus the matrix R has to be unitary.46) suggests to define Poisson brackets of matrices by the correspondence {A. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.M .2) has to be taken into account. In view of Eq.39) If the dependence on a is not contained in p^1' but in F' (cf.e.40) Equations (9.e.37) (9.21) and (2. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.B}:=i[A. (9.40) can be considered as quasiequations of motion.p(°\ao)]da.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H. = Tr(R^Rp^).B].35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da. (9.p]. j£ = i[H. We set therefore M^:=iH. (9.168 i. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.
(10.2 States and Observables Every state of a physical system is represented by a certain ketvector \u) which is an element of the corresponding Hilbert space "K. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u). we establish their Heisenberg equations of motion.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. 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. To insure that the 169 . 10. and consider timedependent perturbation theory. the Schrodinger.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Thus in particular we establish the uncertainty relation for observables in general. Heisenberg and interaction pictures.
Pj] = 0.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. and assume that A = A(p.3 represent the three Cartesian coordinate directions of x.B] = ih. Such a correspondence with classical mechanics is not always possible. classically qiPi = PiQi. however. y. we must have (u\u) = 1. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. 10. E. Naturally one can also write (10.Qj] = 0.2. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. such as for instance of the angular momentum operators Li.g. (10. Here. The phase space coordinates qi.1 U n c e r t a i n t y relation for observables A.* a conspicuous exception being for instance a spin system. we assume this now. 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.Pj] = ihSlj. (10. which clearly does not hold in the case of operators.Pi are called fundamental observables.2.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. \pi.170 CHAPTER 10. Pi • Pi. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.4) *Even in those cases where this is possible the operator correspondence may not be unique. where i = 1. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1.y. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. q). .2) (F(A)) = {u\u) <"lf.q). [qi.
(A{A))(B{B)){B{B))(A(A)) (10. B}= ih. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). Thus AA = AA = ( i 2 ) V 2 .10) .2A(A) + (A)2. AB defined by the mean square deviation AC=((C2)(C)2)1/2. (4. 111 (10.(A)2)1'2 (1 °= a) ((A2 + (A)2 .(A)2 = (AA)2.7b).. Then [A.. (10.9) ((A2) + (A)2 .5)) (for A\u). (10.2(A)2 .8) and.B\u) G IK as vectors) we obtain (for an application see Example 10. (3.B.0)1/2 = AA. C = A.B] = ABBA > ^h.12) the Schwarz inequality (cf. = \\A\u)\\2\\B\u)\\2 (10.(A) = 0..11) (A2) = (A2) .2 States and Observables then their uncertainties AA. we have A2 = A2 . Since A = A— (A).{A)2)1'2 {{A2) . using Eq. Applying to the expression (10. Ai = = = ((A2) . Eqs..1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). so that (A) = (A) .12) where \u) 6 "K is a ketvector representing the state of the system. (10. (10.2A(A)}  {A)2)1'2 (10. (10. (10.10.13) AB = AB = (B2)1/2.6) B:=B(B). (10.7b) = = [A.9a).7a) (B) = 0.
172 CHAPTER 10. From this we obtain (uiJBu) 2 = i22 + / 2 . and (b) when (AB + BA) = 0. Zi (10. (10. for a complex constant c. (10. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. 4. (10. (10. AAAB>I=h.18) >R2 + I2. (10.4). The condition (a) is satisfied when. We can therefore rewrite Eq.13) (AA)2{AB)2 i.17) as had to be shown. and with Eq.16) I = h.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). (10.3.15) as (u\AB\u) = R + iI. since A and B are hermitian. R= (AB + BA).e.14) Hence (u\AB\u) = /^(AB where. (AB + BA) This means (AB + BA) is real. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. and hence for condition (b) 0 = = tSee Sec. Zi + BA)\ + ^ih. A\u)=cB\u). . The equaltosign applies in the case when (a) it applies in the Schwarz inequality. Zi Zi (10. (10.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*.
implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. = 5R2 + G 2 > 9 2 = hi2(Lk)2. the following uncertainty h2{Lk)2. Example 10.e.i relation holds: (ALi) 2 (AL. ALX 10. i. = 0. i. {u\LiLj\u) = R + iQ. Separating the product into symmetric and antisymmetric parts. . For states \u).20) or. such that Lz is "sharp". i. For these we have in the onedimensional case the relation [p. for which one component of L is "sharp"? Solution: Applying Eq. (Lz)2^0.z.y. (ALy)2{ALz)2 > ~h2(Lx)2.10. (Ly)=0. ALy = (u\L±\u) ^ 0.13) to operators Li — Li.7a). (A{A))\u)=i(Zc)(B(B))\u). implying {Lx)=0.) 2 > = x. (10. What does the relation imply for states u) = \lm).e. Ly are not "sharp". + LjLi\u).e. (Ly)2=0. in view of Eq. (Lx±iLy) ^ 0.e.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. The observables like angular momentum are functions of p and q. m\lm)). q) = ih. h2(Lz)2. i. (10.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. Thus in this case Lx. (ALZ)2(ALX)2 > \h2 = (Ly)2.Lj] = (LiLj + LjLi) + ih£ijkLk.3 OneDimensional Systems 173 Thus c = — c*. 9 = h{u\eiikLk\u). we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. 9ftc = 0. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2.
174 CHAPTER 10. as remarked earlier.23) = ihg(q2) (10. (10. This is why the operator p requires a representation in the space of eigenvectors of q.e.p] = ^ihp ~ Proceeding in this way we obtain [q.A(q.p) can be expanded as a power series in q and p. (b) its spectrum is necessarily continuous.34d)) [A. i. We have (cf.A(q.p2] = [q. .pn]=ih^(pn).21a) and (10.22) d ih—(p2).p)]=ih^^and \p. Then (see below) we can derive for A the following commutator equations [q. A — A(p.i h ^ ^ .p)] = .C\. Similarly we obtain \p. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. and (c) the spectrum is not degenerate. Q2] = [p.21a) Equations (10. i.q). they do not have a common system of eigenvectors.qn] = ih^(qn).21b) (10.e. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).BC] = [A.p}p + p[q.21b) now follow with the assumption that A(q. We obtain these equations as follows.B]C + B[A.p] = ih we obtain therefore [q. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). (10. Using [q. Next we have a closer look at the operator q. (3. q]q + # > v] = ~2ihq and \p. Eq.
e.27) (10.1 The delta function for K — oo. U] = ih^= aU. Replacing in Eq.21a) A by U.1 how the delta function arises in the limit K —> oo. (10. (10.3 OneDimensional Systems 6K(x) 175 Fig. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. . We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization).e. 10. 10.28) i.10. we obtain [q. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. and a a cnumber.24) lim SK(x — x') — S(x — x').f^ ^ J—oo dke ikx —ikx' (10.25) It is shown in Fig. i. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. = eipa/h = Uia)'1 = U{a). > We begin with (a). U is unitary. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.26) it Here p is an observable. an operator.
or qU(a)\x') = (x' + a)U(a)\x').31) . q any other observable F(p.e. q) can be constructed representing a dynamical variable. cf.q)\i>) = (i>\U^UA(p. CHAPTER 10.00). Since this holds for any arbitrary value of a in (—00.30) (10.q)U^U\ilj).3. Thus all these values belong to the spectrum of the operator q. or qU\x) = U{q + a)\x) = (x + a)U\x). The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').e. that with U(a) = e~ipa/h = U. Evidently every eigenvalue has only one eigenvector. (10. and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p. i. 10.1 T h e translation operator U(a) We saw above.00). Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). (10. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. we obtain qU\x) = U(q + a)\x) = {x + a)U\x).29) This means: U\x) is eigenvector of q with eigenvalue x + a. WU = 1. (10. we can see that also p possesses only a continuous spectrum (from —00 to 00). Eq. the spectrum is not degenerate.29). which means that the spectrum of this operator is a continuum. 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. ^Observe that {iji\A{p.176 i. b a cnumber).
i. (10.a).x')5(x" = (x" + a\c*c\x' + a) x').x") . \c{a. U(a)\x') = c\x' + a).32) Since U is unitary. (10.x') = = i.e) (x \p\x ) = . since q\x') — x'\x'). (10. we have q\x' + a) = (x' + a)\x' + a). for instance. i. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).33) (10. \x') = U(x')\0)./ (10. (10.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . (10. the operator U(a) acts to shift the value x' by the amount a.10. (10. i.e) ~ 5{x' .37) With the help of the latter expression we can calculate. The operator is therefore called translation operator. We choose the phase such that U(a)\0) = \a).3 OneDimensional Systems 177 Moreover.x")c(a.x" .e. (x'\p\x").x" .39) .e. (10.x" . It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .36) i.35) (10. offdiagonal elelemts of the {x}representation of the momentum operator.e.38) h 6(x' . it follows that 5{x" .hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x").31) and (10.5(x> .x") U{x'\p\x" % pz.34) {x"\U\a)U{a)\x') c*(a.32) we see that \x' + a)ocU(a)\x').e. In the expression of Eq.x')\ = l.e. Comparing Eqs.
e. i.to)m0)).f dpipeipx. The operator U(t.t0) = HU(t.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .45) (10.to) exists which is such that^ m)) = u(t.t0). (io.40) 10. In this case the classical Hamilton function does not depend explicitly on t. so that U{t. .tQ) = eiH(tt<Mh. (10.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. (10. For the way the state of the system develops in the course of time we make the following postulate: Postulate: A linear operator U(t. H\uE(to))=E\uE(to)).4. See Eqs. 6'{x) = !. it follows that 5'(x) is an odd function of x. Quantum Theory: The General Formalism We mention in passing that. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.178 CHAPTER 10. 5. a conservative system.4i) We assume first of all. . (5.° ° d(p) (ip) e"« = S'(x).e.44) (10. since 5(x) = . Differentiating U(t.to) is called the time development operator. 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)).27) and (5.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i.x") = (x'\p\x")*. i. to) with respect to t we obtain ihjtU(t.L fdpeipx. This is in conformity with our earlier postulate in Sec.e.35). isolated from any measuring apparatus which would disturb the system and hence its behaviour in the course of time) and let \tp(t)) be its state vector at a later time t > to (with no interference in between).43) (10.
t0) [U(t + e.47) e. (10.10. t0) (10.t) = 1. t0) = H(t)U{t. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.t0)]1 Let us set for small values of e: U(t + e.t) = l^eH{t).z / dt'H(t')U(t'.48)).t) and therefore [U(t. (10.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. i.45) has very general validity. Ufa.41) we obtain ^(t 2 )> = = J7(t2. Evidently U(t. From Eq. Then with U(t + e. The operator H is d efined by this relation. (10.t0) or ih—U{t. (10. Hence U(t.t0)U(t0.to) = 1.to) can be expressed as an integral as the solution of Eq.t)l]U(t.t) = l. = U(t0.t). t0) = E/(t 2 .t0)\^(t0)).4 Equations of Motion 179 We now demonstrate that Eq.to) (differentiating we obtain immediately Eq.tiM*i.*l)^(*l)> = I7(*2.to).49) .48) with U(to.*l)^(*l. But obviously (cf.t)U(t. (10. (10.t 0 )^(*0)> U(t2.44)) we have U(t.e. (10.48): [/(Mo) = 1 . Eq.
51) Since </.e. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.(£ + dt)) = U(t + dt. The eigenvectors of these observables are also constant in time. Eq. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.53) . t)\t/>(t)) =U~ and since H is hermitian. The operator U(t.180 CHAPTER 10. i.5o) ihjtm)) = Hm)). the predictions of different descriptions are identical. 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. the instant at which a measurement is performed on the system. the Schrodinger equation ju(t. (5.t0) m0)). 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. Now let \tp(t)) be the state of the sytem at time t.e. it follows that U(t + dt. requires the identification H(t) = H= Hamilton operator. Another such description is the Heisenberg picture. Comparison with the equation postulated earlier. and these remain unchanged under unitary transformations.41). Quantum Theory: The General Formalism We obtain the differential equation for the development of states of the system in the course of time by differentiation of Eq. >(*)> = i.t) = l % ^Hdt) \il>(t)) Hdt (10. (10. Then \Mt)) = u(t.30). to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. (io. Since the only measurable quantities are moduli of scalar products.t0)\ii>s{to)) (10.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). (10.
i.55) and using Eq. . subject to a continuous change.e. As(q.59) This equation is called Heisenberg equation of motion.p) in the Schrodinger picture dAs/dt = 0. t0).48). but here.t0)\ips(t)) = \ips(to)) = constant in time. However the associated observables are timedependent.56) U^[As.e.e.i). we consider only cases without explicit time dependence of As.HH]+ihU^U. t0)AsU(t. for instance in cases of nonequilibrium or if part of a system disappears through absorption. even if As is not timedependent.55) Thus AH is explicitly timedependent. For As(q. as a rule. 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.4 Equations of Motion and \^H) = U\t. (10.57) and rf[As.HH]. if ^ = 0. in order to obtain the Heisenberg picture (description).(tf HU)(rfASU) (10. (10. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.H]U = = = U]ASHU AHHH — U^HASU .With explicit time dependence. Here H is the Hamilton operator in the Schrodinger picture. Differentiating Eq. so that a matrix element (i. we have to take into account the contribution dAs/dt. Thus. a scalar product) remains unchanged: AH(t) = U\t. 181 (10.H]U + iW]^U.10. (10. In the Heisenberg picture it is HH = U]HU. = so that ihjtAH or ihjtAH = = [AH.p. {AH. i.HH]. (10.58) HHAH (U*ASU)(U^HU) [AH. (10.54) Observables transform correspondingly with a similarity transformation. It does not always have to be the case that As does not contain an explicit time dependence.
Example 10. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture.21b). This holds also in the case of the Schrodinger picture. For an extensive discussion see B. in the case of the fundamental dynamical variables qi. In our treatment of representations. In particular.H] = 0. (10. Orfanopoulos [224]. since in the transition from one to the other the commutator remains unchanged. A.HH}=0. i. We see that formally one obtains Hamilton's equations of classical mechanics.Pi): and where the expressions on the right follow from Eqs. (10. An observable CH is called a conserved quantity.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. [CH. Finally we add a comment concerning conserved quantities. we used in essence unitary transformations of matrices.Pi we have in the Heisenberg picture (replacing in the above AH by qi. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. provided the Poisson brackets in the latter are replaced by commutators. if ffCH = 0.21a) and (10. 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.182 CHAPTER 10.e. Solution: We have for an observable A (A) = (i>H\AH\TpH). The "pictures" just explained — with regard to the time development of a quantum system — are not to be confused with the representations of the theory. . In the treatment of the pictures we used unitary transformations of operators (and vectors).13) to derive the uncertainty relation for energy and time. We end with a word of caution.61) Thus conserved quantities commute with the Hamilton operator. In many cases the Schrodinger picture is more amenable for explicit calculations. We also observe that the momentum pi is conserved when \pi.
\HAM\2) \(nA. a position or momentum observation is independent of t. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. i. (10. i. > (10. H] = 0.H]) = 0. (10. is displaced by A A in the interval A T = TA. We construct the following vectors using Eq.p) we then have 0=^M) = U[A.13) and (10.64) = 0.e. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .66) This means.62) The Hamiltonian H does not depend on time.7a). (10. at ^±AH>h.62) and dAs/dt > \WAH\i.4 Equations of Motion Then.10. (A). since \IPH) is timeindependent.e.H}\f)\2.e. we can replace ([AH. (10. and TA O C (10. the "centre of mass" of the statistical distribution of A. of A\<t>) = a<£)) are then called "good quantum numbers". AW = {A(A»V>. . this is practically the same as at time t for times t with \t — t'\ < T.)\2 ( = \{i.63) so that with Eqs. If r is the smallest such characteristic interval of time. H\tl>) = {H(H))\f).2 > h.68) To put it The eigenvalues a of A (i. i.Such a characteristic time interval can be defined for any dynamical variable.e. dt in.15): (AAf(AH)2 Using Eq. (10. (10. then if a measurement is made at time t'.67) 1 d(A)/dt is oo and AE = 0. 2 d{A)/dt " and. 183 dV ' \ dt >"•"»» + ( £ ) • (10. if we set AH = AE we obtain TAAE (10.65) and rA AA d{A)/dt' . Let \ip) = \"4>H) represent the state of the system. in a different form: The eigenvalues of an observable are called "good" if [A.HH]) Hence we obtain by ih(A). This means that for a physical variable A(q.
69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.e. i ) = J dE'^El^)eiEltlhfEo{E'). /oo (10. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. First we obtain from Eq. (10. Quantum Theory: The General Formalism 10.tyEt/hdt = = = / J—oo dt J dEJrl>E. states with sharp energy (AE = 0). States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".(x)fEo(E')e«EE'Wh 2irh f dE'^E.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 .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.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x . the wave function ipEo(x. (x)/ E o (E')5(E . (10.70) by From Eqs. (10. have around EQ the form ipEofct) = i..70) we can determine the function fEo{E) integration with respect to t.70) we obtain: oo / oo TPEO (x.H ) ( £ .72) = V'EoW 7/2h + 0i(EE )/h  oo . t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.t) — ipE(x.184 CHAPTER 10. i. (10. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. (10.)exp(iEt/h)). The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. This means that in the case of decaying particles.Eo(x)eiEot/he'^2h9(t). for instance. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.71) From Eq.69) and (10. as is observed. the particle density diminishes in the course of time (t > 0). then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. the wave function ^ ( x . Recall for instance radioactive decay.69): oo poo p / oo *PEo(x. This means.E1) 2irhfEo{E)4>E(x).
6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. (10.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). that** EQ. (C/(0) unitary). and is called "interaction picture" or "Dime picture". Considered as a function of E.105) and (20. We saw previously that if the state of a system is known at time to. (10. (10.10.g. This description contains effectively parts of the other two pictures and is particularly useful when Eq.e.e. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula.e.41). 10. (10.toM{to)).e.45). of course. Let U^°\t. 185 i.t0). W)) = U(t. Eqs. where U is the solution of Eq. i.11). (10. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. i.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). i.45) can be solved only approximately (in actual fact.e.75) (10. to) be an approximate but unitary solution of this equation. They are called "resonance states". . ihU(t. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. The state with lifetime r — h/'y is not a discrete state.76) **For specific applications see e.74) Thus the solution of this equation is the main problem. 7 > 0. £/ = £/W(Mo)^'(Mo). every unitary transformation of states and operators defines a possible "picture"). i. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. (14. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j.to) = HU(t. ^(*o))> then \xp(t)) for t > t0 follows from Eq. We obtained the result (10.
i.e. (10.e.78) (10. (10.e.t0) 0 = l. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. [/' has only a weak ^dependence (i. Now let H^(t).e.81) But [/(°) is unitary.186 Then CHAPTER 10. with Eq. £/(°)£/(°)t = 1.82) (10.e. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). (10. In addition we have (10.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i.83) ih^U^ ishermitian.79) ^'~°i. = HU<®U'. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0.e. Quantum Theory: The General Formalism ih^(uWu') i.e. H(°Ht) = ih^uW. i. varies only slowly with t).h^U> = at ( io. i. with Hamiltonian H = H^ + H'. .
84) U^ where H' is also hermitian.78) at (10 85) (10. U^H'U^U' (10. (10.10.89) where ^s(i)) and As are state and observable in the Schrodinger picture.86) at = U^HU^U'. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.U^HU^U' H'fi'. i. H = H{®{t) + H'. (10. (10.85) = where we used Eq.92) .85) multiplied by U') on the right of Eq.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). (10. (10. U<®\Mt)) = \Mt)).90) = U^H'\^s(t)). (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^. lfe(*)}.78). (10. (10.80). (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. and obtain from Eq.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. We also set H^U^H'UW.e. (10.
e. In order to obtain the equation of motion of an interaction picture observable Ai. ot i.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. near zero and \tpi(t)) is almost constant in time. (10.e. we differentiate with respect to time the operator defined by Eq. (10. dAs/dt=0) .95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. (10. comments after Eq. i. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)).94) Then we have (normally with dAs/dt dt (io=8o) — 0. (10.89). the vector varies with time (different from the Heisenberg picture).89)^0. Since H'j is assumed to be small (i.e. cf. (10. so that with Eq.188 CHAPTER 10. we have i. H'j = UWH'UW. i. (with ihjtAI(t) = [AI.60b) with iff0) instead of H. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq.H\0)}.e. i. (10. A^t) = U^AsUi0).e. a perturbation so that H ~ H(°)). (10.94) _Hf)Al 1 + lhdAl+AlHf).e.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .55)) at + imm<Ms_u(o) at dt _umH(.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).90) and replacing the left hand side of this equation by the right hand side of Eq. ^ (0) l^(*)>.60a) or (10.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. (10.
i.80) as E/(°)(t. (10. (10. We assume again that the spectrum and the eigenvectors of Ho are known. i. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')). ^ \<Pn){Vn\ = 1n (10. We have the Hamiltonian H = H0 + H'. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. H' = H'(t). i.e.97) The equation to solve is Eq.e.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. H0\ipn) = En\(fn).t0)H'uW(t. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em.10.7 TimeDependent Perturbation Theory 189 10. ih±\Mt))=Hi{t)\Mt)).93). Since .100) (10.87) and (10. where according to (10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture.e. (10. (10.98).t0). {ipn\<Pm) = Snm.99) (1098) Instead of the original Schrodinger equation we now solve Eq. (10. We obtain the operator U^ from Eq.i0) = e<Ho(tto)/ft.102) We assume that before the perturbation H' is switched on at time to.
104 ) Here we insert Eq. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^).107) Jto With this we obtain from Eq.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. Quantum Theory: The General Formalism HQ is independent of time.106) We insert this expression into 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'.103) for m — n and use Eq. i. (10. the time dependence of \ips } can be separated as for stationary states.102) and truncate the series after the first perturbation contribution.^ V ™ ) .103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„. (10. we can write l 4 0 ) ) m = elHot/hWm) =e .190 CHAPTER 10.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 . (10. We 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. Hence we set ^> = £°«(*)l4 0 ) >n. (10. (10.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).e.108) .
t o = 0. (10.112) for a^0. (10.109) 10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. (10. » — < (10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0.110) With this we obtain from Eq. if in the scattering of a particle off some target its momentum p changes to p'. where the momenta of the aparticles form a continuum. but that otherwise the perturbation is timeindependent. A further example is provided by adecay. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.Em)t} \{tfn\H'\ipv h 2h(En .Em)t' (<Pn\H'(t')\pm) = Wnm(t). We assume that the perturbation is switched on at some time to.10. (10. Thus we set H'(t) = H'0(t).En (10. for instance.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .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^. This is the case.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En .109) the transition probability Wmn(t) = jp Jo . < KO?t . or if by emission of a photon with continuously variable momentum an atom passes from one state into another.111) Different from Eq.
and no other state would be nearby. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig.io. 10. . we would have En — Em ^ 0. In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. 10. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.Em)t} r i 7 TTT. Otherwise. though sufficiently small.— = o (En 2h En 2h5{En .114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .192 CHAPTER 10.Em).2 The delta function for £ — oo. Hence lim St(a) = 5(a).ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. Fig.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . (10.2. > 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. so t h a t Wnm would be zero. if the energies belonged to the discrete spectrum.E7m)t} = (10.
75.W . 266.e. so that initial and final states belong to the continuum. (10. Example 10. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.3: Application of test functions to formula (10. In the case of the functional (10. sin 2 (tx) ax i—( * According to F.117) provides the outgoing particle current (see Example 10. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t.r Z x 1 ^ ) = 4>{Q). .117) were derived by Pauli in 1928. If we are concerned with pure scattering. (10. then Eq./ t>oc TZ J_00 dx . Then. Let p(En)dEn be the number of states in the interval dEn.116) The transition rate V is the transition probability per unit time. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . pp. in the article of W.113). (10. for instance from a potential. Pauli [227].115) and (10.113) Use the method of test functions to verify in the sense of distribution theory the result (10. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. Fermi gave this formula the name "golden rule" . Fermi [92]. _ 2irh AE > > Se.118) where 8e is the separation of states around En which in the case of the continuum goes to zero.113) this is* hm . We evaluate the integral 0(0) = 1.8 Transitions into the Continuum 193 as equal. 142. Schwabl [246]. p. (10.4). if p(En) varies only weakly with En. i.117) in many applications. F Jo . f See E.10.* In view of the usefulness of Eq.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation .
e " 1 * 1 = 2 lim — . Then ^ . Hint: Use for ip„(x) a plane wave ansatz with box normalization (volume V = L 3 ) and obtain da from the ratio of outgoing and incoming particle fluxes. E x a m p l e 10. .? 2TT / L \3k2dkdUn 1 /V(x). The ingoing particle flux is therefore v / L 3 .a 2 ' i that for a = b = t .^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . 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. Ryzhik.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. p = 1 we obtain I = .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . .194 CHAPTER 10. with v the velocity of the particles. 491. and hence .. . 1 . .947. S.ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. x . = 0 . iknx .. Quantum Theory: The General Formalism with the help of Tables of Integrals.<p). L \3k2dkdUn It follows that T./ cfc a n p 2 +fc 2 . 2 T / ^ M. * ^ = . . = ( — ) dk= 2ir (—  2irJ k2dkdfl. .. formula 3. Gradshteyn and I.tan p 2 + a 2 .k. dEk = h2kdk//j. M.( ^ V V • < ih 2 _ i k n . and is obtained as ih.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. . Jin § m = See I.. p. . . . . 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). 2 .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). 77^ + . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . Hence the result is 0(0) as expected. Then p(Ek)dEk Since Ek — h2k2/2[i. V +(ab)2 a _x 2p6 7 m —^ '. § We have r Jo \p\x dx sin ax sin bx  V. K = k 0 .„I. E = h2k2/2fi. ^o n . with n .
10. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>). by which we mean without leaving the Schrodinger picture. . Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.x)^(E'.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn).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.122b) S e e also L. For a possible application of this result.121) . 1 rfx^°)*(x)^)(x) = 8mn. (10. 193.x) = S(EE').t)+ / n a{E./continuum = <San(£)</4°)(x. Thus we set ^(x. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. specifically in the case of the Coulomb potential.69).^ We start from the Hamiltonian H = HQ + H'.10.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. pp. We expand the wave function V( x . 148.t) = ^On(t)^(x.t)^°\E^t)dE (10. (10. (10. Schiff [243]. see the discussion after Eq. we present another derivation using the usual method of timedependent perturbation theory. (21. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known.9 General TimeDependent Perturbation The differential cross section is defined as da •• Theory 195 outgoing particle current x r2dfl ingoing particle current da / /i V (hko)/(lJL3) J dxV(x)eil where for purely elastic scattering k = ko.t).^°>(E.120) In the case of Eq. (10.
127) i n c a g e Qf c o n t i n u o u g £.196 CHAPTER 10.E') = 5{E ..124) and obtain ihSdn(t)ipn^)eiEnt/h + San{t)En<pn(x)eiEnt'h = San(t)(HQ + H')vn{*)eiEntlh = San(t){En + H')vn(x)eiEnt/h. (10.. = e^EE'^h5(E Thus with Eq..126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10.t)dx. Observe that these conditions do not contradict Eqs. The second term on the left of this equation and the first term on the right cancel out...123) We now insert these expressions into Eq. into dip (10. i.125) where an(t) — dan(t)/dt.e.x. (10. . / *.t)rJ><®(E'. (10.126) + H')<pn^)eiE^h}.128) . Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x.127) we obtain from Eq.129) . Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10.x). since f rJ>(°>(E. (10.„. ^0)(E.iJ5 »*/Vn(x).„„.x. .E'). (10. (10.x.. V>(x. f 6kn in case of discrete tp's. (10. . t) = San(t)<pn(x)eiE^h.122a) and (10.t) = eiEtlhy(E.122b).i) = e. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). We multiply this equation from the left by </?£(x) and integrate over all space. . dnptMrnW = { ^ _ En) (10. Next we use the orthogonality relations .120).
(10. (10. to perturbation theory in the lowest order.e. they are fixed by the initial conditions. (10. determine the state of the system at time t — 0. let's say am / 0.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.131) and insert this into Eq. (10.130) along with the parameter A in front of H'kn. SH'kn(t)e^~E^ha^(t). We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.132) These equations can be integrated successively.e.130) H'kn by \H'kn and at the end we allow A to tend to 1. Thus first we have ak0) = const. Comparing coefficients of the same powers of A on both sides. (10.130) we develop a perturbation theory. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . (10.120). since it is equivalent to the latter. 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.133) The numbers ak = const. 0.10.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. i. and we set ak°}=5km or ak0) = 8(EkEm). Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. SH'kn(ty^E^l^\t) (10. i. In the first place we replace in Eq.134) .130) This result should be compared with the Schrodinger equation (10. We restrict ourselves here to the two lowest order equations. In order to solve Eq.
) ' X V(x')dx'. Equation (10.k .137) .e. (10. (10. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. H'^y^"^^/hd£. In this case we obtain from Eq. ip) in the context of timeindependent perturbation theory.ml2sin2{^frn (En . we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. (10.e. ifm{kl) = f J —oo (10.135) / Jo L Hfrfc ~ Em)/h Hence the probability to find the system at time t in the state n ^ mis (with Al) 22 i rjv I1 r i r IW a(l)m2 K ( j l = i^kl\t) = H'km e^E^dt> = H'km 6 \nnm\ ei(EnEm)t/h i(EnEm)t/h _ j [EnEmy 4g. (10. Integrating the second of the equations (10.Emy in agreement with Eq. (10. What is the physical significance of Eq.136) Comparing Eqs.e. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . so that ihakl\t = oo) = 0. (10.198 CHAPTER 10. The amplitude / B o r n is a Fourier transform in spatial coordinates. ipm) into the state k.135)? The quantity ak\t) is the amplitude which determines the probability that as a result of the perturbation ^H'km the system makes a transition from the initial state m (i.135) and (10.135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0.135) We put the additional constant of integration equal to zero. 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(9.132). whereas ak is a Fourier transform in time (a distinction which is easy to remember!). i.111).136) we see that both "amplitudes" are Fourier transforms and look similar. i. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.
such as the electromagnetic field (the quantization in these contexts is often called "second quantization".m. 11. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. 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.y. ( 1L1 ) We let pi. Z<i = Z) y(r) = _ M ^ . Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory.2. 199 (1L2) .Chapter 11 The Coulomb Interaction 11. meaning the inclusion of the creation and annihilation of field quanta). r=  r . The Hamilton operator is then given by w= jt + ji_^M. Let r : = (x.i be respectively the momentum. the position coordinate and the mass of particle i with i — 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.Ti.2 Separation of Variables. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.
4) P = p i + p 2 . Z) and r = (x. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . . 2Pip2 m\m2 pip2 mi + m 2 . P = pi + p 2 . (11. r} representation they have the differential operator forms Setting R = (X. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc.3) for the individual position coordinates we obtain ri=R+ •^ —r.200 CHAPTER 11.2/ mi+m2 The mass mo is usually described as the reduced mass. where m 0 = mim 2 . so that as expected according to Eq.U7&. (11. (11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. Solving Eqs.5) The momenta P . /'Pi p = m0 P2 A v. p are such that as operators in the position space representation or {R. y. (H4) \mi m. Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' . mi + m 2 iTi'2 r2 = R ^ —r. so that as expected we obtain the expression for p in Eq. Y. r = rir2 mi + m 2 (11.3) and centre of mass momentum P and the relative velocity p/mo by setting r» ._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. mi + m 2 va\ (11.4). But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2.
= ^. (i. Pj over to operators ?i. for which we postulate the commutator relations* [ri. r = —.10) One can convince oneself that the transformation from (ri.Pj} = ih6ij. we mean n = rx s x. with M := mi + vn^R = ^ P. etc.+ ^(r) 2(mi + m 2 ) 2m 0 (H.y. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. .pj.8) and so W =777 v + 7 T .9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. P on the one hand and those in r.Ri.Pj).+ P. p = W.16) Although we write ri. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.pj]=iMij. pj.e. (11.r2) to (R.Ri. (11. The canonical quantization must always be performed in Cartesian coordinates. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. p on the other are completely separated.pj.13) * and = M ' 1V1 P = 0. r) is a canonical transformation. (11.2 It follows that p2 2 P2 n2 n2 n2 (H.11.j = x. (11. Angular so that Momentum 201 ^ ^ . Hence Hamilton's equations apply and we obtain. 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).il) ""Op' i.2 Separation of Variables.12) (11.10) we set H = Hcm + H. 2(mi + m<i) 2mo 2m\ 2m.+ ^2. [Ri.z). Pj (which in the following we write again ri. Ri. (1114) m0 We observe that the equations of motion in R.15) In accordance with Eq. (11.
202 CHAPTER 11. i. V(r) = V(r). 0 < r < oo. (11. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).21b) + E.24) . df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11.Q 2M' + V(r). Thus the twoparticle problem is practically reduced to an effective oneparticle problem. where Hcm$(R) = and Hip(r) = with ^total — ER (11.e.22) If.r) = *(R)V(r). (11.17) (11.23) y — r sin 9 sin tp. 2mo (11. (11.21a) L*+v{r)] ip(r) = Ei/j(r) (11.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. (11. ip: x z = r sin 9 cos ip. r ) = J E t o t a i*(R.20) r2MAR_ $(R) = ER*(R). it is reasonable to go to spherical coordinates r. = rcos9.r).18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . as in the case of the Coulomb potential. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . 0 < 9 < ir. 9. The Coulomb Interaction nc H = 2m. —n < tp < ir.
e.3. k = 1.2 Separation of Variables._0di})\ x We now demonstrate that 1. n ( . (11. 2 1 9 / 1 9 .A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq.32) (11.31) . if i.3. i( i.pr]=ih.33) i.2. j . = 0. i—>0 (1L25) (11.2.12 — I2.26) However. k an anticlockwise permutation of 1.27) We can now write (see verification below) p V = h2Ar^ where 2.11.j. (11. = —1. _ h2 \ . (11.k a clockwise permutation of 1. We have I2 = (r x p) • (r x p).28) . (l + .24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.29) .3. = (p2 + ^ V (r + 0).2. (11. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«.5r = .1 it is shown that pr is hermitian provided lim n/>(r) = 0. (11. with tijk — + 1 .i ' . .j.)In Example 11. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. otherwise. where (r x p)j = ^eijkrjpk. is the operator of the relative angular momentum. that the above expression follows from l:=rxp=ift(rxV). i.
41) o = r pr.39) + ih).43) r2(p2p2). (11.40) (11. (11. (11.ih)pr and x 2 2 9 9 2 2 Q O (11.j. Eq.(r • p ) 2 + 3ih(r • p) .27) we obtain r{prr)pr = r(rpr .ih5jk)pk (11.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . * ' S Since r =xz we have (cf.35) j.44) . (11.ihr • p ) . The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.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.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p . (11. This means that l2 = so that 9 9 1 (11. (11.38) + y* + z . r— = r • — or or (11.ih) — rpr(rpr With Eq. ih6jkrjpk (11.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.42) (11.37) .15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.ihr • p r2p2(rp)2 < + ih(rp).k Using Eq.k ~ rjpkrkPj) (11.
<p)*rip(r} 0.Pril>) J / r= dQs rcj>(r. 11.<p))ril>(r. ip 6 X>Pr C 7i.e'iP^} .d_ r dr the operator pr is hermitian.<p) (11.e.45) oe Thus in Eq.2. Angular Momentum 205 Comparison of Eqs.l2} = 0. h f l d —(r<l>*(r.Q ' 2mor2 with the condition (11.<p G [0.11. <p)* . i. r—>0 il>(r. (pr4>. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.2TT].26). Then {<j>.d. <p)r2dfls = .e. 6. or J .. /'(r.e. 0.— (riP(r.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3). 9 € [0.<p)*?(ri. Pri/Oi V0. (11.30) I2 is to be identified with the square of the angular momentum operator.1 Separation of variables We deduce from the expressions of H and l 2 that [H. 6.<p)drdna " r r=0 J \ll§r~(r4'(r. (11.2 Separation of Variables.<p))drdns. (11. i.ip) = (<£. ¥ p)r 2 drdn s = (p r 0. pr = . S o l u t i o n : Set d£2s = sin.6.48) . <p))r2drdns % J r or r<t>(r. 9. reR3. <p)*Pri>(r.47) Example 11.e lim rip(r) = 0.. r := r : lim rip{r) = o i .7r]. prip) = = I <j>{r. <p) d °° .(r.e.28) and (11.6l.e.6d0dip. H 1 2mo pi + K) +V(r). 8.46) (11.<P) = Etl>(r.! /').30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11.[ i J With partial integration with respect to r this becomes (<l>.[ 4>(r.
lx or l . (11. we have [h. 3 .ly]=ihlz.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions.49) By cyclic permutation of x.l2} and correspondingly [lx. [lz. ly] = [yPz ~ zpy. i. for any two of the components there is no common complete system of eigenfunctions.51) (11. [ly.xpz] = [ypz. These operators here play a role analogous to that of the quasiparticle operators . y.206 CHAPTER 11.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .e. i. Example 11. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. ?3 = z).e. (11. 11.e. so that [lz.pz]x = ih(xpy . The Coulomb Interaction i. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.52) This means that l .lz]=ihlx.50) Thus the components of 1 do not commute pairwise.ypx) = ihlz. they are incompatible variables. z]px + py[z.3 Representation of Rotation Group Our objective now is to develop a representation of the rotation group in analogy to the energy representation of the simple harmonic oscillator. 2 2 2 = 0. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. Since 1 = r x p .lx] = ihly. (11. 2 . In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.q2 = V. (11. xpz] = y\pz. [lyA = 0. zpx .2: Verify that [h. H and l 2 possess a common system of basis eigenvectors. /z> *xJ = l <i\}ylx + tx'j/Jj [h.53) Either of these operators is the hermitian conjugate of the other. However.lz (or l . z we have altogether [lx. We therefore first search for a complete system of eigenfunctions of l 2 .l2] = 0. zpx] + [zpy. g i = x.
y . We write or define: / . l\l) is eigenvector of lz with eigenvalue (IQ — l)h. (11. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. [ig. (11. assuming a finite dimensional representation space.r) (11 = {l02)h{l)2\l) (11. (n. and with Eq.Z_] = [ l 2 . (11.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. since lz is hermitian.59) = 9) h(l0 r)\lr).51). (11.i] = m. i + ] = [l2.+ ll. (11.e.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. Therefore (lo + l)h cannot be an eigenvalue of lz.r) := (Z_)rZ) so that lz\l .).Y\l) = (lQr)h(lY\l).54) l 2 = \{l+l~ + 11+) + ll = l+l. (11. In the present context the operators are called shift operators for reasons which will be seen below. (11. (11. (11. (11.3 Representation of Rotation Group 207 A.54) On the other hand we obtain from Eqs.54): lzl\l) = {llz . Hence we must have Z+I0=0.58) {l0 : max. [i+M = 2%iz.52): 0 = [ l 2 .57) i.lh)\l) = {l0 .54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). (11. A^ in the case of the linear harmonic oscillator.60) Similarly for l+ with the help of Eq.h~lz(1156) (11. Similarly we have lg{l)2\l) and more generally lz(l.61) But lo is by definition the largest eigenvalue of lz.l)hl\l). With the help of Eq.i+] = m+. .11.49) we obtain the following relations: [iz.
60) and (11.\lr) terminates at some number (let us say) r = n (n a positive integer). (11.e.. l\l . l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 . (11.56) CHAPTER 11. \l).n)}h2\ln). i.Z_] = 0.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .+ 11+) + l2z\l) = i+i. (cf. (11. (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz. Eq.\ll). (11. (11.e...64) i.68) .e.n).y\i) l)h2\lr).67) i (11 60) = But according to Eqs.66) It then follows from the second of relations (11.63) + l ) ^ 2 .From Eq.n) (1 = 60) {l)n+1\l) = 0.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). (11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.h)\ln) {(lQnf(lQ.54) (n = 57) (11. [12. i. The Coulomb Interaction (l+l. IQ(IQ 1 0 (11.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.l2\l) = l0(l0 + l)h2l\l).208 Thus for l 2 we obtain: V\l) ' = 11. i.e.55)) we have 12Z_Z> = l. (11. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_..60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.
.. —loh. Messiah [195]. p.n)} = (l0 .m .70) they satisfy the eigenvalue equations (with m = l0Jo l.m) = mh\l.n){l0 .m) = IQ(IO + l)h2\l. i..e.l0): l2Z. R. The nonvanishing matrix elements of 1 = (l+. \l — n) are eigenvectors of lz with eigenvalues loh. *D. As in the case of the harmonic oscillator we introduce normalized basis vectors. lz\l.m\lz\l.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . y/(l0±m + l)(l0^m)h. (IQ — l)h.. m + l)h.1). (11.m y/lo(lo + 1) — (m — l)m\l.m)\l. 17.. (11.m) (l...lz transform the vectors \l). II.70) According to Eq. We therefore write the 2/o + 1 orthonormal vectors: \l.m + 1)(Z0 + m)\l... The basis vectors \l).m). In the following we consider primarily the case of integral values of IQ. (11. Brink and G..m±l\l±\l.71) can be established in analogy to the method used in the case of the harmonic oscillator.m). § See e.Z 0 ).m) : Mo)..m) with — lo<m<lo2 (11.m) l\l. lo = l0n.65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2.69) Prom this it follows that IQ must be either half integral or integral.1).  U o ..m) = = = = y/lo(lo + 1) — m(m +l)\l.63) and (11.3 Representation of Rotation Group so that with Eq. y (lo . Satchler [38].. where lz\l.2. The dimension of the representation space is therefore n + 1 = 21Q + 1.l)h. 24. The operators l+. The phases are chosen such that§ l+\l. p. we skip this here.m) — mh\l. \l — n) into one other...11.m / + l)h . lQ = ..m) = = mh.l)h \/(lo + m + l)(l0 ..{lo . M. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator. We write these* for integral values of IQ: \l. (11.g. . Vol.l_. (11. The normalization factor in Eq. According to Eqs.n .m).l. lz) are then: (l. A.n) 2 .
—lo + 1 .l0)=0. In these coordinates we obtain: h l± so that h2 = xpv . $ oc eim*..<p) = i)n2Y^{e^).yp. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip).73) 1 d d sm0 06 \Sm9dd) + (11. The Coulomb Interaction Z+Mo> = 0.<p). . mhY£(9. 11. ip) = e(9)$((p). CHAPTER 11. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ .With the ansatz Y(9.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation..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.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. 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. . We .. as this is sometimes described. as above. ip.<p) = i0{io + lzY£(0. with i2Y£{e.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9). (11.76) d2 (11.e. where IQ = 0. and m = —IQ. i. as remarked earlier. for the simple "rotor". .2. . (11.1..77) $((p) = m2<b(y).210 Then. We now perform the analogous procedure for the angular momentum or. /o. (11. 9. d2 sin2 9 dp2 _' (11.75) the variables contained in the expression (11. l_\l.
ypx = tfrK(11. is given by (2Z + l ) ( Z .r o ) ! YT{e.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.<p) Jo = 6mm'5w. This is the case if one demands that l 2 and lz possess a common complete.0) is real and positive.82) .^ = *(n . These functions are defined in such a way that they are normalized to unity on the unit sphere.i». (11.76) and (11. / 3 / 5 J — casO. orthonormalized system of eigenfunctions. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.n>). The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.t»'\Y7n((l'.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.80) y? Furthermore Y°l = V^fcosO).(p) are called spherical harmonics. We mention in passing that in spherical coordinates d lz = xpy . so that im(ip±2ir) The functions Ylm(8.81) V l07T V47T y? = Y? = One should note that the functions Y™ are square integrable only for the given integral values of m and I.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. (11.* ? ^ . (11.V>)Yir'(8.\ = W . and that their phases are such that Yt0(0. (11. Uniqueness implies immediately that m has to be integral. (11.(5cos 03cos0).=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.11. m > 0.79) .
85) (11. (11..2.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .88) y = ^aiXi+K.88) we obtain for every value of x ^2 ai(K + 0( K + * .p. the designation s. (11.1.x2)Pf .86) y = 0. i=0 (re lowest power > 0).e. (11.e.e.l)xK+i~2 i . Setting y = (lx2)m>2Pr(x) m ..x2)y" .d.x2)y" .90) .e. (11. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.2xy' + 1(1 + \)y = 0.. Instead of the values I = 0. oo (11.212 CHAPTER 11. we would now have to search for those values of I for which the solutions are square integrable.87) we obtain the equation (1 . 22 (11.1(1 + l)]xK+i = 0. For the solution we use the ansatz of a power series. exp(imip). i. i. If we did not know yet that I = IQ = an integer > 0..2(m + l ) x i f + [1(1 + 1) . (11.m(m + l ) ] i f = 0. in the replacement ip —> if + 2K..g.x2)P[' .2xP[ + 1(1 + l)Pi = 0. f. Consider the equation for Pi(x)..h. i. of Ytm.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function. is also in use. It is instructive to pursue the appropriate arguments.89) Inserting this expansion into Eq.m = 0. (1 . i. respectively. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .
96) + V(r) . This is precisely the case.26)) yi(r = 0) = 0. these are of little interest to us here. first line of Eq. Eq.78) to be square integrable.E Xl(r) =0 (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.91) from which we deduce that K = 0.Qr2 with (cf.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. (11. Murphy [190]. but also those for a different series for a.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. For \x\ < 1 we can obtain from this the coefficients of a convergent series.2 + 2mo dr 2mor V{T)E Viir) = 0 (11.1.92) This is a recurrence relation. For the solution in accordance with (11. We have therefore (multiplying Eq. We see that this happens precisely when I is a positive or negative integer or zero. (11. l(l + l)h2 _2mo + 2m. . 93. (11. ai(re + !)« = 0. .1(1 + 1)].46)) p. M. However.93) where Xl(r) is the solution of the radial equation (cf. Margenau and G. p. they have to be polynomials (recall the orthogonalization procedure of E. when the series terminates after a finite number of terms. > 1. (11. (11. Eq.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. l 2 and lz are the solutions of the Schrodinger equation of the form (11.*" 11. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . Schmidt).5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.94) from the left by r) h2 d2 h2 2 + H! + \)IL.11.94) ^For further details see H.
(11. The equation also has an irregular solution. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. (11. Equating the coefficient of 1/r2 to zero. since it satisfies neither the condition yi(0) = 0. nor the condition of normalizability for l ^ 0. He + . oo) of a onedimensional space. and others.1)).1. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. L i + + . an analogous consideration applies to hydrogenlike atoms like.96). which behaves like 1/r1 in approaching r = 0. for instance.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 \ ). In the first place we investigate the behaviour of yi in the neighbourhood of the origin. and substitute this into Eq.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. s = 1 + 1. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r.102) . The Coulomb Interaction (11.1)+1(1+ 1) = 0. We now consider the hydrogen atom with V(r) — —e2/r (i.100) with (11. we obtain the socalled "indicial equation!'' s(s . 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. (11.e.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. Z\ = Z2 = 1 in Eq.214 with CHAPTER 11.e. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. r m0 = meMp me + Mv (11. i.99) e ip(r) = Eip(r). Obviously the irregular solution has to be rejected.
i. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. W. i.+if>x/2 vi(x).x ) $ ' .11. (11. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. see e.g.107) equation. We can see from Eq.529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.105) t h a t for x —• oo there is a solution behaving like exp(—x/2).105) we obtain the equation for V[(x).108) This equation has the form of a confluent x$" + (6 . . Magnus and F.v) vi(x) = 0. for E < 0 they > decrease exponentially. (11. We set therefore Vi X J. *The factor 47reo appears in SI units. (11. (11. p.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. hypergeometric (11.105) 11.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0.104) W i t h these substitutions we can rewrite Eq.6 11. It remains to determine the function vi(x) for a complete determination of the solution.e. Inserting the ansatz into Eq.103) °y moe2 ( = 0. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . (11.e. T This equation is known as Whittaker's equation. Oberhettinger [181].e. (11.88.a * = 0.1 Discrete Spectrum of the Coulomb Potential T h e eigenvalues Here we are interested in the solution which is regular at the origin (i.6. (11.
(11. it is convenient to use only the factorial notation.(0_i)!(6 + n l)!n!' (11. (11. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. The function T(z) is defined by the integral /co T(z) := / eHzldt. the factorial for an arbitrary argument. (4) r ( i ) = V5F.J. 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. called dupZication formula. particularly pure mathematicians.106). 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. with many of these around in some calculation.112) The gamma function is a nonvanishing analytic.e. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .z(zl)\ = z\. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. (2) r ( z ) r ( l . since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq.216 CHAPTER 11.! ) ! = imlFI' c a l led inversion or reflection formula.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (5)r(n + l ) = n ! . and for both cases. meromorphic function. . (11. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . ^ ( 2 ^ ) ! = 2 2z z!(z .z) = ^f^. and hence the *Some writers. (z)!(z . i. However. more precisely.1/2)!.
and so to the number of different values of m. n' = 0 . . (11. which is thus seen to be a consequence of demanding the square integrability of the wave function.1) + n = n(n .» • » • » .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. The quantum number n' is called radial quantum number.v) T(l + lv) _ ~ K sin{7rQ/ .2. Eqs.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ .V))T(v . • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. if the series terminates for certain values of u.101) to (11. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).1. every n. 1 .1) + n = n2.11.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. .. for z. One defines as the principal quantum number the number n.e. (11. which are also described as its nodes.e. Thus the infinite series must break off somewhere.I) r(vl p)sm{n(i/ I .104)) 1 n =v = K.e. in our case: a = 0. . = n = Z + l + n'. i. 1=0 (11. .. 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 .) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. . the orbital or azimuthal quantum number I can assume the values 0.110) as follows: T(l + 1 + p .. i. With (cf.. This is achieved. must be a polynomial. i. 2 . For every value En. d = 1. (11.
1. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. stand for I = where n denotes the principal quantum number and s. nf . The spectrum of the hydrogen atom thus has the form shown schematically in Fig. ne.p. The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig..1 = 0 (called principal series.1 = 1 to n = 2. 11.1 = 0 (called sharp series.\ I = 0 or 1.1.1 Hydrogen atom energy levels. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. . / = 1.l = 2 to n = 2.'. (b) The spin of the electron was not taken into account. hence p).. nd. respectively. hence d). and from n.. The following spectroscopic description of states is customary: nl —> ns. (c) The nucleus was treated as pointlike and not as something with structure.. and hence the fine structure of the energy levels is excluded... I = 0 to n = 2. 2 .. . from an initial state n > 2. np.4. 11. .§ One also writes n2l+1l.'. from n. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. 0. I = 0.1 = 1 (called diffuse series. from an initial state n > 2..218 CHAPTER 11. . hence s). .d. ' O n e describes as Lyman series the final state with n = 1. One describes as Balmer series the final state n = 2.
however.z) = (r + k)\ Lkr(z). . • (11. 11. Unfortunately the definition mostly used in mathematical literature differs. Apart from factors.114b) yields the binding energy of the states. Vol. for instance.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.11. .117) The function *Lk(z) is a polynomial of degree r. S. several different definitions are in use. in fact.2 (11. (11.p)\(2l + 1 + p)\ The spectrum (11.x). 844. a + 2:*) = t v . (u. and I. We mention here three different definitions used in the literature.115) is effectively the associated Laguerre polynomial normally written Lk(z). p.2. With the help of Eq.r f . 84. "1'f„+11" P] „.= t . . Each can be recognized by reference to the generating function. • fc.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). Gradshteyn and I. W. I. l .116) Laguerre p o l y n o m i a l s : Various definitions in use! The polynomial obtained by terminating the hypergeometric series (11. Ryzhik [122]. p. (11. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. . For clarity and later use we denote the function used there by its normal form L^. see.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. M. .6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. (11. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr).115) ' =0 (ri . formula 7. Unfortunately.6. the polynomial contained in y\ of Eq. Appendix B.414(3). 2 .2l { + 2.118a) where Lk. We distinguish between different definitions in use.1. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. Oberhettinger [181]. so that one always has to check the definition. k + l. Messiah [195]. 3 . Magnus and F.r = 0 . • ( .
3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. Sneddon [255]. L2(z) = l2z + z2. Example 11. which we write ^Lk. . ^ F o r this definition and the following equations we refer to I.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. . I..2.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.3): e2t/(lt) (1 . and Ll(z) = l. . Ll(z) = 33z + ±z2. L1(z) = l .„\*Kk. is defined by^ dk (—'\\k(r^\2 (11. N. 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.121) In Example 11.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. Messiah [195]. **I. p.122) "See A.z . 162 . L\(z) = 2z. pp..164. Vol. 163. Sneddon [255]. It is said that this is the definition preferred in applied mathematics.** The associated Laguerre polynomial here.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.d z—^z + {k + l~z)—+r Lkr{z) = 0.220 CHAPTER 11. r r Z r=0 v tr k k (11.„M„ z"*L. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. N. (11. Appendix B.1. There at some points n + 1 is misprinted n + 1.
118a) and (11. (11. '"''Comparison with Eq.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp. Jfpd>2s{p))eptdp. The limits will be determined later.120)) it)k& zt/{lt) ~*P p=k fc (11. N d rk 0.. (11.124b) Comparing Eqs.4 + 2z. Example 11.120).124c) Example 11. Then.119) derive the generating function of associated Laguerre polynomials (11.. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp. (11.11.121). and partial integration. and obtain with this the normalization condition (11. *L\(z) = . and *L\(z) = . (11.4z + .124a) (i*) pi The orthonormality condition is (cf. i L1(z) = l . (11.107) implies k = 11 + 1 and r = n + I.. . ^L\(z) = ..119): ty"(t) + (a + 1 .1 8 + 18z 3z2.8): (11.6 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 . Solution: We consider the following equation of the form of Eq.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.z . ^L2(z) = 2 . (11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z).1 .3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.
(11.. We demand (to be considered later for integration contours) the following condition. Hence the expression in square brackets vanishes. j p = (P 0 "V?. t > 0.q) and P~ t ( . p and one has (as polynomial of degree 6. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1).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. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. agreeing with Eq.a = 21 + 2. and (4) around p = 0 if b > 0 and integral.. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. Thus in the case of Eq. 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. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . case (3) similarly.^ .p + 1 = 1/(1 .118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. Case (2) implies the exponentially increasing solution ~ e + t ./ ( l — <?) —— = G(t).p ) « + i q c + i ( i .p ) o + i 7 t = 0 . Then dp = dq/(lq)2. with requirement of a finite solution.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. since for n > b : (b — n)\ —• oo. (11. (3) around p = — 1 if a + b is a negative integer.222 CHAPTER 11. the only possibility is the last. It follows that 1 d i I _i." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt.] from the above equation: C(p) := p(p + l)s(p)ept = 0.( a + l ) p s ( p ) + 6s(p) dp 0. . (2) from —1 to oo. 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. 2TT_ Jq=0qb+l(lq)a+l 1 .) 2 J J p 6 + i ( l . s bK> (lq)°+i (11. this is the rejected irregular solution.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)} . which removes [.
—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. (11. Clearly only the second form is sensible.6.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.11.?). Jo (11. where Fnl(x) = yi(x) (11. Then Nni is a dimensionless normalization constant and (note that we use the associated Laguerre polynomials as in the book of Sneddon [255]) Fnl(x) = xlex^LZ.125a) and the factor a~ 3 ' 2 with a = Bohr radius. (11. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = .3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! . m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O. Eq. so that I = (c + a)! 1 f .111). .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.—i. is introduced for dimensional reasons.(x). and 0 if cj=b. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as .127) . 11. The wave function of the state (n.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.112). I.
1 + 1) (11..2Z . For the ground state (n = 1.7.n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .2.! ] + (21 + l ) 2 . 2 (11. = 0) we have ) r /u =. (11. This behaviour is indicated in Fig. from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix ()r2dr 2 ^i + /„. a (r)is = ~a.3fc + 2).128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).131) . and for the explicit derivation of the second case below Example 11. The Coulomb Interaction One finds in particular (see also Example 11.v (6n2  6nk + A:2 + 6n .(l) n+l. + .(2Z + 1) + 1} = ^[Zn21(1 + 1)]. thus on the average the electron is the farther away from the proton or nucleus the larger n is.224 CHAPTER 11.129) n2a' Analogously we have roo j = p2 I r n \ 2J ° J0 'V^yr nl _ na J^° xF^(x)x2dx _ na f^SJn+K 3 ) \na) na {6(n + Q2 . 11.3(2Z + 1) + 2} 2 {2(n + 0 .130) We see that the mean value of r is the larger the larger n is.6(n + Z)[(2Z + ! ) . (11.
132) (11. = l a nl n N'nll rFnl. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. or (r ) ~ n a 2 2 ^ [ 3 n 2 .2 Hydrogen atom wave functions.J n 1s 2s 3s *r Fig. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . we have and tp becomes particularly simple. with n +1 replaced by n and 21 + 1 by k.e. i. Problem 5. \f{r ) ~ n a.6 The Discrete Spectrum of the Coulomb Potential r. pp.l ) n ] =  [ 2 n 2 + n] an[ n + (11. 173/174.^ We obtain therefore for the quadratic deviation (r) .2 I 3/2 Kl r 225 \2 1 R„.134) *These are taken from I. In the case in which I assumes its maximal value n — 1.2 1 A r = V (^2) .( n . 11.(r) = x W 2 n + 1 =.11. v ^ V2n + 1 (11.j .133) for large r.20. S. / ! J — .. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 . Sneddon [255].
Thus. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. Since 6 is a cyclic coordinate. we consider in Example 11. .135) This expression becomes very small for large values of n. i = 0. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.e. However. but rather like classically in a plane (i.6. r. \/v vT 7 v^ See also Example 14. for these values of I it is not distributed with uniform probability over the spherical surface. The Coulomb Interaction or TT = ^ = (11. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. In order to obtain an impression of relativistic corrections in a simple context. (c) The nucleus was treated as pointlike and not as something with structure.3. Example 11. the states Ytl are predominantly concentrated in the xyplane). „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler .e.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). the electron remains practically localized within a spherical surface of radius (r). relativistic corrections are of the order of v2/c2 ~ 0(En/moc2).226 CHAPTER 11. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. and separating this in cylindrical coordinates r.r. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.— > where ampler is the length of the semimajor axis of the elliptic orbit. = ngh. in which periods are obtained from the BohrSommerfeldWilson quantization condition. i. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". p2 = p2 + — pg • eV. 6. This means. we have pg = const. pJ = —— and j> Pidqi = nth.4 the relativistic equation in cylindrical coordinates.
( v c + i ) . where r = ^/x1 + y2 + z2. 11. <p = <p.2. Schiff [243]. which correspond to a new choice of polar axis. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'.3. .. 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. as also indicated in Fig. 2 2 Therefore the energy is no longer a function of n = nr + ng.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16.* See L. where the constants are identified by comparison. (11.136a) (11.9. which are related to Cartesian coordinates in the following way. r\ = r ( l — cos#) = r — z. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . (p). The deeper reason for this are symmetry properties of the problem. it is possible to separate the Schrodinger equation in other coordinates.T). 11. y=y/%qsanp.136b) £ = r ( l + cos#) = r + z. ip are the spherical coordinates.V .(p .11. z=^rj). 86. y — 0: V ^ c o s tp. where ip = 0. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. p. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. Whenever there is degeneracy.6.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines.
there is really only one basic equation to solve here. We rewrite the first of these equations with the help of the relation M (w ^  m + e) .136b) the Schrodinger equation of relative motion (11. . The Coulomb Interaction Fig.228 CHAPTER 11.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. 1 . so t h a t the variables can be separated.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. with zaxis as axis of rotation. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . . = 0. for uniqueness of the wave function. In terms of the coordinates (11. . a separation constant. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A.3 Parabolas £ = const. and r\ = const.22) becomes — as we show in Example 11. (11. 2 . 11. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike.
ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).o. " .114b). We leave the calculation of the degree of degeneracy to Example 11. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . as well as the Laplacian A = V and the Schrodinger equation. Eq. 1 . p = /?£ and cr = /^r?. E x a m p l e 11. . (11.6.105)) Hence we write I = (m — l ) / 2 .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0.± ( m + 1) = 0 .e. 2 . . ds2 = dx2 + dy2 + dz2. . . i. Adding now / . . .11. . m0Z1Z2e2 ^ 2 = n (say). To obtain a closer analogy with the separation in the spherical polar case.106) and (11.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . n 2 = n" . In the spherical polar case we had (cf. 1 . .5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric.125b) (dividing by p 1 / 2 ) £ oc e . . . 2 . (11. . T h e n by comparison with the spherical polar separation the answer is as in Eqs. we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". from Cartesian to parabolic coordinates.
.2) and summing over m. We then sum all possibilities and select the coefficient of ojn~1 in the result.2 . (11.[df + dr]2 . W. Thus. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly).e. </. last chapter. z = (£ — rf)/2. In the case m ^ O w e have n\ + 77.. Oberhettinger [181].2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . see Eq. (11. when m = 0 = 77. " . (11.1 in this expansion is seen to be n. / J„^2 . 2 2V? V^ One now performs the square of this and obtains similarly dy2. We use the method of selector variables. \— < / — cos ipdr] — \/£r] sin ipdtp.g<p 9 \ . N2 where g^dS. and so dz = (d£ — drj)/2. > . Solution: For the wave functions ip = £(£)J\f(r])&(ip). all cross terms cancel in the sum of 7 the squares. we have to find the number of values of the quantum numbers n\. Observing that the highest possible value of m is n — 1 (i.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. i. and attach a factor to to each term introduced in the wave function. . etc. 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. (n _ !) n = n2 §See e. n the double sum ]T]2 is TI — m. = e±im^e("+^'2{par'2L^{p)L^ (a). are elements of length. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. / — cos ipdl. n 2 (1UJ)2 The coefficient of a . Similarly we proceed with dx and dy.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. Thus dz2 = . and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.114c). We consider first the case m = 0. 712 and m which give n.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n .g.2 = n — 1.2)d/2.l)a + (n . The Coulomb Interaction Solution: We begin with z of Eq. 2 _ = (9id02 + (gvdvy + /„ . Here we have (see preceding text) 711+77. 2 + ^dV . i.e.2d? dr)}. for instance. Beware! For m = 0 there is only one mstate. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .136a).tx2 .l ) ( n . dx = d{ y £77 cos ip) = . . i. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. {g^Y.2)d} = (n. Magnus and F.. 0 there are two.136a). for m / .1) ~ .e. as t h e two signs of t h e exponential indicate. Eq.230 CHAPTER 11.E E n l „ . + l 4 {^^)drl f i + v \ .e. Thus we want to obtain the coefficient of wn~1 in £2 .
With the help of the wave functions we derived above. we would have to use degenerate perturbation theory and obtain the same result. and determine the average value of v for a given state.) = ^eF(p .7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. with Z\ —> Ze.drid<fi oc (p + o)dpda. (11. Z2 —> e. = j (2ni+m+l). (11. L ni\ ny. where F is the electric field.128) if one makes there the substitutions n —> n\ (or 712) + m. with the connection » (11. which will then give the change in energy of that state to the first order (i. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.11. . if we expressed the perturbation in spherical polar coordinates. However. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. s 0/ .6).128). Ikr+k.6 The Discrete Spectrum of the Coulomb Potential Example 11. we can relate the expressions Iq to those of Eq. .•* n (l) = / Jo It follows that . ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. . i.124c). .e. (11. 8. i.r.™(p)] 2 By contour integration (see Examples 11.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in.127) and setting n = r + k. 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.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. fc — m. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£..e.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I.e. with z = r cos 9. . these expressions agree with those of Eq. Sec. I. In fact. Thus e. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—.8 and 11. with RayleighSchrodinger perturbation theory). with Eq.1 (cf. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. eF{z).g. {p — a) — = dxe~*xkvLkn (z)] 2 .a).—r~n = 3(mn 2 ).
c positive integers.124d) of the alternative integral representation obtained in Example 11.232 CHAPTER 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 .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.9: 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.p ) r + 2 [(lp)(lg)]"l+1(lpg).b.g )(l.9 )« Solution: Using the generating function for both associated Laguerre polynomials. The Coulomb Interaction E x a m p l e 11. (n + r . dt = ds/a.1)! n\ This verifies the expression I™t in Example 11.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l . E x a m p l e 11. tt„:_ (1 . s = at.( m + 2 ) is equal to the coefficient of (qn — qn_1) in (1 — pq)~(m+2\ The coefficient of qn in the expansion of (1 — pg)~ M is pn(n + fi — l)!/rt!(/i — 1)!.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .p ) ( l . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 .7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.p ) ( l .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 .. the integral / is the coefficient of pnqn in the expression _ / .^~T^——^ ) = coefficient of (pn .9 we consider integrals with an arbitrary power of t. .1 ) ! (n . (11.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11. In Example 11.a.3 by one power of t.pK — — —L .p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . (11.3: /"OO I := Jo et[L^{t)]2tm+1dt.x ( n + m)! ( n .l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.128). and is a special case of Example 11. the integral arising here becomes with Eq. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. 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 .3).l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t . _ „ .
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 «. ° ^ + a ( l . Using the contour integral representation (11. c + b — 2r = l (i.w(r) = and / a ) = r\(P ~ r .124d) of an associated Laguerre polynomial.l ) * x coefficient of a* in the expansion of 7 ( a ) . c + b even) and/or j / = 0.^ ( l .?(t)t a e.L.™( r > (1 + a)P r £^0y^w Jc+6_2r = . c + b odd).6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. .p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . we have to have y + (c + b — 2r) = 1.a _ 1 = [1 + a ( l . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.e.1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.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. i. so that = (q+fc±£) .Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.r ) ! a ' .f ].111) of the gamma or factorial function.e.e.p) . i.q)(l . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.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 .(i) = i ! ( . (lp)(lq) + a(l q)(l .(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.e.p ) ] . we have 2r = 6 + c = even. c + 6 — 2r = 0 (i. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l .11. .a t [Lg(i). (a) j / = 1. — (cr)!r!(6r)! Kv. 7£. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . where (a+**£)! c j n Jo = .
^A very readable source to consult.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. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r).1%^ 11. 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. cannot really — and this means in a rigorous sense — be applied in this case.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. These results are seen to agree with the expressions 1^. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. Since such a rigorous treatment is beyond the scope of this text. b\ V cited in Example 11. is the work of J. + (!±<)«. also with regard to earlier literature. .)!' f^)(. Taylor [267]. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. 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. the standard principles of nonrelativistic scattering theory. t We have the Schrodinger equation of relative motion of particles with charges Z\e. . R. however parallel to the Coulomb case. In view of the slow falloff of the potential at infinity.7.~* Hence (recall that 7£ c (i) = i!(—l) l x coefficient of a% in / ( a ) ) (a+^)!(/3a) r r in (£f^)!(^)!(^±£)! t J = ( a + b)\((3a) > 6! _ (a + ft)! = [a + lb + l ) a .^)(.
. i. . In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). a(a +1) z2 ] (11. Oberhettinger [181].b..z)). b.143) and v = ikrj. where ^. (11.144) *(a.b. i.140) and using (11..e.+ ^  + . that we encountered previously.7 The Continuous Spectrum of the Coulomb Potential Then Eq.z).M) = l + . (11.z) = Wi(a. v = ik(r .z).136b). pp. (11. (11.140) The simple Coulomb potential permits particularly simple solutions.146) See. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a.141) Inserting this expression into Eq. with the asymptotic expansions (11.b. what is the > behaviour of 3>(a. z = rcos<9. 86 .145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). „ az 1. (11.138) becomes 235 ^A + k2 . W. (11.^ W ) = 0.e. According to Eqs.142) This equation is of the form of a hypergeometric equation.108) and (11.. for r — oo) • ipr = Aeikzf(r . Our question is now: How does ipr behave for r — oo.109) we can therefore write the solution ipr = Aeikz$(i>y. Magnus and F.z) + W2(a. for instance. A = const. (11.11. (11.87 under Kummer functions. .
l.236 n\Zzab CHAPTER ^T(n '£ fo 11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.154) .kz+l^ktrz)) C . i.149) (11. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.151) with Eq. = r cos 0.(11. (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.z)T(l .ik(rz))} =^ + ^d.i 7 ) (11.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.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) .148) . We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. with 2. if tbr were „ikr ibr ex el"z + f{6) r (11. > A = Aeikz[W1(i1.7 ln*(rz)) k(r . Without t h e logarithmic phase factors.150) ^(rz)]1"^ 1 + 0 T(«7) i.e.e.a) (z) na)"~ r(la) T(ba) n\ ' (11.152) Since z = r c o s # .z) + la)T(n + b.l.b.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].153) The asymptotic solution represents t h e stationary scattering state (E = const. (11. J\") ci(kr~/\n2kr) (11. The Coulomb Interaction W2(a. (11.^l*7 l + O (11. a comparison of Eq.
2im. also called phase anomaly. the current density is * h . For the incoming wave ipi = exp(ikz).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.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. 11. (11.7. In other words.n .Q hk mo v0. that we encounter here is a characteristic of the Coulomb potential. the effective range of the Coulomb potential is infinite. we see that this expression is the exact analogue of a current density in classical considerations. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. do dQ K 12 K Fig.156) . We define as current density the quantity h J := 2imo [^(v^)v(v^n.. The problem of the logarithmic phase.4 Variation of the observed differential cross section with 0.11. As a consequence of the masslessness of the photon. (11. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. 11.155) Since the gradient is proportional to the configuration space representation of the momentum. z — — oo. The quantity /(#) 2 would then be a measure for the scattering in the direction 9.
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. That this formula retains its validity here in quantum mechanics is something of a coincidence.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq. thereby screening itself off from the surroundings. is called scattering amplitude.159) as the Rutherford formula which can also be derived purely classically. i.e. in the current this would in any case contribute only something of order 1/r. This implies that scattering takes place for the attractive as well as the repulsive potential. (11. 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. We recognize the result (11. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . Correspondingly the other part of (11. for 0 —> 0.238 CHAPTER 11.154) we chose the prefactor of the incoming wave in (11.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. We make a few more important observations.152) as 1. (11.153): 4fc2 s i r  The expression /(#). 'screened ZlZ<2e2 cr/r0 ° . In Eq. which yields the scattering cross section. We see that a depends only on the modulus of the potential (i. the square 7 2 ). since in actual fact the Coulomb potential occurs only in a screened form.e. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). In reality this divergence does not occur.
k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. be given by ^ f dk AL. whose maximum moves with the particle velocity. (11163) (11. A wave packet with momentum maximum at k = ko (with momentum p = hk) is therefore given by the following expression in which xo can be considered to be an impact parameter.2 fr2 . and if H is timeindependent one has the stationary wave function ^(x. The wave packet is a superposition of waves ip(x.x° ' r j . _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .11. i. (k). of the equation r fi. and which provides the uncertainties arising in quantum mechanics.t). (11. t 0 ).8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.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. / dk 7^340(k)4(xxo. t0) = e.165) ^ ( x ) = £ fe Vk(x).164) Ek~Eko + {kk0)v0h. 2 (11162) where Vk(x) = ^ ( x .t) = e . t) = U(t.i i f (''°>/Vk(x.166) Let the incoming free wave packet at time t > t$. The particle velocity v is v = — . 2mo V + F(x) k (11.t).yJ ^ j A k o W 727)3/2 e 0. which in the case of free motion (zero potential) are simply plane waves.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * .4. 11. the wave packet in the absence of a scattering potential.e. (11. v0 = .161) where Ak0(k) ~ A(k — ko) and ^ k (x. m0 m0 and V.e.^ ^ % ( x ) . i. of continuum wave functions. 11. t 0 ) . t 0 )^k(x. i.e.
(2TT)3/2 eik* + —fv(P.x °Vk(x). k = kez.166).10 we show that the solution V'k(x) of Eq. (11.172) h h k0 {rv0(tt0)} (11.e.167) by replacing the plane wave exp(zkx) by this scattered wave.xo.e.173) .169) The wave packet in the presence of the potential is obtained from the free form (11.iB *(**°)/ R °'tj + 1 pikr e * .^ ( * . Then with k = ^(ko + k . With Eqs.k0) ~ k0 + and the expansion of E^ in Eq. Eq. this solution > has the following asymptotic behaviour in which f(6.161) and (11. The Coulomb Interaction In Example 11.240 CHAPTER 11. ^k(xxo) ^k(x) ~ = e.ko) k0 (11.x 0 ).169) and (11. i.ip) is the scattering amplitude with respect to the incoming plane wave. From Eqs. (11.171) The phase of fk(0. the stationary scattering wave.167) we have ^k0(xx0. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf. i. (11.t o ) / V k ( x .<p) + Ol Akr l (11.165) ko • (k .154)) that for x — oo. (11.k 0 ) 2 ~ Jk20 + 2k0 • (k .x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11. (11.i) dk (2TT (11.163) we obtain </>ko(x . ip) could be carried along but we ignore it.170) A ko (k) e .ik .t) = J ^ A k o ( k ) e .
r ' ) = 47r<5(r .10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.154).to). t) . 11.r')U(r')ip{r')dr' /4T7 (as we saw earlier) .to) has t h e form of t h e wave packet in radial direction as indicated in Fig. (A + fc2)t/. Solution: We define the Green's function G ( r . (11. '.5. T h e scattering amplitude is t h e same as t h a t given by Eq. *.11. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.fco(tto)/ft]^0o)(x/xo. t) f.r ' ) . 11. E x a m p l e 11.167).8 Scattering of a Wave Packet so that with Eqs.(«)/ ipko(x .x 0 .^y /k0^. the solution can be written iMr) : (2TT)3/2 ^dr'G(r. and determine the Green's function G(r.e. (11. U(T) = ^ V ( r ) . (11.174) r /k0(^.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.y)eiko'Xoe'[fcor£.5 Scattering of a wave packet with scattering centre O. Adding a suitably normalized solution of the free equation.r'). however.171).*0ko (x .z Fig. A particular solution of the Schrodinger equation is — J G(r.r')V(rW).x 0 . Here ^ (x' — xo.r')l/(r>fc(r'). . i. with the energy of the maximum momentum of t h e wave packet. r ' ) by the equation (A + fc2)G(r.
we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' .O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. 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.6 The contour C+ in the upper half of the k'p\ane. These are the socalled outgoing and ingoing Green's functions. displaced slightly away from the real axis as indicated in Fig. which is such that the contribution along the infinite semicircle vanishes. e > 0 small. 11.fc ) oo dfc' 1 d nik r 7rr dr / .242 CHAPTER 11. dk' ik'*r 2TT2 fc'2 .fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. Thus we consider the contour integral taken around the contour C+. The Coulomb Interaction We find a set of Green's functions G(r. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r .6. Imk . Hence we write G(r)G+(r)./2 dk'd(—cos 6)dip iki.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 . . one obtains another Green's function given by G_(r)= r eikr. 11.Rek' kfc/2K Fig.=k+ie/2k 1 d f.
the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems.11. i. .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. Eq. Thus we proceed as above. b.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). 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. Eq.0 + W2(l + 1 + iy.e.178) (11. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. under the stated conditions 2/07 = Z\Zie2 = —Mo.£). b.g.9 Scattering Phase and Partial Waves 243 11. where Mo is a parameter in Chapter 16. Then £^<fc + (2l + 2 £ = ~2ikr. (11. z) + W2(a.z2 27fc 1(1 + 1)' yi = 0. whose continuation to infinity is again given by (cf. If we proceed from the radial Schrodinger equation. b. Before we introduce the general form of the socalled partial wave expansion.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. i. (11. e. (11.176) " 0—A ~ (I + 1 + iy)<t>i = 0. Then yW (11. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. z). (11. 2ikr)] [Wi{l + 1+11.140) (previously we solved this only for the bound state problem) § .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. (11.2Z + 2.145)) F(a.e. 21 + 2. z) = W^a.2l + 2.
n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .244 CHAPTER 11.180) ^ ' (11. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 . +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) (11.146) and (11.j syyll iy il( _ .' . The Coulomb Interaction Using Eqs.ry) = e"^r(Z + 1 + ry).l e*"" + T(l + 1 + H ) T(l + 1 . (2kryr{i" V 7 . .181) so that eldlT{l + 1 . (11.183) r(Z + 1 + 17) := Reie.i f c r + i 7 In 2/cr+i7r(i+l)/2 + We set T(l + 1 + 17) = r(z + i + i 7 ) ' r(z + I .i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 . T(2l + 2) (2fcr)*T« (Alliy ikr .«7) „ . Then e2i5< = Reie Re~ J2i9 9 = Si.ii)' (11.
. (11. However.114a) reveals). i. .. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. i. i.11. That the energy E appears in Eq.e. with Eq.1. In the present quantum mechanical considerations I is.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). i. considering I —• an(E). kr) the regular solution with the following asymptotic behaviour: r 0 (11._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).e. (11.n ' .139) n' = 0.e.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. a positive integer.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function).e. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element. g = e«* = rSi + 1+ *V*ri (11.187) contains almost the entire physical information with regard to the Coulomb potential.. One defines as regular Coulomb wave Fi(j.188) in the form y/E. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. 245 (11..2. (11. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . considering I as a function of E extended into the complex plane. For instance the poles of S are given by Z + l + z7 = . as we saw. where the continuum of the spectrum starts.
For the Coulomb potential a typical such trajectory is indicated in Fig.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. we obtain the partial wave expansion of the scattering amplitude as follows. 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. (11. and thus in general.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. 11. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.7. we obtain trajectories which are known as Regge trajectories. The Coulomb Interaction and plotting these for different values of n.7 A Regge trajectory of the Coulomb potential. Singh [252]. 11. (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.246 CHAPTER 11.191) . In these cases. 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.
g. Magnus and F.i f c r . 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.eikr\P^cos9). The incoming plane wave can be reexpressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase. . (11. (11. The potential disturbs this phase relationship (by delay or absorption or redirection).22. Oberhettinger [181].11.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.195) with the scattering matrix element S(l. r e. W.~ J > Z + l)Lil*e~ikr .192) Thus the incoming and outgoing spherical waves have a definite relative phase. There the expansion is given in terms of Bessel functions Jv which have to be reexpressed in terms of the asymptotic behaviour (1 2) of Hankel functions H„ ' . 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 ' . 21 .^2l + l^S^k) lik (11.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.196) "See e. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. The mathematical expression can be found in Tables of Special Functions (there cf.k)=me2iS^k\ in which r\i^\ indicates absorption. pp.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.r ^ V ^ Wostf). (11.
This page is intentionally left blank .
In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. can be found in any traditional text on quantum mechanics.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. states of a finite lifetime. Razavy [236]. In fact. These newer methods will be presented in detail in later chapters. i. It is not surprising. namely the instanton (or more generally pseudoparticle) method. which a particle in classical mechanics would never be able to intrude. was developed only in the last two to three decades. that quantum mechanically a particle may have a small probability of being in a spatial domain.g. i. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. to explore first much simpler models in order to acquire an impression of the type of results to be expected.e.e. therefore. which in one form or another. in texts on quantum mechanics is rare. however. It is very instructive. as — for instance — for trigonometric or double well potentials. that the explicit derivation of such quantities in nontrivial contexts. M. 249 .* *See e. For various general aspects and applications of tunneling we refer to other literature. in general. Similarly difficult is. one of the main and now wellknown methods to perform such calculations. the computation of finite lifetimes of a quantum mechanical state. Such a vital quantum mechanical phenomenon is socalled tunneling.
(12.e.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x.250 CHAPTER 12.t) (12.t) = HiP(x.1) as the probability described by ijj(x.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. t) was developed by Born in 1926.4) (12. so that p = T/>2 is the probability density for a spatial measurement at time t. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x. ..t). i.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. r (12.t) that the one particle concerned is somewhere in space at time t.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics. ihr(x. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.t)\2 =1 (12. Quantum Mechanical Tunneling 12. M^ Then dt + Vj 1=0. According to Eq.
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. Gosdzinsky and R. 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. (12. 1 0 ~x o a= f See L.12. 2m E 2 V0 > 0. t For the singular delta function potential a modification of these demands is necessary. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. the problem is analogous to that of the scattering process from a potential in a 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. as shown by P. ~hr> k = n^ (12 7)  . Delta potentials in more than one dimension do not permit bound states and scattering. even for discontinuous potentials. 12. where „ mpVo V(x) =+V0S(x). although in general small. higher dimensional delta potentials do not have properties like that in one dimension. pp. which is the case we consider here. Nonetheless. This case plays a special role.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.6) . This effect is known as "tunnelincp.33. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory.2 . However. 'Maybe the reader dislikes being confronted again with the onedimensional case.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. In much the same way that we define there reflection and transmission coefficients. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. The Schrodinger equation for this case is* ip" + [A. Thus it is possible that the probability for the particle to be in such a domain is nonzero. we can do this also here. the equation can be integrated at every point. If ^>(x) and V^>(x) are known at every point x. Tarrach [119]. Schiff [243]. as we mentioned above. If there is no absorption. 29 . Such a region is for instance the domain where V is larger than the total energy E.2a6(x)]il> = 0. From this follows that both p and j have to be finite and continuous everywhere.
1) .= 2a^(0). .(ik . except in domains of the potential V.ikB) = 2aC or 2a(l + B).ik(C .[^']o.e.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes. (12. a 2 — a 2 a — ik a2 + k2 ' . Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.13) ~A R= for x > °(12.2 a J 5{x)4>(x)dx = 0.' : It follows that ik . Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).: (i2iQ) Ce for x > 0. i. i. a — ik £Qr < { eikx a eikx •t" .12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. (12.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.e.ikC = i.12a) (12.e 1 + B = C.252 CHAPTER 12. (12. i.9) for ij.11a) ikC . [<//]0+ . e Q (12.8) (12. In the region x < 0 we then have both types of waves: the incoming or incident wave to the right as well as the wave reflected from the potential. 2 (12.11b) 2aC. The continuity conditions at x = 0 are therefore for V : and according to Eq. (12. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x . that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .e.
i 8 ) .e. Proissart [223] or other books on scattering theory like R.2 = . See e. R.e. 2m 0 (12. ( 1 2 .ip'(0) = .K C . Newton [219]. K = a.2 a C .2 o ^ ( 0 ) .17) da#(s) 2 = 1.(K)C = .19) Thus the delta function potential supports exactly one bound state. the quantity T. i. at k2 = —a2. oo (12. possesses simple poles at a ± ik = 0. . poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. = . and C follows from the normalization: OO A. We write the solution ^ so that ^(0+) = ^ ( 0 .12.e «x = J \ CeKX for for x > 0. Omnes and M. or h2 E = —a2. i.K 2 < 0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.a l s l. f Ce~KX c. Thus ^ = Ce. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0.g. when normalized to 1.) and V>'(0+) . z<0. as coefficient of the outgoing wave. 'Note that the physical pole has negative imaginary part.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. (12.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. .
1. x€[a. 12. Fermi [92]. —a]. (12. The latter contains also specific applications in nuclear physics. oo]. x G [a. p. 58. Schwabl [246]. \x\ < a.22) + De Feikx + Ge~ikx *We follow here largely F. . or in the lectures of E. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.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. V(x) V=E a v=y Fig.a]. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.20) for for \x\ > a.254 CHAPTER 12. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.4 Scattering from a Potential Well We can immediately transfer the above considerations in a similar way to scattering from a potential wen* as indicated in Fig.1 The square well potential. (12. We want to consider the case of stationary states for E > 0. 12. 55. Quantum Mechanical Tunneling 12. p.
26) K^gifoa—i/ca fc.27) v ' Similarly Eq.. . ( V C n vD ). (12. I (12.ikBeika = inCeiKa + DiKe~iKa. (12.24) yields F ^ G. (12.12. ).25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.e. Fikeika .f )e" t\ K\0—tKa—ika .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.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 \ .24) The first set of equations can be rewritten as / ( p—ika / p—ika pika pika U pika A \ \ / \ / 1 pina Kpina p—ina Kp—ina \ \f ° ){D ) \ B ) ° or A N \ Bt i. fc Af(a) = . M(a) .ikGe~ika. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12.iKa ? ikAe~ika and . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . (12.
(12.e. . xe[a.32) [cos2«.35) with with with x G [oo. \R\2 = T(E)^sm2Ka (12.^esin2«a]e 2lfca \r\ sin 2/ta Thus if we now set G = 0 in order to have only an outgoing wave proportional to exp(ikx) to the right of the well. = < Ce~lKX + DeiKX .4T(£)e ifci.(1 + e 2 ) sin2 2na 1 .a .37) . Quantum Mechanical Tunneling Hence with the complex quantities IK (12.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.oo]. (12.a].34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .e.256 CHAPTER 12. i.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.34) into Eq.e sin 2na IJlika e B = r)F sm2na. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.a ] .31) we have A AT(a) = )MFo B ^r)sin2na . (12.—2ika [cos 2/ta + \e sin 2«a]e~ (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. i. (12. x € [a. we obtain A = F( cos 2na — .
in the domain of A imaginary and K real. (12.e. 83. 2m0\E\ h? ' (12.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. However. (12.e.12. an equation of the type f~1f = g19.44) . Dwight [81]. B.43) Thus we set k — —i ' H . they have solutions. (12.40) we obtain cot(na) — tan(/«j) — i K IK . and k real. (12.41) i. (12. which is satisfied if either / = g or / = — g'1. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)].31): 1\T(E)\2 257 1 .e. We see in particular from Eq.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.39) 2 i. .4 Scattering from a Potential Well On the other hand according to Eq. (12. 2\k K. nor when both parameters are purely imaginary. Li ( — H— I sin 2na = 0. i. (12. for VQ < E < 0.12. formula 406.42) These equations have no solution for K.e. (12.38) as expected. i. p.35) and using Eq. cot(fca) = i— or tan(fta) = K K .
258 and obtain the equations CHAPTER 12.45) These are the equations that one obtains from the connection relations for the case of Eq. and is itself infinite there.49) Taylor expansion of the denominator in powers of (E — ER) yields . The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. (12. A = 0.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . Schwabl [246]. This can be understood. .35)) for E > 0 sin2Ka = 0. . F. „ . Quantum Mechanical Tunneling . (12.g. from Eq.' We observe that T(E) has poles at those values of E which supply binding energies.42) which then leads to bound states with associated even and odd wave functions.e.e. (12. n = 0 . (12. Eq.65. Also. Then according to Eq. . (12.fl = ( .48) We expand T{E) around these energy values. At their positions cos(2/«z)£. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1.46).47) These states in the continuum are called resonances. 2Ka = nn. (12.e. 2m0\E\ Ktan(«a) = — y——^— .50) See e. pp. 64 . k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. . ± l . i.34).46) This condition implies the eigenvalues E^ER = n2^^ V0>0. (12.l . t2 2 (12. The remaining wave part becomes exponentially > decreasing. 2m0\E\ and Kcot(Ka) = y—^—L.20). i. around the values given by Eq.c o . We now return to the scattering states. t&n(2Ka)ER = 0. T(E)e2ika = — r . i. The amplitude of the transmitted wave assumes its maximum value where (cf. (12. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . = 2/rn (EER). (12. ± 2 .l ) n .
A potential of this type is known as a PenneyKronig potential.2 The PenneyKronig potential.e. .51) i. 12. V(x) x=2a x= =2a . = 0. V(x) = 9(x + a)VQ6(ax). 2a<x<2a.52) Here ER is the energy of the resonance and 2/F the lifetime of the state. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. a 3 a. T(E) has poles at E= ERi: (12.5 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. This result is physically very plausible. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.2. s \ V0 /• ~\ 'X . 12. These are precisely the resonance states.12. We restrict ourselves to the period — la < x < 2a.' ''' 5a 3a \ a ~~~.' ' 5a \V / Fig.^ l / . . Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. 12.
53) *w«w^( « j)'*^{i i ^ .2. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo..Vo) n.55) The Hamilton operator of the problem. i.56) Thus the problem has a twofold degeneracy: Associated with every eigenvalue E there is an even and an odd eigenfunction and h2 n 2 7T 2 E — En > 2m0 4a 2 (b) Next we consider the case VQ > E.. with V as parity operator.2 = W h 2 _„2 = 2mo(E . E^En = ^ ^ t n = 0..260 CHAPTER 12. and hence for = Eip(x). (12.e.. and we choose (12. i.e.1. . VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). (12. . x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\..a ) = 0 = ^(a). HV^(x) = H. In view of the periodicity we have ^(x) = ^(x + 4a).2 < a . We set . = EVip(x).e There is the additional boundary condition (for Vo = oo) ^ ( . This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. the differential operator is even in x. 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.
il>]?0..Be Ka (12. we shall not exploit here. This wave function is no longer restricted to the interior of a box. 12.2).2a].12.. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . however.59) Asink(x + 3a) with x & [~2a.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a. xe[a. that these are such that for VQ — oo they approach those of the previous case. BeKX±Be~KX with xe[a. We > obtain even and odd functions ip± by setting ^ = ±1. (12.2a. We demand these for even and odd wave functions ip±(x) which we can clearly construct from ip(x) (as examples see the dotted lines in Fig.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka]. ^kAcos{k2a).58) D sin k(3a — x) with xE[a. The central region of the solution is the classically forbidden domain. V'(O) = const. of course.63) = KBe~ F = . (12.2a]. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. = Ka ±Asin(k2a). (12.60) The boundary conditions that ip± have to satisfy. — a]. ±Asink(3a — x) with x€[a. (12. We therefore have different boundary conditions. are ^_(0) = 0. We expect.a]. which. (12. V+(0) = const. VV(°) = °. This means that i>±(x) = <  = ±1. At x = ± a we now have to connect ip and ip' of neighbouring domains.a]. so that (W meaning Wronskian) W[tl>+. — a].62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] .
i.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .e. where Akoce'2™. F ^odd (12.66) Inserting these expansions into Eq. (12.65) and we expand tan(2afc) about nir. in the present case it is easier to continue with the above equations. i. Taking the ratio of the equations in both cases. say.e.2 e . 12.3 Level degeneracy removed by tunneling. Quantum Mechanical Tunneling Fig.e.even.. (12. i.64) we obtain ~[(2ak k .Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.nir) + OU2ak . say A. for VQ — oo.+ • • • 1! cos^(n7r) 2ak . + 0 ( e . It is plausible therefore to use appropriate expansions.nir + 0[{2ak . and is — as we see — less than that belonging to the lower sign case. (12. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—. we obtain k v ' eKa^eKa  tanh(«a) v ' These are transcendental equations for the determination of the eigenvalues of the even and odd eigenfunctions. = AA.— (1 ± 2e~2A + O (—). we set tanhx = l . » we expect the eigenvalues to approach asymptotically those of the case Vo = oo.67) The value of k belonging to the upper sign. the odd solution with. However.mr)2}. is the one belonging to the even solution. i.262 CHAPTER 12.e. (12. k = fc0dd.68) . In general one now rewrites the equations in matrix form. (with reinsertion of nir/2a for k) 2ak ~ nir . For large values of Vb.nir)2}} = 1 ± 2e~2Ka.2 a .
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.^ 0 = °' Hence in dominant order •00 — exp .70) xo = (a + b) 2' . (12.** " W ~^7~ Fig.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. ^v + ^[E. as indicated in Fig.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. 12. 2mn.e. w(x0) •• = = mogy(x).v = 0.. V ^ 0 are given by the equations ^o + " J . p. b > a. Solution: The wave functions i/>o>Vv in regions V = 0. . i. exp .4 A car meeting a bump in the road. Example 12. 0.2 dxy/V(x) (12. Classically the car is unable to overcome the bump. exp r rb rb / •J a 2mo (V(x) .4. 12.3. 1pV 2ro 0 4>o '. '. .. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. and is a reformulated version of a problem set in lectures of Fermi. 57. Fermi [92]. Compute this probability. .12.V{x)\i. but quantum mechanically there is a finite probability for this to succeed. y(x) = sin and y(xo) = 1. 12. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation.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. 2m0E „ 2mnE . .
formulas 3. p. h = 1. We can now evaluate the probability.72) Assuming a mass of 1 ton of the car (i.2 . Jahnke and F.y) is the beta function (see Eq. and assuming a length of b — a = 100 m of the base of the bump. (15.**. : 2 ^ B ( ^ .1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.621. Emde [143]. n I .e.1. which give /•TT/2 dx sin'M .4 x 10 3 9 ]. ^ ) = 2 ^ . Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. (12. **For instance E.2. S. Ryzhik [122]. 1000 kg) and using the following values of the natural constants: g = 9. p. . 14. 950. We find™ (with ( . p.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.1.l .71) The integral can be looked up in Tables of Integrals. M.17)). Gradshteyn and I. 369 and 8.807 m s . dx sin ' x •• Looking up the value of the factorial in Tables.1 / 2 ) ! = v^r) TT/2 /. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12.264 we have ro / dxyfV(x) Ja CHAPTER 12.9191 ~ 1.055 X 1 0 _ 3 4 J s .384. one evaluates an approximate probability of exp[6. we obtain ( .
we consider the linear potential in three space dimensions that forbids the particle to escape. in the book of G. 265 . Thus.* In Chapter 14. whereas the first case permits only a continuous spectrum. we consider the corresponding case of a particle above the flat surface of the Earth. This is the problem of Galilei in quantum mechanics. in Chapter 15. each of which propagates with the classical momentum and energy of a Galileian particle. p. This is the quantum mechanical version of the problem of Galilei in one space dimension. * A highly abridged treatment of this problem is given.2 The Freely Falling Particle: Quantization Under quantization of the freely falling particle we understand here the solution of the Schrodinger equation for the linear potential. for instance. 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.3. 144.1 Introductory Remarks We distinguish here between three different types of linear potentials. 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. Finally. 13. Then considering the probability distribution determined by the squared modulus of the wave function. Siissmann [266]. Example 14. the last case allows only a discrete spectrum. An additional reason to consider the linear potential at this stage is its appearance in the following chapter in connection with the matching of WKB solutions across a turning point.Chapter 13 Linear Potentials 13. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.
t) = mogx + V0.t) = m0g with potential V(x.2) which is the case with a discrete spectrum.e. g > 0. The present case is an exception and a good example of its applicability.4b) .3) In general the momentum representation is of little importance.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. V(x) = —mogx. We see " that the case g < 0 can be related to the reflection x — —x. and yields rriQX = mog. We consider here the case of Eq. 13. i. (13. (13.266 CHAPTER 13. The equation of motion follows for instance from the derivative of the constant energy. A different > "linear potential' is the socalled "confinement potentiaF V(x.= l V 27T J_oo 1 dkeikxijj{k.t).t) = c\x\ + V0. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x. (13. Recall the classical treatment of the freely falling particle. c> 0. (13.t).t). (13.4a) f°° ${k. from d_ m x2 0 dx + V(x) 0.1) where g is the acceleration due to gravity and we can put Vo = 0.= l V 27T Joo dxelkxip(x. (13.t) = .2. x > 0. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle.1). Therefore we use the Fourier transforms il){x.t) = .
(13. (13.13.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.9) . We rewrite this equation in the following form th (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k. multiplication by a test function and subsequent integration).7) applied to every quantity in Eq. To this end we consider the following substitution in Eq.< (13. (13.t). The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.t) dketkxiP(k.t) h2k2 d ~J(k.t).4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.6).6) the latter equation becomes d ~( m0gt ^(^M<=+=?. (13. i.6) and wish to solve it.d m +imo9 .6) the argument k is then replaced by this.8) With this result and the substitution (13.e.e.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.3) yields us now by inserting for ip(x. (13. we obtain the differential operator on the left of Eq.t) m0gx /2TT m0gxip(x. Equation (13. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt. d \ ~f1 dkr[ k + m0gt ^r't (13. Applying the total derivative d/dt to this function.t) dk nikx k=oo '2K (im0g)il)(k.7) In the wave function on the left hand side of Eq.t) the expression (13. • fd ~(' m0gt %h .t)im0g^k.
the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.268 CHAPTER 13. Thus addition of both indeed leaves the expression on the right hand side of Eq. We observe t h a t t h e solution is actually a function of kt — k — mogt/h. i.= k rmt.10) k by m0gt back : k —> k h The solution of Eq.6).6).6). (13. the following result h 2moi tp[k\ ^—.12) . t ) .t ) ' U ( f c .11) solves the partial differential equation (13. Our next step is to reexpress the solution (13. On the other hand the second operator expression implies dtj>o imog^—ip(k. To obtain the solution of the original equation (13.11) in a neater form. Solution: We apply the two operator expressions on the left of Eq. We have thus obtained the solution of the transformed equation.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc. we have to reverse the substitutions and replace in (13. (13.11). (13.11) *[k?f. (13.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. We define the wave number kt as kt.i) . (13.1 we verify t h a t this solution indeed solves Eq.6) is then $(k. Thus we have first (the second term arising from the upper integration limit.e.gh [ d t ' j f c + ^ ( t ' .t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13. E x a m p l e 13.t).t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.6) to the solution (13.6).t In Example 13. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.10) Note the amplitude function ipo(k) in front.
t) into this integral and recalling the property (13. _ k t  h2 3 Inserting here the explicit expression for fct.— [It(k) .15) we obtain / = .o ut = 2mo —kf 2mo A.18) h =k mogt p h S' h .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.17) = J_/ C X ) h2 [It(k) .19a) . . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.11). we obtain rp(x.t) = .2„2+2 mfaH 3^3+3 . we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * . we have ip(x.} fA ' _ _tkt 2 k+ t) 2 k .I0(k)]. 27T ^ .13.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13. (13.13) rn0_t+2 + mlg2t3 . Inserting ip(k..t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With .14) (13. m0g (13.16) Next we evaluate the Fourier transform of ip(k. (13.4a). .+^f (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.gfci + 2 g2m0 2/i 2 ' (13.o o 2 —  exp iktx h mZgH' .t) defined by Eq.11) can be written j'A ^.
270 and CHAPTER 13.ut> = Jo t2 g2m0 i 3 k t . 7 u>t'dt' (13. (13.23) . (13. 13.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.t)\2 for t — oo (distant future or past).t) as a decomposition into de Broglie waves. To this end we convert Eq. 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 . The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.20) (13.o o Jo This result can be interpreted as follows.19a) exactly as for a falling particle. Linear Potentials f dt'.2.22) 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 .19b) h3 3 we can rewrite ip(x.21) In other words.v = kh/mo. Such a decomposition is possible only for homogeneous force fields. These wave vectors vary in accordance with Eq.18) into a • different form in order to be able to perform the integration with respect to k. (13.t) =y= dkip0(kt)) exp i< fctx V27T J .t) as 1 I00 ip(x.20) expresses the wave function tp(x. fm0gt\ \ 2h ) (13. the relation (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 associated kinetic energy of the particle is mv+ = huJf.
ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f.29) . m0xcl ht m0gt 2h gt O v: °° 2hti (13. rnogt m0g2t3 tX^r~. We evaluate the Fresnel integral in Eq.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.2 The Freely Falling Particle: Quantization The integral (13. (13. integrate and then take the limit fi —> 0). 2h h 24ift ' mogt^ r m.26) d£e ie £ kt^ m0' 0.13.t) = 1 1 oo 271 27T /. Then ip(x. TOO#A ft and.18) thus becomes ip(x./m 0 a.28) (for a verification replace a by a + ifi. (13.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.t) 1 exp r .e. (13. fi > 0.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ . /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. being now independent of £.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. It follows that ip(x.24) We now set fit / 2m 0 V" 0 V m0a.QX 2 ^ 0 7 (m.26) with the formula 1/2 (13./oo dkip0(kt) exp l x ixI exp oo . may be put in front of the integral.
30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i. These are incoming and outgoing waves.t) ~ W —exp i V irlt m ktxd J cot.1) has only a continuous spectrum. For stationary states we set xl){x.31) and obtain from Eq. kt and ut 13. (13.33) oo for E <C \mogx\.dA 4>o(h (13.1 A turning point at x = —XQ.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. 13. .e. (13.t) = eiEtlhct>{x) (13. For x exponentially increasing and decreasing solutions are possible.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0.272 CHAPTER 13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . V(x)=m 0 gx V=E Fig. Linear Potentials Replacing here x by the large time value x c i. this expression can be brought into the following form: if *\ 0 ip{x. xc\). 1/2  ±^(2mlg/h^x^ / 2i (13.
exponentially decreasing. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. However quantum mechanically this is possible with a small. The point — XQ. t). 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.35) It follows that /W tm0g \ 2m0 i. at which the classical particle bounces back.t).e.e.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. i. t) = Eil)[x. but also d 00 dt Eijj(k. i. and we can establish for the resulting continuum solutions the orthonormality and completeness relations.e. at which E — V = 0. J \ „ (13. Akx (13.e.1. and hence is continuous and extends from —00 to +00. which is not possible for real particles.36) m0g \ 6m0 with c = const.00]. we .4b).e. probability. so that ih dtp{k. i. E)dk. cf. is reflected. t) = — = / V27T J . 13. (13. 1 f01 dketlcx^{k.e.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.t). H2k3 4>{k) — cexp Ek (13. The Fourier transform is given by the relations (13. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00.t) = dt (13. its momentum p would be imaginary there.34a) Then d &k 1 f°° ih—il>(x.13. i.0 0 dkelkxE^(k.4a) and (13. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. i. Fig.6) that also for 4>{k.34b) It therefore follows from Eq.
/ <>~ikx VZ7T J (13. (13. (13.e. 13.M *K Joo Setting dt cos ( ts n Jo st).36) (with c replaced by (13.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. i.43) y=2^x+2^E.36) into (13. </>E(x) = .L dxeikx(f>E(x).e. (1340) i. In a similar way we can verify the validity of the completeness relation. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').39) dk4>*E{k)4>E.40)) into (13..£') = <*(#£')• 1 27rm0g' = /o . Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). h? h2 .a : ' ) . (13.42) The Airy This integral can be rewritten in terms of the Airy function function Ai{s) can be defined by the following integral: 1 f°° Ai(s) = — / eft exp i ( st . apart from a constant phase which we choose to be zero.41) J—oo Next we insert (13. V 2lT J Mk) = .37) (eikxdk. / dE(j)E(x)(t>E(x') = ^ s .38). so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .38) we have for <j)E{k): OO /"OO / oo (13..L [ dkeikxMk).
whereas the other is of exponential type.44) is then expressed in terms of the Airy function: . so t h a t the integral becomes 4>E{X) = 2vr ^/mQg / n oo dtexp oo h? * 2m£g y \ 7 5 * " 3 M' Choosing /2mfo\1/3 A= ^ . (13./ 2m 0 E (13. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential.. .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. the function 4>E{X) i and setting s y = — = 1 .^x + ^ E (13.45) One can now derive normalization and completeness integrals for the Airy functions. / s M .J * .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. and we shall see t h a t one branch has periodic behaviour. (13. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.46) This follows since with Eq.^ .
i. (13. We want the integral to exist and so desire that lim e**^ = 0.r 3 sin 30 ).. i. or n <6 <TT o for r > oo.f dzJ*W 2TT JC with z = reie. (13.\zz.2 The contour C for s < 0 with ends in the angular domains (a) and (b).e.TT < 30 < 0.4 The Saddle Point or Stationary Phase Method The integrand of Eq.e.. (13.TT < 6 < 0 for r > oo.43) contains the phase (with x — z): > *(*) := sz .276 CHAPTER 13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 .51b) . .51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0. 13. o (13. .r 3 cos 30 J + i ( sr sin 9 . (13. Linear Potentials 13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.49) (13. Thus we must have sin30<O. (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .
51a) and (13.2.4 The Method of Stationary Phase 277 The integral (13. —i ${zs) = szs . so that exp(±2i9s) — 1 and — s — rl. Then the ends of the contour lie in the domains specified by Eqs.55) s < 0. and the main contribution to the integral comes from the a region around the saddle point. (13.57) $"(z s ) = 2% s.58) .( \ / ^ ) 3 2z.53) Since we shall have the contour C in the lower half plane.55) into (13. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.54) i ¥~s)3/2' (13. we choose zs = iy/^s.e. provided its ends at infinity satisfy these conditions. (13.56) 2n(s)^Jc But I daefr"?*"^.e.z3s and V^s + .S )V2 (13. and we set with a real: z := zs + a (S)V4' (13. 0 = &(z8) = s~zt s = za „2„2i6>. Then 6S = ±7r/2. 13. (13. i. oo] as indicated in Fig. Such a piece can be found in the neighbourhood of the socalled saddle point zs. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.45) exists for the contour C.51b). Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.48).13. at which the phase becomes stationary. where the derivative of the phase function vanishes.54) where a is the new variable with —oo < a < oo. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. i. Inserting (13. (13. (13.52) = exp(±i7r) Let s be real and s < 0. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.
44) we can convert Eq. V 2h2 (13. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. According to Eqs.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z).42) into an equation with a more appealing form.45) 4>E(X) ~ Ai(z). For (—s) — oo the integral converges towards T/TC.32) with the solution (13. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). This is an important aspect in the WKB method to be considered in Chapter 14. (13. Ryzhik [122].278 CHAPTER 13. Sec..44) and (13.2/32u = 0 (13.491. M.59) In Example 13.64) tSee I. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. S.s ) V 4 exp \(s) 3/2 (13. as one can see from the following equation^ ri' + Ffz •2^2.2 we show that for s — oo: > Ai(s) 1 .32) therefore becomes dz2 + zd) = 0. With the help of the substitutions (13. 8. (13. Gradshteyn and I. (13.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. cos . 2 s 3/2 . (13. V^s / 4 V3 (13.56) decreases exponentially around a — 0 for real values of a. > so that Ai(s) 2 v ^ ( . (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. .60) The reason for describing the point zs as a saddle point is that the integral in Eq._ 1 .
Choosing the contour C to lie along the real axis. Example 13. cos  .s 3 / 2  . It follows that for s —> oo: Ai(s) ~ . Thus there are two symmetrically placed real saddle points on the real axis.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s). P Introducing a new variable a by setting p = ai\a\. changing to the variable p.59) and (13.z ± ) 2 * " ( z ± ) .51a) and (13. 27T Ic Jc 3 This time s is real and positive. we obtain f° Joo dp=—[a+ i\a\]da. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. the conditions (13.13. 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 .2: Periodic asymptotic behaviour of the Airy function Obtain with the saddle point method the asymptotic larges behaviour of the Airy function Solution: We have (cf.. (13.60) 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. Then. with *"(z) = 2z. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .51b) are satisfied. * " ( z ± ) = =F2Vi._ .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13.^ e ^ / V2 4 [°° Jo dae'2^. Eq. .
since f'(zo) = 0 and (z — ZQ)2 = —x2. .111).dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). ^/2KJW1. n!= / e~zzndz= / e~z+nlnzdz. z o = n .3: Derivation of the Stirling approximation of a factorial Obtain with the saddle point method from the integral defining the gamma or factorial function the Stirling approximation. i. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. 13. (11. Thus here / ( z ) = — z + n l n z .280 CHAPTER 13. Solution: Briefly. Then. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix.*.3 The saddle point at z = n + ix. 13.e.Be 2 Fig. ~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq. and hence integrating as described above through the saddle point indicated in Fig. Jo Jo \ . / " ( z o ) = —n/z2 = —1/n.3. Linear Potentials E x a m p l e 13.
and the role played by h in our considerations here. B. the linkage of solutions above a turning point to those below. Kramers and Brillouin. 292 .317.* The method had. named after its first promulgators in quantum mechanics: Wentzel. Wentzel [282]. The WKB result is therefore frequently described as the semiclassical approximation. for misunderstandings pp. ' T h e most prominent earlier expounder is H. Dingle [70]. *R. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. A critical overview of the history of the method. will there be that of a coupling parameter.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. Jeffreys [144]. 316 . A.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. been developed previously by others in different contexts.e. § See 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. For the history see pp. however. where *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. It goes without saying that such comparisons are very instructive. i. H. 281 . Froman [99].g. of misunderstandings and other aspects may be found in the monograph of Dingle.e. ^ The central problem of the method is the topic of "connection formulas". Brillouin [37].Chapter 14 Classical Limit and W K B Method 14. Kramers [153] and L. O.295.
the function S being called phase function. the energy has a definite value E with V(r. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. for which the observation of position and momentum is independent of the time at which the observation takes place. For stationary states ^/. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T). so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14. 14.7. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e.282 CHAPTER 14.2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0.t) = eiEt/h7P(r) and \V(r.* We illustrate the use of this condition by application to several examples. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.t)\2 = \iP(r)\2.3) . In this equation we set. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . A(r) and S(r) being real functions of r. Classical Limit and WKB Method E — V changes sign). V>(r) = A(r)eiSWn. the derivation in Example 18. (14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh.V(r)]^(r) = 0.1) (14.g.
(14.13) .11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.t)}2. (14.V(Y) + and 2VA • VS + AAS = 0.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14. These equations are equivalent to the Schrodinger equation.1) and separating real and imaginary parts.6) 2mo dt now with A also a function of t. we obtain the following two equations: E . V(r.* VAVS+AAS + V(r)A.14. We can identify Eq.t)eiS{r't)/h. (14. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r. (14.10) m 0 ^ .+ VAVS + AAS = 0.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.Q A 2i 8A 1 (14. We set in this equation. (14.12) VA A (14. h h (14. h2 ih— = H$ A + V(r) * .10) with a continuity equation by defining as probability density p:=*** = [A(r.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.4) (14.
(14.20) With this we can rewrite Eq. (14.10). ie. Since J = — A2VS = PVS.10) we obtain = — [V(A2) • VS + A2AS}.9) for the phase function S.15) and (14.14) A2VS. (14. Since we obtained this equation from Eq. (14.284 CHAPTER 14.n.16) this becomes m 0 1 rr + m o V . . ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. the implications of this equation are also those of the equation of continuity. (14. Next we investigate the significance of Eq. (14.18) ot ot This is the equation of continuity that we encountered already earlier. ^ + V • J = 0. + h2 AA V = — ^ .J = 0.16) BA 1 moA—. (14. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. mo <™> (14. m0 But now (14. (14.17) With Eqs.. TTIQ (14.9) as 9S 1 mo^ 2 + T. (14.21) . %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.01.+ AVA • VS + A2AS = 0. we have mo—(A2) + V(A2)VS + A2AS = 0.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS.
(14. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. This expression with h in the denominator shows. This means. The particles have no interaction with each other.e. r o t v = ( l / m o ) r o t g r a d S = 0. i. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. V • J = 0. 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. with the help of Eq.V ) v + V 7 = 0.14. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. . Such series are typical for expansions in the neighbourhood of an essential singularity. (14. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). a singularity different from that of a pole). This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. where.1/h. that the quantum theory implies an essential singularity at H = 0 (i. they are asymptotic expansions.2) for the probability amplitude. we obtain ^ V 5 +  v ( v v ) + V V = 0. However. This result is very general. which for very small values of h yield the dominant contributions.23) (14.24) d 1 dr d •— = d Lv _ .22) (14. so that V ( v • v) = 2(v • V ) v + 2v x rotv. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.20). > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0.e. In the first • place we require the ansatz (14. and hence''' ^(m 0 v) + m 0 (v. d — = (14. The equation of continuity then describes the stationary flow of a fluid. the wave function \I/ describes effectively a fluid of particles of mass mo. We deduced this equation solely from the phase S in the limit h — 0. dt 2 Constructing the gradient of this equation.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv.
the method is widely familar under the abbreviations of the names of Wentzel. or h2/2mo(E — V). The expansion in powers of h2. (14.[EV(x)}y = 0. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large. . such as "LiouvilleGreen method' and "phase integral method'. Messiah [195].286 CHAPTER 14. As remarked at the beginning.4) and (14.$ The basic idea of the WKB method is to use the series expansion in ascending powers of h and then to neglect contributions of higher order. (14. 6.§ y = A(x)eiS^'h where W(x) = S(x) + . Classical Limit and WKB Method 14.3. similar to our earlier procedure with A and S real functions. i. z (14. We consider the onedimensional timeindependent Schrodinger equation y" + ^.26) We observe already here that an expansion in rising powers of k.3 14. There are also other more specific descriptions. i = eiW(x^h.27) (14. Dingle [70].30) ax *For details see R. B.In A(x). h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V). p.28) The calculations we performed at the beginning imply the following equations (compare with Eqs.2. 318.e.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. (14. is an asymptotic expansion. §We follow here partially A. We set.5)) and 2 ^ f dx ax + All=0. Kramers and Brillouin. Sec.
10)).V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• . + (14.28).A = const. Since the expression (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 . .V) and (14.35) can be integrated.36) Equation (14.2mQ(E . Eq. (14. 3 / 5 .2rn0(E V) = h2 lS'A 2S'0 (14.14.33) 3 (S")2 . But first we note that: A dA dx2 2 = c{S')ll\ = A 15^ 2 S' + ^ = A ax 3 (S") 4 (S')2 2 2 S' With this Eq. (14.35) .32) is therefore (S'0)2 + 2H2S'QS[ . r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .29) becomes (note that this is still exact!) (S')2 . (14.4 (S>)2 1 S'" 2 S> (14.31) \a.3 The WKB Method 287 Equation (14.30) corresponds to the equation of continuity (i. Li This expression can be substituted into Eq. (14.32) s" = sz + h2s'( + •••.\ In S'. (14.e.
and instead is asymptotic which we shall not establish in detail here. (14.39) where a and /3 are constants. p = h/X) k{x) :Then S'n +.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .288 CHAPTER 14. (14. E > V(x).27) we obtain now with Eq. 1 S e e R. Eq.and So = ±h / — — A J k(x (14. Dingle [70]. y/2mo(V(x) .39) and (14.41)? We note first that the expansion in powers of h2 does not converge.37) For Eq.41) We now ask: What are the conditions of validity of Eqs. B. . we expect that in any case So » ft2Si. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). Chapter XIII. i.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). Classical Limit and WKB Method has the form of a momentum. (1. The solution is seen to be periodic provided k(x) is real.' For the expansion (14. (14.33) to make sense as an asymptotic series.40) rx dx v "7 W)\.e.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . (14.38) h y/2mo(EV)' (14.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14.19). (14. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. it is reasonable to set (cf.
43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.36).V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.45a) . 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. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .14. E > V(x). Eq. Since k(xy 1/52\S'0 (14.1 / 2 ]" [§0SS). (14.3/2 s&T + 3 {So Il« we have (cf.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.42) [(^).3 The WKB Method We have 289 S0 = ±h and Si follows from (14.
46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. In a similar way we have for E < V(x): l'(x) <C 1.290 CHAPTER 14. at which E — V{x) changes sign? We note first that in this case near x = a: E .3.V{x) ~ Or .48) behave in the neighbourhood of such a point called "turning point'. (14. 14.V(x) = V(a) . In terms of X the original Schrodinger equation is y" + = y = 0.48) The (dominant) WKB approximations are exact solutions of this equation. V(x)) (14. (14. *2 A(z) = V / 2m 0 (£ . around the classical turning point.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. (14.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l .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). How do the solutions of Eq.e. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. Roughly speaking one therefore requires E to be large in both cases. Classical Limit and WKB Method we can rewrite the inequality (14. and hence (with UIQ = 1/2. the WKB approximation is in general invalid (see below). In the neighbourhood of E = V(x). i.a)V'{a).47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. h = 1) .
3 after stating them first here.1.1 The regions around the turning point at x = a.48).48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. On one side. 14. We assume we have a turning point as indicated in Fig.14.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. since this requires different approximations there. 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.e.3.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. which connect one domain with the other.1. 14. (14. 14. The transition is provided by matching relations. in the domain E < V the WKB solutions are exponentially increasing or decreasing. i. in the domain E > V the WKB solutions are oscillatory. As indicated in Fig.1 there are WKB solutions on either side of the turning point and some distance away from it. (14. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . and on the other side. 14. 14. V=E Fig. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. We shall not derive these conditions here in detail but rather make them plausible in Sec.
H— J =^ v/exp + m (14. Note also the extra factor 2 in yi. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. in order to make the formulas independent of whether the potential is rising or falling.52) For a potential rising from left to right. 14. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. has the same asymptotic behaviour in the domain i « a a s Ay\. This means the asymptotic part suffices only in t h e exponentially decreasing case. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. Thus.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 / —. one could simply replace everywhere /•a / pa • • • b y / Jx J x . a interchanged.51) It can be shown t h a t the second W K B matching condition in the opposite direction (and for the potential rising from right to left as in Fig.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. 14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . as in Fig. (14.292 CHAPTER 14.50) x S> a.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . (14. however with the limits of integration x.
Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.a). 2 1 X (14.51) — always with the decreasing exponential on the left — is valid in both directions. is singular at a turning point x = a. (14.14. y" + Toy = o. the matching relations have to be altered accordingly. whose exact solutions are the dominant WKB approximations.48). the equation becomes approximately y" = (xa)xiy. Therefore we consider now this original equation in the domain around x = a. i.54) . (14.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.3 The WKB Method 293 In that case Eq. 14. If higher order contributions are to be taken into account. One should remember that the matching relations above are those for the dominant terms in the WKB expansion.Airy overlap* Fig. as we saw.e. (14. 14. 1 ~ (x .2 The overlap regions around the turning point at x = o.53) Since at x ~ a.46). the equation V approximately linear around turning point V=E Airy solution WKB .3.
. > for a. (14. (14. formulas 10. (13. (14. — a ^> 0. A second solution is written Bi.60).59). (14.291. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. Stegun [1]. (13. R. 448.4. In the following we require the asymptotic behaviour of the solutions Ai(—z). Cf.57b) UV2 3 11 by J dx .a)3/2 ^ /2 =^ 3/2 o Q.^ 4 I 7T x — a <C 0.294 CHAPTER 14. A. B. as > may be seen by referring back to Eqs. In particular we encountered a solution written Ai. J (X (i4 58)  Replacing in Eqs.57b) To the same degree of approximation as the equation y" = (x — a)xiy.55) the equation becomes the Airy differential equation d2y dz2 zy.1 / 4 by \^.57a) and (14. (14.Bi(—z) for \z\ — oo. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. With the substitution z = (x . we have r x * f x ' /2(x a)i/2d*=^(x .5960.a)X\/3.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. i I 772 \3 + > for ia>0. This is given by the following expressions. p. Dingle [70].56) We have discussed some aspects of solutions of this equation in Chapter 13. or M. Abramowitz and I. p.
F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. and hence must be proportional (in view of the uniqueness of the solution there).l. These solutions.l v/2m0(EV(x))dx 7r = (2n + l). n = 0.2. 14. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). by Ai(z).1 to the quartic oscillator.(zf2 by J we obtain the matching relationsX (14.3 The anharmonic potential well. In this way different asymptotic branches of one and the same solution are matched across a turning point. however.51) and (14.. 14. are only valid in a small interval around the turning point at x = a as indicated in Fig.52). (14. As an illustration of the use of the WKB solutions we apply these in Example 14.14. As we approach a limit of this interval in going away from the turning point.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 / — = .2. In the direction to the left as indicated in Fig. we enter the domain of validity of a WKB solution. We summarize what we have achieved. In this sense we have now verified these relations.3 The WKB Method and 295 •. e. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. 14.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution. Example 14. The latter domain is a small region where the solutions from either direction overlap.g.59a) .
b). (14.60) Using the matching relation (14. Classical Limit and WKB Method where x = a. 14. . extends only over a length of very few wavelengths and is therefore valid for n small. We argued earlier that the WKB method is suitable in the case of large values of E. 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. Let E = V(x) at x = a and x = b.l. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . regions 1 and 3 in Fig. /n = s u ^ TJcos^ya TJ+cos^a T J . (14. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. The condition for this is ya{x)=yb{x) for xe(a. 2 n = 0. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. According to Eq.51).. . Sometimes this distinction becomes imprecise. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. Then around these points the potential would be well approximated by a linear potential (as we saw above).^ .2. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.(x) = c ' v Acos I / — I for a < x < b.62) The condition for this to be exactly satisfied is — as we show now — that the Bohr—Sommerfeld— Wilson quantization condition holds and c' = ( — l ) n c . 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 ( .y o dx TT T provided 6 _ 4 /. Hence we search for solutions which are exponentially decreasing in the far regions.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). 14. i. ^ A = (2n + l ) . . 1 .. n = 0 . whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. . j/i.3. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. (14. 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. Here we are interested in the discrete states (the only ones here).59b) Solution: We consider an anharmonic potential well as depicted in Fig. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. . (14.44) this implies large values of n.3.296 CHAPTER 14. (14.e. as in the case of the harmonic oscillator itself. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. b are the two turning points. 2 .
with n — 0. 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.x number of turning points h.x number of turning points TTH. The corrected form* is. . This relation is sometimes wrong. Sommerfeld and W.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. . equivalently. ri2 I Jqi :pdq n + 1 — .1.e.4 BohrSommerfeldWilson Condition 297 14. A. pdq = n + 1 — . .14. and knows of no published form or script. Wilson by supplementing classical mechanics by Planck's discretization.63a) or. . Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. (14. B.63b) *The author learned this in lectures (1956) of R. In this socalled "old quantum theortf (i. Bohr. (14.64) (14. n = 0.1. (c) V Fig.4 Three cases with different pairs of zeros. Dingle. 2 . 2. . 14.
the case of tp(q) — 0 at q = gi. together with q — — oo. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. There are three cases.e. The WKB Method One pair of solutions is (as we know from the earlier sections of this chapter) tpT(q) oc 7T7Iexp 1 sin (_J)l/4 Cos 1 = JdqVJiq) F if f(l) is positive. / • n an integer > 0.4(b). this is an exponential zero. Before we continue we recall from Eq. i. where /(g) is positive.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. 0(g) ex . In this > case the wave function for g > go is . where /(g) is negative. if /(g) is negative. In quantum mechanics we have dq2 + Ki/j(q) = 0.4(a).e. Jqi ^ fpdq J =2 pdq = (n + l)h. We have the case of trigonometric solutions.e. i. this is a trigonometrical zero. this means dqyf—f = (n + l)7r.65) We use the knowledge of these solutions to find the eigenvalues of the equation. (14. Case (a): The case of two trigonometrical zeros. (14.298 CHAPTER 14. i. 14.66) J<1\ Thus in this case the socalled old quantum theory is correct. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. 14. i.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. where n —. Since sine and cosine have zeros spaced at intervals of TT.67) Case (b): Next we consider. (14. as illustrated in Fig.
4(c). (14. and • • for the wave function to be exponentially decreasing to the right of q'0. 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 .70) We can therefore summarize the results in the form of Eq.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). the sines must be proportional. (14.63a) or (14.63b). this is the case of two turning points. Hence we have pdq= i n +  . Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. i. Solution: As in the case of the quartic potential.3.69) pdq = (n + l ) " 4 h. 14. Since both cases refer to the same region qo < q < q'0.68) (14. E x a m p l e 14.e. Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. As illustrations we consider Examples 14. Thus in this case the socalled old quantum theory is wrong.14. i.2 and 14. (14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. or Pdq (n + l ) " 2 h.63b) to obtain the eigenenergies of the quantized harmonic oscillator.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.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . For the wave function to be exponentially decreasing to the left of qo and using Eq.2: The harmonic oscillator Use the relation (14.e. Therefore their arguments differ only by (n + l)ir. ri\ / J do y/f(q)dq (n + 1) 1" (14.
5 A particle above a flat Earth. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. Thus there is one turning point at E = mogq.71) For m 0 = l x 1 0 _ 3 k g .63 x 10~ 3 4 J s . and h = 6.2 3 J x (n + 3 / 4 ) 2 / 3 .807 m s . permit this although with rapidly diminishing probability.5. in principle.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. Example 14. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. 4 / x CHAPTER 14. whereas a turning point does.28 x 1 0 . one obtains E ~ 2. as illustrated in Fig. where g is the acceleration due to gravity. Calculate its quantized energy.300 With potential V(q) = 2ix2mov2q2. so that E = hu{n + 1/2). 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+ • (14.2 .V(q))dq The result is E/u. g = 9. . 14. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . 14. :: V=m0gq Fig.
h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .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 ( — = .72) \\2dxp=Qdx^{Eh Hence.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.75) which verifies the formula immediately.l.14.7.73) Jx! rp2a dx I^<E 2 v) = ^ n — = . 9 .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 . and u> is the oscillator frequency in either well. with p = modx/dt.( 1 — = — = it — = h\ n H — 2TT OJ 2K J p dE V 2 In the case of the harmonic oscillator we have hi n E (14. Chapter 16) one obtains the quantization relation (for Mo = const. (14. E x a m p l e 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 £ = .5 Further Examples E x a m p l e 14. Solution: The proof is contained in Example 18. In the case of screened Coulomb potentials (cf.2. 1 . ra 2/ = 0.f V ft h JXl h 2TT (14. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). d dE V)={n+)n. (14.5 Further Examples 301 14.= — <*— = h(n+\.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.
y2)(y . It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.6 The cubic potential. 14. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. 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 . q = y+ a .~ \ (14. The q term in this equation can be removed by transforming the equation to a cubic in y.e. (14. (11. Eq.6. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.133)).6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). which determine the imaginary part of the energy. Fig. Example 14.y3) = 0. where .a .302 and T CHAPTER 14. i.77) Determine (a) the turning points.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. 14. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.78) (a) We determine first the three turning points at qi = <?i. (14.<?2i<?3 for E > 0 given by E — V = 0.79) . where 1 2 • 1 2 y = Q.
gi ~ y {V3e} < 0. _ \ / 3 — tan(9 2 V92 . M i l n e . q3 = a2[\ .e.9 2 [ .6. i.g i ) ( g 2 . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . 78: sin 120° = v ^ A c o s 120° = .120°)] = .93 2a2 [ . 92 = a2[l . 0<6O°.= s i n t 3 v 3 (14. H e r e cos(<9 ± 120°) = .g i < 1.\tan0^ v/3.0^6O°. 91 < 93 < 9 < 92(14.cos 0 + cos(6> + 120°)] = a2 92 . B .T h o m s o n [196]. 14.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . V cose' + sin 8 (14.14.c o s ( 0 + 12O°) + c o s ( 0 .a 2 [ l .< 1. e > 0. we h a v e qx = .80) I n s e r t i n g t h e r o o t s yi. V3 + t a n 0 2 . (14. D . M .2/3 i n t o q = y + a 2 / 3 .85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0.5 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. B y r d a n d M . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . 37. 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.9 l ) f f / du s n 2 u en2 d n 2 u .08.83) 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 . 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 . (l_fc' 2 a + fe'4)l/4. 92. 6 = 0 i m p l i e s E m a x of E q .1 / 2 . (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. 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 .04. q3 ~  .3/2. f o r m u l a s 236.k 2 2tan# \/3 + tan! .2 c o s 0 ] . p .93).c o s O + cos(6» .' : ± 6 0 ° w h e n E = 0. p . 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. g 2 ~y{3}>0. 80 a n d 361.9 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 .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.{ V 3 e } > 0.q)(q . F r i e d m a n [40]. (14.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 .79) a n d t h e n s e t t i n g m o = 1.2 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.2 c o s ( 0 + 1 2 0 ° ) ] . p . 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. D w i g h t [81]. § T o h e l p we cite H. F .9 3 — = = 2 " 3 " [ . tan60°=%/3.84) 92 . ' P . (14.81) m a 91 . (14. 93.ui=K(k).78)) c o s 30 •• 4P3 ( ' \ aeb2} .gi = 1. 212. p . . we set 0 = 60° — e. ui=sn 1 \l— — = l. a n d t h e a n g l e 0 = 0° t o E = £ m a x .
91 . as in (14. (14. A:') = # ( * ' ) • (14. 0 (14. sinV = q3 qi ~ = 1. Whereas the integrand of the above barrier integral is effectively a momentum.fc'2 + fe'4)"1/4. E(u) the incomplete elliptic integral of the second kind.^ + ^ ) (l + fc'2)2 . We can evaluate this integral again with the use of Tables of Integrals. V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .S . V?2 . (14. g= 2 . G(fc) =86aS (ITS£ li[fe. D.sm t %/3 £\.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. V 93 .k). Thus Iwen=lgF(ip.2(fc2 .91 F(TT/2.= —^ = k a2[cos0+^] v ^ + tanfl V=". the integrand of the integral across the well is effectively the inverse of this momentum. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. .91) a2[cos0^] y/3tanfl ? L .304 CHAPTER 14.86) 15 tan. with K(k = 1) = oo.^ V3 / V V3 y a / c o s e J. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .84) we obtain Carrier = 2a 2 sin6> 2 b( .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . 2 2 . (14. 1+72 = and J o s 0 + ^ = (1 . (24. .87) Inserting G(fc) together with the expressions for the prefactors into (14.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .92) ''P. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .85). formula 233. E(u = 0) = 0.32)). Next we evaluate the required integral across the well from 91 to 53 at energy E. Eq. p./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. 72. The WKB Method elliptic integral of the first kind.qz) (14. F.89) This expression will again be obtained later with the help of configurations called "bounces" (cf. 91 < q < q3 < 92 bJc.00. E(u = K(k)) = E(k).~ y ^ ) a< [ COS 5 + .?)(?3 . Friedman [40].
to represent a physical decay rate.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. 2a 5 b K(k) + 2 E(k).99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. pp. we obtain 15 .14. and hence at E = 0. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1.85) (14.** However. a°b n = 0. (14.63b). (14.88) in rising powers of fc' . 905. we obtain E a6b2k'4(l + k'2). S.3. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. so that even in the case of the ground state the zero point energy will contribute to the prefactor.4 2n + 1 fc' = 4 — ^ — . (14. Gradshteyn and I. 32 (14.113.96) Expanding the coefficients of the elliptic integrals in Eq.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . 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. M.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . .5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (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. we obtain > .80). Ryzhik [122]. (14. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. 906.97) 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. With q = 1/z the integral (14.2.114.l. Thus one obtains from (14.3 and 8. formulas 8. (14..93) where w is the harmonic oscillator frequency in the well.
103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. (20. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.73)..J + 0(VoJb). 15 a . this has not yet been done.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results. _ T T • (14. (14.96) for the imaginary part of the energy as in the Breit.. so that we obtain 4 .100) 'barrier = ~0°b In [ ^ — j . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .ba / 25a5b\n+2 ±l_l eT5 a f >.Wigner formula (10.„. also Eq.— (14. /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .e x p [ .102) . The WKB Method For k' —• 0 the first and the third terms dominate. the result agrees with the ground state path integral result (24.e. 8 Bfc ..101) (14. It follows that (for one complete orbit back to the original turning point) .> tf = = 2/wfiii = ±tv8a°b——P./^~^l^ab ^aub .fc'2 + fc'4)1/4 * ± t ^ . .178)) call the Furry factor set equal to one.. 7 ~ —.e. _ exp[2/barrier] >.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. i.7 .j .306 CHAPTER 14. 2n + 1 — / 26a5b \ . We observe that with \j2je as 1. i. n ~ V27T . V (14. (18.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . .2 7 b a r r i e r ] Z Z7T where (cf. (0) i hw En = E\> .
K. along with inclusion of angular momentum and spin effects and their interactions.* 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. K. The study of this potential. the Coulomb potential. in fact.e. There are numerous. Lane and T. T. that this is indeed the case. 307 . that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. This potential became widely popular in the spectroscopy of elementary particles. Rosner [230]. describing free fall under gravity. Quigg and J. Yan [82]. Gottfried. [233] and H. Kinoshita. [232].M.1 Introductory Remarks The particular linear potential we considered in Chapter 13. L. Rosner [269]. [231]. i. L. B. Thacker. indications from the nonabelian generalization of quantized Maxwell theory.g. See particularly E. at least indirect.Chapter 15 Power Potentials 15. after the realization that quarks as their constituents might not exist as free particles and. now known as quantum chromodynamics. C. D. led to a spectrum which contains only scattering states. which permits only bound states. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. 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. Eichten. C. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. Quigg and J. the classification of the various states of nucleons and mesons and other particles.
u ~ r i + 1 (/ + l)R(r). m\m<i M= : .1. where \x is the reduced mass of the two particles of masses mi. These particles therefore cannot be separated by any finite amount of energy. </>)R(r).V 7 = const. (15. and the normalization [ dr\y(r)\2 = 1. and u'(0) > [u(r)/r]0.4) Instead of considering the linear potential.3) with the boundary conditions''' u(0) = 0.2 The force The Power Potential F = . Power Potentials general power potentials. A > 0. v > 1.2) tt(r) = Ylm{9. This force maintains this value irrespective of how far apart the particles are separated. (15. mi + m2 and r is the relative coordinate.4>)ru{r) = Ylm(9. f dr{u(r)}'2 1. we return to the Schrodinger equation in three dimensions.m2. and so u'(r) .r acting between two particles is constant and directed towards the origin (of the relative coordinate). For a central potential V(r) we write (15.6) = (I + l)u(r)/r = R ~ rl.e. 15. ~ (I + l)rl (15. R(r) = ^ .308 CHAPTER 15. We leave consideration of the logarithmic potential to some remarks and Example 15. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). In order to have a clear starting point. i.£ ] * ( r ) = 0. h2 2n V 2 * ( r ) + [V(r) .5) u'(0) u[r) r=0 = R(0).1) i. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15.e. (15.
63b): 2 and V(0) = 0 (15. and there identify the Swave states with those states of the onedimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq. (15. and the turning points are at a — — b := — xc. (15..4) in the threedimensional case. (15. d5.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.8a) ^ JO = (2Ar + l ) £ .1. (15.2. 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. x > 0. or f l H " ^ ) ! . which has been reduced to an effective onedimensional problem in the polar coordinate r. (15. We therefore proceed as follows. .8a) we can write the quantization condition (14.7) k(x) = h/y/2m0(EV(x)). 3 .63b) for the onedimensional Schrodinger equation.15. ."MA. so that with Eq.. odd integer ^=(nll)vh.. We now want to find a quantization condition for the threedimensional problem. This means that we transcribe Eq. (14. In order to to obtain the . Then k(x) = k(x).V(r)) = (2nl) +1 N.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!).9) into the threedimensional case in the following form: / Jo C dry/2n(E . We also assume that V'(x) > 0.8b) This condition implies that we have only one pair of turning points. is to be interpreted as the principal quantum number in three dimensions. N = 0. (15.2. .63a).10) Here rc is the turning point and n — 1.
6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E . s. Now. . c (nj\irh. (15. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/.\rvc = 0 . E A fv. (15.17) The integral on the right is the integral representation of the beta function B(x.310 CHAPTER 15. Power Potentials eigenvalues E —• En for the radial potential (15. _ —vr" *dr. (15.15) I ' = ^ 1 (( 1) B{x.12) and (15.14) Prom Eqs. dr = .y) defined by (xl)\(yl)\ y ^ (x + y .11) rc = (15.13) dt .13) we obtain _ X_ „ _ so that A£_ (15.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. (1516) (15.'(1*)1/2= J1F\lt)y1dt.
The case v = 4 is that of the pure anharmonic oscillator also discussed by M. $T(z) = {z. 15. for instance simply graphically as indicated in Fig. (15. i r ( i ) = r ( l + i ) = ( i ) ! . One can convince oneself now. 15.55).19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. Weinstein [281]. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. ^ '"Compare with Eq. .2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. (12. T(z + 1) = z\.1)!.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . and n — — ]7r/i.1.15.1 Approach to square well with v —> oo.
discussion at the beginning of this section).21) (4nl).oj = ^/2X/JX.41) for the . 15.61) h2 = I. Physically it does not make sense for the probability amplitude to have a discontinuity there. > and with Eq. (13.312 CHAPTER 15.oj2/2. we obtain E„.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.> r ( 2 ) A V 2 in \)*\W \it^T& (15.25) "Comparison with Eq. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. 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). Power Potentials In the case of v = 2. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. In the case v = 1. since these are the ones we are interested in (cf. x > 0.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). the case of the linear potential. The potential V(x) = X\x\ (15. (6.2) implies A s fj. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .mo = \.g — —2. For the following we set in Eq. we obtain En = (n .22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.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. the harmonic oscillator. (6.1 + \)hx> = {An • l)£fiw. eigenvalues (JV + h)tkj — (2n . 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.
Eq. i. Ai(s) 1 2v^(s)V4 exp x —> +oo. s = E. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0.29) .27) i.60) applies. (13.x)\x=0 = 0. dz2 According to Eqs.e.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .e.2 Behaviour of an odd wave function at the origin. i.56) and (14.15. (14.e. 15. (15. In the domain —a < x < + a .x).59) applies. i. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . (15.e.28) In particular at x — 0. i.26) For z = s := E — x —> —oo. \(s?» (15. this wave function has the required exponentially decreasing behaviour at infinity.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. E > V or s > 0. and for s —> oo the trigonometric behaviour of Eq. (13.e. Ai(s) "v^ G ' cos s 3/2 _ t (15.
51716 6. Co.e.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . copyright of NorthHolland Publ.3.04017 11. (15.93602 12.= ) = «. i.78671 7. of course. . *This Table is reprinted from C.33811 4. p..30) ! ^ / 2 ..08181 5.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large.. It is interesting to note that in the dominant approximation both methods agree.82878 1 2 3 4 5 6 7 8 9 10 2.94249 9.82814 i.02137 10.28) is really only valid for s large and s — oo.03914 11.32025 4.78445 7. L. It is possible.1 demonstrates the quality of the WKB approximation by comparison with the exact values.^ = !(2nl). 201.31) (an odd function has an odd number of zeros!).02265 10. Quigg and J. Rosner.00852 11. but for large > values of E Eq. Table 2. the eigenvalues En are determined by the zeros of the Airy function.93528 12. i.2. = im — (7r/4). to obtain the zeros of the Airy function numerically.94413 9. Table 15.e.00767 11.En from Ai(£„) = 0 2. n = l.52056 6. (15. The expression (15. [231].314 CHAPTER 15.e. Quantum Mechanics with Applications to Quarkonium. (15. so that n — 1/4 ~ n. with permission from Elsevier. (2/3)E3/2 or 3 / 1 2/3 E — En — .29) implies that cos (  ^ . Power Potentials Table 15.08795 5..
34a) **The most frequently quoted reference for this result is R.3 The ThreeDimensional Wave Function 315 15. P. D. We indicate briefly here — without entering into further details — how these quantities can be calculated. ee. (14. .g. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. In order to be comparable with experimental data these investigations. have only a finite lifetime and decay into other particles such as electronpositron pairs. i. i.• •) by a mean value. which are inversely proportional to the lifetimes.** r ( # ^ e e ) o c *(0) 2 . Weisskopf [239].e. We saw previously (see for instance Eq.e. e. F. of course. In general these quarkantiquark pairs. See also J. Jackson [140]. r <53) 1 3  dr cos 2 () (15.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs.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. van Royen and V. e.g.e.15. The resulting masses agree in most cases very well with those extracted from experimental observations. can be shown to be proportional to the modulussquared of the particle wave function at the origin. *Note that this makes N a W K B normalization constant.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. the charmanticharm meson ^ . We shall encounter WKB normalization constants at numerous points in later chapters. cos2()^— / 1 fr° 2 > w . The decay widths T. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. in Chapters 24 and 26.
e.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.39) in / .34b) VMEvy Variation or differentiation of Eq.2) and (15. so t h a t approximately l N2 / Jo k(r)dr .l N n .37) (15.rcu w X{r)dr = Kh . From Eq. (15.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( ..4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 . & (15.1 )n ..38) N : sin V ^ Jr dr' 7T (15.316 In accordance with t h e relation CHAPTER 15.35) N2 . (15. dE a—— / dn J0 so t h a t with Eq.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.40) VWY .\ \ * n = l. Power Potentials this averaging implies a numerical factor like 1/2. i.2. (15. (15. (15.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. A = (15.36) \i dEn irh2 dn According to Eqs..
* Using the result (15. The case of arbitrary positive power potentials as above has been considered by R.l ) n . MiillerKirsten [29].37) we obtain: 1 (2M^)1/2 47r ft. K.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. E. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. and hence with Eq. . MiillerKirsten [28]. S. for instance.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. 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). Bose and H. § S. MiillerKirsten [147]. Itt^pocn2*"1^"^. (15. W. J. Hence Eq. we have ^'(0) = ( . 3 . 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e.g.2. Kaushal and H. Bose. 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. W. K. J. J. . (15. Hite and H. S.23) for the energy levels of the linear potential.1 ^ ( 2 ^ „ ) 1 / 4 .38) implies 317 (15. .44) (15. the expression (15. Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. W. .15. for various potentials: (a) For a confinement potential of power v.41) I*(0)2=_L^)^. G.
— oo < z < oo.^ ( 3 ^ + l)55p^«(3^l) + .I l n f ^ p . Eq = i q .g ^£+(<*Pl»riy = 0. W. i. q = 2n + l. 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 . . (11.. Solution: Details can be found in the literature. D.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).L 2 + U{z)]<f>. Power Potentials *(0) 2 oc .. (15. Example 15. = (I + 1/2) 2 . S. J." 1 11 L . rewrite the equation as ^ z + [ . M. ip = exp[(z — c)/2]. Lifshitz [157].46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14.2.. *=^f. (15.46) and the classical Kepler period of Eq. .1. We observed there in the case of the Coulomb potential that the period is proportional to n 3 .. U(z) =  dz zf3e2a/?e2*.c = —a//3.e.318 (c) and for the Coulomb potential^ CHAPTER 15. Bose and H. MiillerKirsten [29].L2 h4 = 4/3exp[(2a — /3)/f3]. 7 With the substitutions r = exp(z — c).5. in agreement with Eq. E' = E + gln(r0).75). and there Eq. n = 0. K.133). (14. Landau and E. . {S = ^ > 0. = 1(1 + 1).j .^ +•••.
that no matter how far the particle is scattered away from the source of the Coulomb potential. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. A very similar potential is the Yukawa potential V{r) = g2 — . The result is a screening of the Coulomb potential which thus attributes it a finite range ro. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. This effect sounds unphysical. It is clear that for ro — oo the > potential becomes of Coulomb type. Historically this potential arose with the realization 319 pVr . In reality. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. and in fact. 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.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . so that like charges are repelled and unlike charges are attracted by each other. This implies. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state.Chapter 16 Screened Coulomb Potentials 16. it is. 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.
Since. (16. . from the relation Pi + P 2 + M2c2 = 0. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies.* The relativistic equation of motion of such a meson when free is the KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x. one frequently resorts in calculations to the expansion of the potential in rising powers of r.3) where for the real potentials we consider here.e. the mathematical expression with an exponential is not so easy to handle analytically. in fact the parameter [i represents the mass of such a spinless meson.ii k .1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. (16. 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 . i. 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)\ (16.2 ) or (p = hk) 1 /" . Heisenberg [133].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.1) We encountered Green's functions earlier.320 CHAPTER 16. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. however. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. and that. all coefficients Mj are real and independent of the energy E = k2.
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. 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.Q. . $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. Froissart [223]. Squires [258]. if>oo. Omnes and M. Frautschi [97] and E. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. A. i.e. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. Wu [47]. 2 . "These were first considered by C. l+n+l = ^P. Bethe and T. J. . Lovelace and D.5) where K — ik and A n is an expansion in descending powers of K. Kinoshita [21]. i. They are usually written I = an(E). i.r). ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H.k. (16. Cheng and T.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.e. See also the other references below. bound state eigenenergies and the 5matrix. (16. to be of the order of 1/k. however. C. Mandelstam [185]. 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. (16.e. .e. S.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i.1 in the above.mo being the reduced mass of the system. S S. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. Masson [180]. T.3) have been considered by various authors.r)=0. £+**PVV t/. f H.1. Potentials expandable as in Eq.k. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. (16.3) in the radial Schrodinger equation for the partial wave ip(l.§ 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. only for the cases of n = 0.16. .4) where E = k2 and h = c = 1 = 2m. physical) values of angular momentum as functions of energy E.{l.
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. Squires [258]. H. W. .4) to z = —IKv and set TP(l. A. tf H . Muller [204]. *See in particular A. in the present case at fc2 = 0 or k2 — Mi = 0. 'In the case of the Coulomb potential (cf.t Considering such superpositions of Yukawa potentials with /•oo / a(/j.ndp. J. Under these conditions the S'matrix is meromorphic (i. Regge [31] and E. < const.** The energy is later obtained by reversion of the resulting series. J. W. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. 16.z) Then x is a solution of the equation VaX= (16. one can assume an expansion of the potential V(r) in ascending powers of r. for all n. (16.19). Miiller and K. o rV(r) I dpp\V{peie)\ regular at r = 0.k. Schilcher [203].)/j. Bottino.322 CHAPTER 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. M. In the following we follow mainly the last two of these references. Longoni and T. 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 conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. J. as may be seen from Eqs. Screened Coulomb Potentials as a function of the energy.15) and (16. W. J. Muller [201] and [202]. < oo for all 0 < vr/2. Section 11.9) the cut starts at E = k2 = 0.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H.z) = ez2/2zl+1X(l. (16.e.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.k.
2 Regge Trajectories where d2 Va = z^ and (cf.7) is seen to be of order 1/K.a + j)Smi(a. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a). (16. W.11b) ^ The associated boundary conditions are: So(a.a + + l).0) = 1.j) (16.b. MiillerKirsten [249].b. j)Smi(a. so that to leading order we have VaX{0) = 0. anharmonic and cosine potentials have been derived in L. (a. a) —b — 2a. By a repeated application of the recurrence relation (16.5) is equivalent to a = —n.j +(a + j + l.a+ l)$(a + 1) + (a.16. . as in the case of the Coulomb problem. {0) X = Ha. The function <&(a.z) = Ha).j) = (a + j l.10) zm$(a)= ^ j=m Sm(a.a)${a) + {a. a + 1) = a — b+ 1. (16.i 0) = 0. K.9) we obtain m l)$(a .j l) + (a + j.a + j)Smi(a. Sharma and H. which means that the hypergeometric series breaks off after a finite number of terms.a — l) — a — l.5)) a = + 323 . Eq. and all Sm(a.j)$(a + j).awhere (a.* Recurrence relations for coefficients of perturbation expansions for Yukawa.9) (16. We also observe that the ansatz (16.10): Sm(a. (a.j) for \j\ > m are zero.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.j) may be computed from a recurrence relation which follows from the coefficients (16. (16. J.1). (16.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. where $(o. all other So(a.11a) The coefficients Sm(a.
j).aj 2 \ [a. We first cite the final result .[ a + l.a\4.12a) where [a.13) one obtains the quantity An(K) and hence with Eq. when n = 0. [a. 1 . a .a]l^a) + ^2(2K)i+1 Y.5) the Regge trajectories I = ln(K).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. (16.l. i. 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).7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except.a + l ] 2 [ . (16. + (16.e.324 CHAPTER 16.An{K).aj 2 (2K)A _ [a. Va+n = Va .a + j]i+l = MiSi(a. Evaluating the first few terms of the expansion (16.o]i = M 0 .' — . [ a . of course. oo _ . This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. (16. [a. .13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. Any term /j. 0 < \j\ < i.n. (16. i R »] = 2K[a. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .1] 2 . Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq.7). for the coefficients M^> of the expansion Details can be found in the references cited above. the latter can be written .[a>a + J]i+i*(a + 3). (16. and hence Va$(a + n) = ra$(a + n). #(a + ra) fjL$(a + n) V. This follows from the fact that D a $(a) = 0.<&(a + n) on the right hand side of Eq.
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. i. we obtain the expansion for the Regge poles. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A. Burke and C.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16. .16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .1) + 3M3M02 + M22n(n + 1) (16. 16.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). Ahmadzadeh. Thus the expected appearance is that shown in Fig.§ This is an important observation which indicates the equivalence of the methods.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex iplane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling.1. .5). G. Tate [6]. However.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .1 by evaluating the first two terms.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . Inserting this expansion into Eq.e. (16. (16. including an exploration of the domain of small energies E.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . of course.2. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. P. Obviously the expansion — which.^ p [ 3 M 4 ( n .1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . With numerical methods one can achieve more.
a)So(a. The conjugate variable to energy is time.0) = (a.0). just as a decay width T represents the width of the resonance in energy.2 a = In See S.11b) we obtain Si(a.^ . 0) and Prom Eq. 0). [a." Re I 1 0 1 2 Fig.O) = Other terms vanish since So (a. Frautschi [97]. Similarly the conjugate variable to angular momentum is angle.1 Typical Regge trajectory for a strongly attractive Yukawa potential. Hence Si (o. and the angle A#. 114. one can expect a resonance with the lifetime determined by this imaginary part. 16. a] 2 = M i Si (a. p. For a resonance with a long lifetime.326 CHAPTER 16. C. (16.—" + In = — . . From Eq. where the imaginary part of I is very small. However. Example 16. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. In fact.13). (a. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig.1). through which the particle orbits during the course of the resonance. (16. ±1) = 0. so the imaginary part ai of I — an represents its width in angular momentum. a] 3 = M 2 S 2 (a. satisfies the relation ajAO ~ h. 16. and the lifetime At of the resonance satisfies the relation TAt ~ h. .12b) we obtain [a. ai is small and A6 is large. Solution: We evaluate the first three terms of expansion (16. For a bound state a j = 0 and the orbit becomes permanent.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.a) = b .
1) (a .2 Regge Trajectories 327 Analogously we evaluate S2(a.2 a ) .13) we obtain 0 (Mo . a)Si (a.a + l)S0(a.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . 1).0) + 0 + 0 = ( a . Solution: Ignoring higher order terms in r.16.0) Si(a. Thus (with terms which are 0): Si(a. . a)Si (a. 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 .0)+ 0 + 0 = (a1). i ) again from Eq.b)(a .4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity".00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . We obtain the quantities S i ( a .6)Si(a.An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 . Sec.a)5o(a. a)Si (a.1.6 + l)a + A. 14.1 ) + (a.n ( n + l Inserting these expressions into expansion (16.1) + (a . 1) Hence 52(a. (a.6 + 1 ) . we have to evaluate (cf. A" An An A" = = = (a.0) + 0 + 0 = ( 6 .65) 2m dz V'A + 2Bz + Cz2 • Z— 0. (a. (n+l)U+^pJ . (16. One sets z = 1/r so that. 0) + (a + 1.al)S0(a.0) = (a .2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.0) (a . 0) + a S i ( a .A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .1) Si(a. . using the Cauchy formula (6.65) 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 .1 ) + (6 .11b).2o)Si(a. .
1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6).* 16. Example 16. .328 Thus here dr K' CHAPTER 16. we can use Eq. 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. The turning points in this case are at r = 0 and r = oo. ±2HK in agreement with our expressions above.2. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. I. Boukema [34] and [35]. i. Screened Coulomb Potentials Mo . by expanding the right hand side in rising powers of v and evaluating the individual integrals.l. however. *J. Here we do not perform this procedure. Singh [252]. n = 0.15). tSee Chapter 11 and V.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. (16.. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. 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 . (16.5) together with the expansion (16.e. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A.t First.3 The 5Matrix It is clear that if we now work through the usual procedure for the derivation of the S'matrix we pick up a logarithmic phase as in the case of the Coulomb potential.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z .
17).16.14).19) we obtain the energy.19) +2M 3 M 0 (3n 2 + 3 n . J. which are precisely the expressions yielding the Regge trajectories of above. .l)n(n + l)(n + 2) + 2M2M02 . J. We observe also that the S'matrix is unitary as a consequence of the result (16.* 16. i. 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. § See H. Now reversing the expansion (16. n = 0.2. Miiller [211]. This has been done§ and one obtains .17) and is therefore real for real values of the potential coefficients Mj.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16. Miiller [204]. W. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16..l. W.l (two additional terms are given in the literature).g. the behaviour of the scattering phase in the domain of high energies. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\..e. .. Paralleling the case of the Coulomb potential in Chapter 11.4 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\. for instance. such as.0[(KZ + Mx)'6} (16. 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 . the following *See e. One can use the explicit expression (16.18) of the Smatrix now to explore further aspects. H.
Mendelsohn [135] and E. in which the leading term is the usual expression of the Balmer formula for the Coulomb potential: K2 = Mi+ 1 ° l4n(n + l ) ^ ( / + n+l)2 4(7 + n + l ) 2 Mio +4(2n + l ) ^ ( / + n + l ) 3 4(Z + n + 1)4 ( l) ^ — .1) + 2M2M.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Zauderer [286]. the transformation was introduced in the form given below by G.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .e.l)n(n + l)(n + 2) . Frautschi [97].330 CHAPTER 16. Screened Coulomb Potentials expansion.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . .f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n . i. 107.< 3 M 4 M 0 ( n .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 . Iafrate and L. E. Large coupling expansions derived in H. VahediFaridi [212] contain as special case the expansions of the first pair of authors. Watson [280] in 1918 and later resurrected by A. 'According to S. C.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. Sommerfeld [256].l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . MiiilerKirsten and N. B. N. J.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.^ 16. p.2)(n . Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. 282. p. W.11) + 2M22M03(9n2 + 9n . k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).* We recall from Chapter 11 the definition of the scattering amplitude F(9. Warburton [279]. A. in the following expression with an "See in particular G. J.
''' The idea is now to consider each term of the expansion (16. 16..21) The scattering amplitude determines the experimentally measurable cross section o given by da (16. The scattering amplitude possesses the partial wave expansion F(0.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral.24) Si(k) being the phase shift.184) to (11. I = n + x.k) 0ikr (16. For this purpose we consider the following integral taken along the contour C shown in Fig. ~:' sin nl (16.2 The integration contour C.2... z=o in which (cf. Actually 5j lrr/2 as in Eqs.k)~l] = ~ k" PiSiW sin 5i(k) and S(l.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0. .fe)P . so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).26..k) = 2zfc L " v "'' v . dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. Eq.1.k) = ^T(2l + !)/(*.22) = \F(9.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1.187)) (16.k)\2.25) Setting Note the minus sign in the argument of the Legendre function.kmcose). c = 1. 16. \x\ <C l . (11.iJ 11 —[S(l.187). (cos 9).k) — e 2i5[(k) ~" p—iirl (16. (11.23) f(l. n = 0.
28) . we obtain m® = ± I dl^±^f(l. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. In order to see the relevance of Regge trajectories in a reaction of particles. t = (qq')2. 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'). for instance.k) Zi Jc TT{L . be elastic scattering of a off b. For the momenta indicated we set (with metric +. 16.—.—) s = (p + q)2.26) P((+COS0) = 2 ( 2 n + l)/(n. Thus the direct reaction may. n=0 which is the usual partial wave expansion. 16. a + d — • c + b. u={pq')\ (16.3. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.27) a(p) © b(q) Fig.n) oo {l)1 Pt{cosO) v ' (16. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig. in which case c = a and d = b.fc)P„(cos0). as indicated in Fig. The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d. 16.332 CHAPTER 16. a + c —> b + d.—.3 Reaction channels and their respective Mandelstam variables. d(q') (16.3.
there will be several such amplitudes which together describe all three processes).29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s.t).0s). M.§ Since the three variables s.g. .30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). the only kinematics we require in the following is given by the relations (e. Thus. the momentum k there would correspond to qs) s = ~4(ql + ml).^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. one then has A(s. u. Ft = £ ( 2 J + l)F.u are not independent. t. 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. t = 2q£(i . B. A(s. Chew [48]. For a Lorentzinvariant normalization of the initial and final states.9s)\2.u) (if any of the external particles has nonzero spin. Symbolically we then have the situation shown in Fig.g. § See G.2 ^ ( 1 + cos6 s ). the variables t and u would describe momentum transfers in the crossed channels. N. Mandelstam [184] and R.t) = VtFt(t. T See e.The variables s. except for poles and cuts which characterize the reactions in the three channels. s = 2qf(lcos0t)(16.32) t = 4(q? + ml). Treiman [25]. u = . One now makes two hypotheses. S. A(s. the dynamics is described by the exchange of quantum numbers.4.(<fe)^(cos 6t). In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).t. 16.t. N.u are known as Mandelstam variables. e. one can write the amplitude as depending only on two.t. Blankenbecler. which is an analytic function of the variables s. if we had been considering the s reaction originally.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.t. etc.et) (16. In a Regge theory on the other hand. if s describes the square of the total energy in the s channel. L.t) = y^Fs(s.31) i l Taking all particles to have the same mass mo.c o s 0 s ) . Khuri and S. (16. Goldberger.. (a) Mandelstam's hypothesis: All three processes (described by the s.16. F. (16.g.
334 Yukawa CHAPTER 16.)e~ •fj. But for integral values of I. 16. the Legendre polynomial Pj(cos 9) does not possess a cut.4 Yukawa versus Regge theory.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. So what can one do? 1 H. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^.e. in the domain of schannel physical values of the kinematical of the ichannel). . Lehmann [162].5 for Yukawa potentials V{r) u: lm cose dr a(/j. the partial wave expansion is not valid in the physical region of the (i. 16. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. 16. But > > wave expansion X)(2Z + l)F. For s — oo : cos 9t = 1 + s/2q^ — oo.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. However. Screened Coulomb Potentials Regge oc(t) Fig.5 The Lehmann ellipse.
16. to use the SommerfeldWatson transform. 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. „ i.^ B 1 A Fig. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. Thus consider now the same integral but taken along the closed contour C shown in Fig. i. all poles of Ql > 0. B Iml 1 \ i All \ r\ 0 / closed . Moreover.6 The integration contour C".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.1 / 2 . sin nl [16.5 The SommerfeldWatson Transform 335 T h e answer is.16. the contour representation (16. N F(6.6.26) but for a different choice of the contour. contour C / / D ^ ^ . 16. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.t). and there is only a finite number N + 1 in the domain of 9W = .e.33) .e.
e. as we discussed briefly in the above. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. K. Predazzi and T. (16. Limic [177] and E.e. p. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. . MiillerKirsten [29]. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. this behaviour has been confirmed in high energy hadronic reactions.e. at high schannel energies. Foissart [223]. if ao(t) is the Regge pole with largest real part. Eq. for high energies in • > the schannel). the amplitude can be represented as a sum over ^channel Regge poles. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. Apart from more refined details. J. i.35) This result is valid for s — oo and t negative (cf.336 CHAPTER 16. of course. Bose and H. i. It follows that under these conditions. In particular. 16. Regge [229]. Omnes and M.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.32)) and finite. W. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. 42. we have (16. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. Regge trajectories have.t)&B(t)sa°V. 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. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. for plots and discussion see R.
The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. .g. and is composed of quarks. The above treatment of screened Coulomb potentials is incomplete. J. a composite particle being one with structure.6 Concluding Remarks 337 important parameter in string theory. and also as socalled "fixed poles" .16. The way to do this is similar to calculations in Chapters 18 and 20.1. the latter are related to the distinction between "' elementary and "composite particles". H. For an overview see e. Related aspects are discussed in Sec. called "Regge cuts". W. Singularities appear also in the form of cuts in the plane of complex angular momentum. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. 20.** Whereas the former are related to absorptive properties of a reaction. like a meson or a baryon. MiillerKirsten [209].
This page is intentionally left blank .
is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult.) which are functions that interpolate between the n.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. 339 . called elliptic modulus. See F. M. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. the most immediate candidate is a trigonometric form like that of the the cosine function. example 14. since many analogous relations hold. 2K or AK. are not much harder to handle than trigonometric functions. 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.0 < k2 < 1. The reason for these difficulties is. p. and are themselves periodic with period e. motivated mainly by the regularities of the crystal structure of matter. 25. The Schrodinger equation with this potential. Arscott [11].g. 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. K being the complete elliptic integral of the first kind. depending on the parameter k. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.* These elliptic equations involve Jacobian elliptic functions like sn(a. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. however. In essence these periodic Jacobian elliptic functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. 19 and p.Chapter 17 Periodic Potentials 17.
1. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .2h2 cos 2z]y = 0. 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.4.1) where h2 is a parameter and — IT < z < IT. the three Jacobian elliptic functions s n z .e. cosine and tangent. Lame equation: H. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. Like the trigonometric functions sine. [205]. W. Miiller.1. need not be afraid of them. (17. ''The reader who encounters these Jacobian elliptic functions here for the first time.7. the Lame equation possesses 2 n + l polynomial solutions. . each with its own characteristic eigenvalue. The most important formulas for our purposes here are collected in Appendix A.2 . Periodic Potentials vide periodic potentials like trigonometric functions. W. For large coupling. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. 17. 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.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. The Mathieu equation. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. which is conveniently written ^ + [\2K2sn2z]y = 0. which does not possess any polynomial solutions. J. In fact. and to relate these to the weak coupling bands or regions of stability. those of integral order. The first of these conditions is already tMathieu equation: R. We might mention already here the much less familiar but more general Lame equation. Both of these domains are important for a host of other considerations.340 CHAPTER 17. or Sw&ve Schrodinger equation with elliptic potential. Muller [73]. (17. This is a significant difference compared with the Mathieu equation. c n z and d n z are handled with analogous formulas. See also the discussion in Sec. 8. B.. Dingle and H. 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. 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. which one looks up in books like that of MilneThomson [196] or Tables when required. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. J. i..
h2 = 0 belongs to the domain of stability. the case of h2 small.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. 17. Although the mathematics literature on the subject uses the physical concepts of stability and instability. the semiaxis A > 0. i.e.e. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. 17. a bounded function. Salem and T.2. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. Looking at the Mathieu equation for h? —> 0. we compare the For interesting related discussions see M. y" + \y = 0(h2).17. . since the solution there is of the form cos vAz. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.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.e. i. as in Fig. It is this additional parameter function appearing in the solution of the Mathieu equation which attributes the equation its reputation as being particularly hard to handle.e. In classical mechanics stability is treated for instance in connection with planetary motion. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. 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). we see that in a plot of h2 versus A. Adding the negative term — 2h2 cos 2z to A.^ 17. these concepts are rarely explained as such there.1. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r).2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. Vachaspati [242]. i. i.
b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz.e. Setting z = 0.342 CHAPTER 17.2/i2 cos 2(2 + vr)]y(z + vr) = 0. y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2).B and a. a = const. y~(z) = 6sin vXz. ° ^ [1 + 0(h2)}. i. y(2vr) 2 a = —7—^ = 1.e. d(z + TT) cos 2z We see that there are solutions which are proportional. (17. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). y+{z) . sin V Az or linear combinations.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). d2y(z + 7r + [A . 2/(0)' yfr) .3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A.2h2 cos 2z]y(z) = 0. we obtain y(vr) y(2?r) a = — — a = —^^r. — • . (17. i. Periodic Potentials equations ^ p + [A . TT. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants.
ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. 2/_(0) = 0.y+(7r)y_(7r) = 0 or abVXa2 . we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. [y+(v) . Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. (17.oWX] . 7 y\z) = ay'+(z) + (3y'_{z).y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.e. \y+>y\ = y+{ir)y'{ir) . and by'+(?r) = —ay/Xy{n).y+(7r)j/_(7r)} = 0. we choose these with the following set of boundary conditions: y+(0) = a.aa][y'_{n) . with constants a and /3. y'(z + n) = a[ay'+(z) + py'_(z)}.3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. This expression is equal to the coefficient of a2 in the quadratic equation for a.a2b2\ a± ~ 2a6\/A . the determinant of the matrix must vanish. Using Eq.4). y'_(Q) = bV\ with Wronskian W[y+.{ay'_{ir) + bV\y+(ir)}a Now from Eq.17. We can write therefore. yV(0) = 0. This means. y{z) = aty+(z) + 0y(z). i. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. 2/'(7r) = a ^ W + ^ . The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. (17.
+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. This is the easiest application of t h a t method since only simple trigonometric expressions are involved.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). we have <7± — CHAPTER 17. _ / ± v 7 ^ V . 8. ** An important consequence of Eq.0 This series may be reversed t o yield the Floquet exponent. 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 . (17. cr+ = 1/<T_. **This point is not immediately clear from mathematics literature. .2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .455A2 + 1291A .6) We can derive various terms of this expansion perturbatively with the method of Sec. where the constants are usually taken as unity from the beginning. <7__cr_ = 1.= — . Eq. Prom this condition we determine later (cf.22)) the boundary conditions of periodic solutions. Setting cr+ = e . T h e result for A is the expansion A = h4 2 .e.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. 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.8) "Observe that for integral values of v (in lowest order of h2). Periodic Potentials g= 2aby/\.e. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 .7. this implies y+(Tr)/a = ± 1 .5) a The parameter i/ is known as Floquet exponent. (17.344 Setting / = 2by/\y+(Tr). i. (17. This calculation is demonstrated in Example 17. <T__ + <T_ = 2y + (vr) .1.. . i. 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. (17. (5i/ 2 + 7)/i 8 vz + — .
124.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17.5) and (17.12) S e e J.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.2 how the expansions for small values of h2 may be obtained perturbatively. W. = J—^ + o(hr). rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). (17. . With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. Example 17. Solution: We write the eigenvalue equation A = v2. (17. For these integral cases of v we demonstrate in Example 17.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . Schiifke [193]. cos TT\/A 32A(A1) 2 .1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y. (17.35A + 8 .3). (17. (17. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. These cases therefore have to be dealt with separately. Meixner and F.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . (17.17.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.11) or smvz or e±ivz. p. VA = ±n. as will be done later. It follows that in these cases i? = n2[l + 0(h2)]. 8.1) 15A2 .10) A = n2.„„ x n.35A + 15A2)/l8 /. we must have a2 — 1 and hence e 2iv/A7r = 1} i. n an integer.e. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.jl2.
16) *These smallh expansions are convergent. (17. by J.14) = = = cos(y+ 2)z + cos{y— 2)z. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. . (17. i. for instance. '. (17. v . (17.2)j/„_ 2 + {v. h2(u. . as shown. Meixner and F. ("• ^ + 2) (o) 2(2^ . so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.v + 2)yu+2].(2) + . Next we treat terms yv+a in j / 1 ' in a similar way.!/) = 2A We now observe that DvVv = 0.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq.e.16) directly into the original equation in the form of Eq. = . me±v(z. siniy + 2)z + sin(i/ — 2)z. up to 0(h2) we have A = 0. .e. and (v. when a = 0 or 2v + a = 0. i.v— 2) 2 ^ + yu+2 Vv 2 y (D = h _2(2u2) 2(2i/ + 2 ) ' Then up to 0(h2) the sum j / 0 ' +y^ is the solution provided the remaining term in Ri. (i/.e.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. e ± i ^ + 2 ' 2 + e±i("2)z. v)yu + {v. W.11).2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. se and me as* oo i y = „«>) + yW + i. The recurrence relation can also be obtained by substituting the right hand side of Eq.. (17.3 and 19. i. (17.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .2) _ __ _ (y. v .h2).3.„ + £ h2i £ P2i(2j)yv+j (17. or A = 0(/i 4 ). see Example 17. Proceeding in this manner we obtain the solutions ce.346 CHAPTER 17.v±2) = l. (17. 17.h2).e. of course. since j / 0 ' = yv leaves unaccounted R\.2) which applies when v is nonintegral. sev(z. With our perturbation formalism — as demonstrated in cases considered in Sees. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes.2. h2). v) = 0. The right hand side of Eq. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .15) Hence a term fMy„+a on the right hand side of Eq. where except. Schafke [193].14) therefore leads to the following nextto leading order contribution (v. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.
Meixner and F. .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.18) Again applying Eq.7). .1 with A = v2 — 2h2A but with v = 1. the solutions are not yet normalized.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. The higher order terms naturally require a little more algebra.v)\ . (17.. p. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. 2 ( 2 i / ..15) (but with v — 1 and a = 2) we obtain the next to leading contribution to the solution as (i) = h2 =_fe2 V {2(2v + 2)}„=1y3 8 y3' This contribution leaves uncompensated the terms amounting to R\ = ~ — 2h (A + cos2z)j/ 3 o = —[Ay3\ 4 \ 2 Hence the next contribution to the solution is 4 V 8 2(4)(2 + 4 ) / 4 V 8 48 Proceeding in this manner we obtain as the complete solution the sum y = y(0)+yW+y(2)+.17.2 ) v(y2."2).. i.. W. thus determining the quantity A. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. Schafke [193].u) + h4 (". (17. . one has to deal with each integral case separately as in the following example. * Note. (17. For v = 1 we then have yi = cosz = y~\. . v) ' ( 2 v + 2) (17. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. fo^ + 2) . 0 = h2(u.14). 121. Example 17.1 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]. Following the first few arguments of Example 17.2—r——— (i/ + 2. however.. 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>.17) Inserting 1 for the stepcoefficients of Eq. „ .R{ . + y~i) (17.g. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . See e.e. J. Solution: We start as in Example 17.
J. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn .24c) below) . 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]. 123) (there the solution is called cei(z.n integers. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. . and v is to be found in ascending powers of h2. Jo dz cos z h2 cos 3z  iC2T It J* cos z H 64 cos 2 3z  64 It follows that 128 and the normalized solution is therefore h2 Vc . h2). We introduce a normalization constant c and rename the normalized solution y then yc. Periodic Potentials h4 r ••• . see below!) one obtains the additional terms given by Meixner and Schafke [193].* The associated solution is ym+ym. Meixner and F. 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. h4 (17. The reverse situation is later of importance. p. With normalization (not to 1. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. 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. W.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. 120.348 From this we obtain 2h2A = h2 CHAPTER 17. In the above we considered the case of v an integer and calculated the eigenvalue A. i.e. Schafke [193]. (17. We consider this case again in an example. m.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). the case of A an integer. Eq.
. ^S. on the left hand side of Eq. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + . fe47T2(A4) 4%/A(A .• 1 + 5n h 2 s 2932 Setting in cosm. 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 . Manvelyan.9b) about A = 4. (17.3 1 x 4 + 8) 274233 (. (17. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . (17. J. S. H.1) (A . o 1• • • . Hashimoto [124].Q . Liang and Yunbo Zhang [189]. W.4 ) + .('108\/E\2/ Expansions of u for integral values of vA have recently been given with a larger number of terms:^ l//i\4 3 el 2J 1 fhy 15 1 2 / + \/A = 4 : v § 311 4320 V 2 ' + 1555200 V 2 + 137 (h\° 305843 (h 12 27000 I 2 ) + 680400000 V 2 133 (h 12 + ••• R . .l ) 2 (A . (11A2 . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.1)2/A TT2 +h8 (15A2 . Gubser and A.1) .9b) and considering the approach A — 4 gives > (A4)2 + .3 5 A + 8)TT2(A4) _ 3 / / 64(A .= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . .17.9b) v — 2 + <5.3 1 A + 8) ( A . Substituting these expressions into Eq. MiillerKirsten.2 Cosine Potential: Weak Coupling Solutions Example 17.3: The Floquet exponent i> for A=4 and h2 small Show that the expansion of the Floquet exponent v around \f\ complex expansion: v = 2 349 = 2 is given by the following iVE/h 3 \2J 7% h (108^/5 \2J 8 1185H (h 31104^X2 Solution: We proceed as follows which demonstrates explicitly how the singular factors cancel out systematically.4)Av A2v A ~ 32A(A . J ..
bV\ y+(0) 2/40) 07.w^.29. We now derive these boundary conditions. for instance.W4W=UtWSta±I(.21b). solutions for integral values of v. 28 .10) applies.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. "These conditions. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.21a) and (17. = and s+w 2tW±I.350 CHAPTER 17. (17. (17. Taking in particular Eqs. Arscott [11].1 Boundaries of domains of stability. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. pp. as also conditions (17.4) into account. M.2. These are defined by specific boundary conditions. i. Fig. 17. .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. may also be found. We verify these for the large/i 2 solutions in Example 17.. 17. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. in F.e.6.
. /7r\j/!_(7r) y L l 7TJ . /TTW^) 2j^T±y+V2 From these equations we obtain y+{ir) in terms of functions at TT/2 by eliminating y'+(Tt) from the first and the third equations. we obtain y+(7r) _ i  2y_(f). f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. which after some rearrangement can be written y+(vr) _ .21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ . o y+(fy(f)ob\/A _ = 1 + 2I&A/X .4(f) (17. Then y+ y 351 U .17. ^\y±M /vr\yL(7r) . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. .
Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.T + . the case of ai with solution cei (of period 27r).' fh2 V2ceo = l .. This is shown schematically in Fig. (17. 17. y > se 2n +2s 9o +i The functions so defined have respectively period 7r. There the second term on the right is nonzero. (17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. ce± = cos 8 64 h4 5h8 b2 = 4 1 .24c) 8 h2 sin 4z \ .5) that the left hand sides of Eqs.• a2n(h2).2): hA 7h8 29h12 + Y l28^04. (17. (17. 27r..24b) 8 h2 z —— cos 3z + • • • . (17.2 we require this relation for nonintegral values of v. These may be subdivided into classes depending on whether the integral value n — 0. 19.21b) must be ± 1 . in this order. of v is even or odd. 2K. . Mr7nA (17.352 CHAPTER 17. (17.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .1). 17.3. 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 .— — — • • • . IT. .24d) v 12 ' **To avoid confusion we emphasize: In Sec. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. and whether the function is even or odd.. A > b2n+i(h2). A 2 y • ce2n=q0i. These are defined by the following boundary conditions.1.. was treated in detail in Example 17...23) We can now see how the domains of stability and instability arise. We see that there are four possibilities and hence four different types of functions. y > se2n+i=qo. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq.22) * a2n+i(h ). sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . (17. A *• b2n+2(h2). Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2.2. y > ce2n+i=qo.21a) and (17.+ .
17. . 17.1 that the boundaries of regions of stability merge to lines.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. 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. 17.3.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.2cos4.3 17. Miiller [73]. Thus in Fig. W. Dingle and H. In the case of the cosine potential.24e) 17. the walls are not infinitely high and hence tunneling occurs from one well into another.> (17. 13824 . We can see from the inverse of Fig.2 The cosine potential.17. 17. 353 .3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . .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. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain. The cosine potential cos 2z depicted in Fig. 17.17„. B. . J.z3 ce2 = cos 2z .hz — 12 . Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. of course.
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 = 0.2 and 8.26) 17. Sees. In the case of finite heights of the potential barriers. We therefore wish to expand it about these points.3.. Ion (17.2 The solutions We insert Eq. n 0. we therefore set 9 A A = 2h2 +2hq + —.26) into the original Mathieu equation and obtain y" + 2h q + rrxr .2. there is tunneling from one barrier to the next...25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. 4 (17. Thus in this case the quantity on the left is not an exact odd integer. Periodic Potentials minima at z — ±ir/2. (17. (17. of course.1. where A / 8 is a remainder. as in the case of the cosine potential.27) .2/i cos z V = 0. Taking the remaining terms of the cosine expansion into account. but only approximately so.3) and is. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . simply the onedimensional Schrodinger equation for the harmonic oscillator. 8. and we set this equal to q.354 CHAPTER 17. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1..
30) D A): COS « = 4z + 1 iq sin 2.31) We make the important observation here t h a t if one solution of the type (17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig. 16/i/ (17.27) we )A = 0. Choosing y(z) = A(z)e2hsinz. z). 17.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z).17. (17. can be . (17. 2 (17.28) A" ±4hcos zA' + 2h[ q =f sin z \ A.29) is known. the linearly independent one. (17.3 Schematic picture of Mathieu equation eigenvalues. another solution.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . and substituting this into Eq.
9) (9.9 + 4) = (9. ? . 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. 2 A A = («.35) But since and so a term iiAq+n on the right hand side of Eq.4 ] ._ 4 . 2[(92 + l) + A].\z To lowest order the solution A is A^> = Aq.32) cos^'+i) [\K . (17.94) \ — Aq4 (17. 2 /i D^Ag = 0.35) not yet taken care of is set equal to zero.33) (17.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17. of course.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. D{q%Aq+4i = 0. when i = 0 — the terms not involving Aq in (17.9) = 0.94) = (9 + l)(9 + 3). ^ ) A + ( ? . ? . (17. i. (9l)(93).4)A. (17. (9. (9.^ 6T [(«>« + 4 ) A ? + 4 + (Q. Aq+4 2Jh .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g.4 ) ^ .356 CHAPTER 17. .Q)Ag + ( ? .30) amounting to Ri ] q = .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. 1 + 4 ) ^ + 4 + (^. (17. With some algebra one finds that 16^ + where (9.36) Then ^4.e. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout.g + 4) .
g + .40) .. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known.4 . can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q. . ) ^ 4._ 8 (17.8)A. h) = A(z.4.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. q.q 4)Aq+4 q + 4 1 q + + + + + <^(.38) Aq8 + 1 2 Clearly we can continue in this way.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. q .37) and so on — left uncompensated — must vanish. . (q.)+ & ! = « ( . an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. (17.q + 4)(q + 4. Thus 0 = (g.g4) 4. +4. Q . (17.35) and (17. Then for A = A(0)+A(D+A(2) + . except those involving Aq.q + 4) (g. q. h).3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. h).4) (q . h) = A(z. to satisfy Eq. (17. A(z. q (17.<J+4 Aq+& + 1 2 1 . .. q.39) which is the equation from which the quantity A and hence the eigenvalue A is determined. (q.17.4) 1 1 (q. q.37) Here all the terms. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq. q .30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .4. q . the sum of the terms in Aq in Eqs.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout. (17.g + 4)(g + 4.q 8)Aq+8 (o)' ^±V(q 1 4.4) (q . A.. Thus in particular A(z.
where Hen{w) is a Hermite polynomial when n is an integer.43) 9 2eh .45) {B) We recognize Dq as the differential operator of the Hermite equation.d2B w dw2 + w(l w dB_ dw d dw 1. . Note different definitions in use.) i ( Q .67). 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. Before we study the eigenvalue equation in more detail.42) (17. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). The differential equation there is Eq.e. (17.41) so that the early successive contributions decrease in magnitude. (6.358 CHAPTER 17. 2 w2 + H* (17. We observe that for the large values of h we are considering here.i ) We encountered Hermite polynomials earlier in Chapter 6." Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (91) H (17.46) 2s(«.44) where D(B) 9 = dw2 $_ !)• (17. i. 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. and discussed there. We return to Eqs. it is found that D(B) z+ — ]B = 0.28) 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 . (17.
49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q.e. The solution with contributions B^°\B^>.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). i. 16hJ (17.7 T H 1 2 Z 4 <1..qi)Bq_4.51) .e. as in Eq. (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh. 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. y — Cexp(2hsinz) and Cexp(—2/isinz).50) C = 0. 1 COS I . (17.47) where the coefficients are the same as before. q)Bq + (q. also the same coefficients in the case of functions C below.e.34). i. Finally we obtain a third pair of solutions..e. (17.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. . i. the expansion for the quantity A being identical with that of the previous case..3 ) \He* ! (9+1) H. with the equation C" + 4h cos z C + 2h ( q .sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.q + 4 ) £ g + 4 + (q. i. A{wz (17.48) This shows that this solution is valid in particular around z = ir/2.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. which is an amazing and unique property of the Mathieu solutions — cf.
w2)He'n .59d) of Hermite functions Hen(w). Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.h) ocB{z. q + 4)C 9 + 4 + (q. (17. 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. 2' the remainder 72.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. show that the perturbation remainder of Eq. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.(w2 + A)Hen.7 T 1 2 Z 4 <1. 17.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).44). (17. Example 17. (17.q. We have thus determined three pairs of solutions of the Mathieu equation. ie 1 COS I . (9+3) (2n 2 + 2n + 1) + .54) where the proportionality factor is complex. each of them associated with one and the same expansion for the eigenvalue.4. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq.q. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. He'n(w) = wHen(w) Hen+i(w).53) This solution is valid in particular around z = —7r/2. 2 J where n=(ql).360 CHAPTER 17. (17. (17. We note that C(z. . A closer look at that expansion is our next objective.4)C g _ 4 .q.h).A Hen .n2(n l)Hen2 .52) and the coefficients are again the same as before. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.4: Hermite function linearization of perturbation Using the recurrence relations (6. These domains of validity are indicated in Fig. q)Cq + (q.
q4) (17.q + 4)Bq+4(w) + (q. ) . the solution y = exp(2/i sin 2)^(2). 17.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .3). We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.)Bqi(w).56) (94.q + A){q + A. . (q.' 361 in 2*f*1>[i(gl)]! = (q. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.i ) ( w ) 2l(.0 and —4.17. with (.3. ^ 1 Hv" '^ ^' ^ ' .55) M2 = M. We can write the solutions obtained previously. e. r j=r.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . in the following compact form with coefficients Pr(q.g.l)[l(q _!)]. l)Ag4 + P2(q. (?) 1 1 + (9.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.4) = (q . q + 4) = (g + 1)(? + 3). no intermediate move to or from q being allowed. (q. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. 2 + ^ M 3 + (17.l)(q .g4)(q4. (b) The allowed moves are +4. . q) = 2[(q2 + 1) + A].57) . q)Bq(w) + (q.g) 1 1 and so on. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. q .3 T h e eigenvalues We saw above that along with each solution the following expansion results: 2A = M 1 + ^ M where Mx = 2(q2 + l).. (q.q + 4) (q + 4. 2)Aq_8} + • (17.1i ..q 4.
P 2 ( g . l ) . .e. i.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.g + 8)(g + 8. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g.362 Clearly CHAPTER 17.58) with P 0 (g.i (g. j) = (g + 4j .l ) .j)] = 2j]jPr2(g. B.1) + (g + 4j.j). Periodic Potentials PlM)=(M+i).60) M2r+l = = *R.3) ^3(9.34) we deduce that Pr(q.j).g + 4)(g + 4. j) +(g + 4j + 4.j) = (l)rPr(q. .g + 4)(g + 4. 2) 1 2 ' ' ^ ' ^ 1 (g. g + 4 j ) P r . Thus we have. Dingle and H.g + 12) 1 2 3 (g. 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.j)Pr+1(g._1) = Mzil. P2(q 2) Pl(g. g + 4j)P r _!(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 + 4)(g + 4.g + 4) 2 1 (g. J. p = (<?><? + 4 ) ( g + 4 >g + 8) .j). x)_ (g. 0) = 1 and all other P0{q.j)^2(9.g + 4) + 1 1 1 and so on.j). M2 = P 2 ( g . (17. j + 1) (17. and in general* M2r = J]j[P2(g.g + 4 ) ( g + 4.g + 4) ^3(9.j ^ 0) = 0.g + 4)(g + 4.j). From the coefficients (17. g + 4j)P r _i (g. Miiller [73]. and the contribution from a sequence of r moves from q + 4j back to g.4. (17. the factor j removing a duplication of factors in the denominator at the junction. j . W. for instance.g + 8)(g + 8. as jP r 2 (g.
^ . S.^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. 17.284g + 57) + (17. It will be seen in Chapter 18 that in the case of anharmonic oscillators. J.92q3 + 70g2 . Goldstein [115].* We observe that the expansion remains unchanged under the replacements q — —q. Ince [136]. the solutions and eigenvalue expansions again have the same type of symmetries. 17. L. In fact we demonstrate in Example 17.5 the following proportionalities there: B[w(z)] = aA(z). B. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. q is only known to be approximately an odd integer qo = 2n + 1. .4.( 3 ^ . [116]. remains to be determined next. Miiller [73]. *E.3.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 .17. We observe that there are regions. [137].61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. of course. where solutions overlap. W. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. Dingle and H. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. as a result of tunneling between wells. a= (8/i)3(9x) [1(91)]! 1 . h — —h. as pointed out earlier. Up to > > this point.62a) f R .2 * +3) + ^ 2 (9g4 . Its precise deviation from qo.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) .
4 Domains of solutions and their domains of overlap. Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.(17. We write and expand this function as follows: .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . Example 17. Solution: We begin with the solution containing the function A{z). 17.1 4 .32). a ._ ^ l CHAPTER 17.62b) *z domains of overlap Fig. and determine the constants a and a.1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .7T H Aq(z) = = C O S ^ .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.364 and C[w(z)] =a~A(z).1 ) ( 7T + . The first term of the expansion of A(z) is the function Aq(z) given by Eq.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .
2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed. we obtain Bq[w(z)} H. A*1) given by Eq. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa. with e.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.i ) 1 (g.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. Since the solution with A is obtained from that with A by a change of sign of z.g + 4) (q + 4._i)]! w*te» 2l(91 (. (17.92q 3 + 70q 2 . (6. i.61) for the connection and Eq.63) In Example 17.g. These are the conditions of Eqs. (17.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . \ql)/2(w) 2i(9D[i(.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. (8. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.36).17. s . In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73].284g + 57) + • The other relation is obtained similarly.e. the coefficient of the dominant factor c o s ( .53) for the expansion. 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 .46). (17.!)(. 17.2. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2. we have to impose the appropriate boundary conditions. with (8h)li''1) (3g2 .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 (17.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 . As we saw there.22). these are conditions imposed on even and odd solutions at the point z = it/2. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin.
80.62b). With the help of the duplication and inversion formulas of factorials. Magnus and F. (17.22) and to correlate the solutions to those for small values of h2.59d). i.62a) and (17. a ± C[w{z)}c_2hsinz^ a (17. He2n+1 (0) = 0. Then y± oc ^ M e ^ i n . 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 = 0.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. give the following expressions for polynomials with argument zero: ~19)"(^)!.366 CHAPTER 17.» a \dw _ C[w = 0}c_2h a w=0 = Q (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. Periodic Potentials We can extend the solutions (17.66) ^E. W.66). He'v(0) = Hev+1{0).66) then follows from the recurrence relation (6. The second of relations (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\ ! ' . The functions defined as in Eq. p. Oberhettinger [181]. i. . and then replacing 2n by i>.g. one obtains the first of relations (17.e. Thus y+ = ce and y_ = se.63) to the domain around z = ir/2 by using the proportionalities (17.e. He'u{w) = wHev{w)He„+1{w).
B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 . and similarly for C and its derivative.3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.39 2~^h? (17.» 2 6 /i the condition reduces to the equation cot{ . and using the reflection formula (8.65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17. Hence expanding the cotangent about these points.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 .11.68) Inserting here expansions (17.1) 7T .! ) ] ! [ .67a) and (17.67d) Considering now the second of conditions (17.Qo) + 0[(q . (17. the solutions would be given by q = go = 3.7.69) If the right hand side of this equation were exactly equal to zero. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .g0)3].106g .67a) g 106g 87 2 14 /i 2 (17.87 4 2 ¥w (17. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.40g3 + 18g2 .39 (17.82g2 .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g . we have cot j ^ ( g ~ 1)} = ~\<(l .67c) 4 2 \dw)w=Q C[w = 0] [\(q .136g + 9) + • • • .67b) <f ~ 82g^ .67c). .( g .
Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i ..40gg + 18gg .120g£ + 467g .368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17. Considering now the first of conditions (17.408g0 + 1089) 2 /i +• (17. Thus we have 2 (16/i) qo/2 T2 4.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .70) except that now qo = 1.. = _a a 6 _4h (17.65) leads to the same result except for a change in sign.5.65) yields the same result except for a change in sign.9.136q + 9) + .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.e. ..65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T . we use \(q)~\(q0) + (qq0)(^J .72) This leads again to the result (17..136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .(17. (17. . The remaining condition (17.40gg + 18gg .528gg + 3307Q .40g3 + 18q2 .136g0 + 9) + • Clearly the last of conditions (17. i.70) + ^ 2 (9?o .
17. this becomes 369 = (16/i)^+1e4h l . Dingle and H. i. The dominant contributions were also found by S.74b) The above results for the cosine potential are in many respects important. 120. M. though approximate. *In Sec.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. The complete. . + ^ 2 (9<?o + 89o " 78g02 .1. 17.88q0 . 9o = 2n + 1. just as any solvable problem is important for the insights it offers in a concrete and transparent form. Goldstein [116]. See also F.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. We see here that *R. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. B. (17.t From Eq.2. 20. (17. The tunneling effect.3 we call this difference 26E„n. J. p. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. . n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.e.. Arscott [11]. example 3. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics.61).87) = Azpteo).. n = 0.3.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . quite apart from applications. and not the separation of these as a result of tunneling.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . results from the boundary conditions. In the literature that we follow here* in all cases several more higher order terms have been given.e. i. (17. minus sign).A_(g0) ^ .3. Miiller [73]. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. W. and that for the odd Mathieu functions se qo+ i and se9o (lower.. (17.
The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). evident through the exponential factor exp(—4/i) in Eq. in effect. yA{z) = Aq(z)e2hsin '. The splitting can also be obtained with some methods of large order of perturbation theory. (17. (17. The explicit derivation of the level splitting (17. results from the imposition of boundary conditions. y^{z) = Aq(z)e~2hsin *.6: Translation of Solutions For the largeh 2 even and odd solutions (17.2. Periodic Potentials this nonperturbative effect. 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. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). 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. the Schrodinger equation with the cosine potential is a special case of several other more general cases. Example 17. From Eq.74b). which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. M±TT) . . _3/+(z) + ____2/_(z).74b) is also important for various other reasons. The exponential factor represents. Thus we do not reproduce every step. T y'_(0) y'A±*) y{z). y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. As mentioned at the beginning.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. Then we obtain first y±(z) = yA(z)±y^(z).2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . . The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. We begin with the solution containing the function A(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. 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. and one can compare the methods.370 CHAPTER 17. Naturally one expects the solutions there to be related to those of the periodic case considered here. 17.
y'_ (0) = \/2(4h . = ±ir).z ) .1 )  . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .VA DV 371 1 V±. Replacing sines and cosines by the functional expressions.31).y]z=„ = [y+yL . Similarly we find y(z±ir) = = and by putting z = 0: V. etc. but also to enable a brief familiarization with the Lame equation.17.( z ) From this we deduce for z = 0. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A . y+(0)~2V2.T ( ? =F 1) p + (0). From the equations for the sine and cosine we can deduce the value of the Wronskian at z = n in the leading approximation for large ft: W[y+. — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .[A'qM + 2hAq(n)] = iV2(q .4ft) ~ 8(q . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r)..4ft) sin j ^n(q .yV+\z^ = [j/+(0)] 2 (<? .4.2hAq(n)] . A' (z = ±ir) = (. since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).g A ? ( z = ±TT).4ft) cos i TT(<J + 1 ) 1 . ( .q). Aq(z) The derivatives can be handled with the help of Eq.) cos < 7r(q 1) [ = .1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.4ft. where Aq(0) = Aq(0) = V2. (17.Ah) ~ 32ft. (l^ifr^i^O + Cl^ifo^j/aCO cos  . A ?q( Z = ±TT) = .3 The Ellipsoidal Potential Replacing VA.7 r ( g ^ l ) U + ( z ) q : i s i n { .4 17. and y+(z = ±TT) = V2(q .%/2(g . The Lame equation has not been a widely known equation of mathematical physics so .• Then y'+ (0) = 0.( ± 7 r ) = c ° s  .( 2 ) . 17. we obtain the conditions stated at the beginning. which is easiest by comparison with the foregoing treatment of the Mathieu equation.c o s  ~(<1 T 1) f y .7 r ( q = F l ) } j / .
f E .* and the work of both prepared the ground for presentday investigations. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A .fi2A.. L. 17. Erdelyi [85]. M.Q. Tchrakian [165]).. so that Floquet's theorem cannot be immediately applied ..1. Periodic Potentials far. as alluded to at the beginning. the nonoccurrence of the Floquet exponent.0. inverted double well and cosine potentials (J.. W.K2sn2u .0 and .372 CHAPTER 17. p.0. Arscott [11] (p. soon encounters grave difficulties and there has not been developed up till now any general theory of Lame's equation at all comparable with that of the Mathieu equation .0. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.75.* As Arscott [11] (p.4sn4u]y = 0. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). 19.." Fig. 194) remarks to parallels with the Mathieu equation:". Ince [138]. 194) remarks. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. H. k) for k = 0. J.. but also cubic potentials.98.5 The potential sn2(z.. and was continued by Erdelyi. Liang. (17. § F .5. . H. The most conspicuous difference compared with the Mathieu equation is. Arscott [11]. MullerKirsten and D.75) ' T h i s means double well. however.5 < z < 5. But more recently it has been observed to arise in various contexts. This line of investigation. *A.
17. Jianzu Zhang and Yunbo Zhang [206].3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2).77) into Eq. (17. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). In comparisons with the Schrodinger equation.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. (17. If we put fl = 0. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. and of the derivation of the level splitting'.1. In order to distinguish the above equation from that with the Lame potential.2k. the usual factor —h2/2mo. the equation ^  + [A . Eq.K2sx?u}y = 0. W. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k. . we consider mainly the Lame equation. (17. The second step is to insert (17. (17. N = 0.77) where q —> go = 2iV + 1. J.75) reduces to Lame's equation and one writes K.e.e. .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. . Although the results have been calculated for the ellipsoidal wave equation. The range of the independent variable u is 0 < u < 2K.\k\ < 1.K) = qK+^±.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system.4. 2 .17.78) We follow in this brief recapitulation the description in H. i. The function sn2u is plotted in Fig. cmx and dnu. The first step is to write the eigenvalue A as A(q.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. in the case n2 — oo. 17. i. (17. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. K being the complete elliptic integral of the first kind. MiillerKirsten.5 for several values of fc. . in front of the second derivative has to be kept in mind (mo being the mass).2 = n(n + l)k2. is the elliptic modulus of the Jacobian elliptic functions snu.
the other in terms of those of an imaginary variable.K — > A(u)=A(u + 2K).e.e. /dnu + kcnu\ \ dnu — kcnu J For large values of K the equation for A(u) can be solved iteratively resulting in an asymptotic expansion for A(u) and concurrently one for the remainder in Eq. one pair in terms of Hermite functions of a real variable. i. C replacing yl.82 . Thus in the third step two more pairs of solutions B. Periodic Potentials .81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). . \/8K (dmz = cn« \ ' F . _ Solving the resulting equations iteratively as before.80) The domain of validity of these solutions is that away from an extremum of the potential. (17. determine their proportionality factors) in domains of overlap (their extreme regions of validity). A(u. A into equations in terms of the variables z(u) = ^—{k' .77). one has to derive new sets of solutions there and match these to the former (i. (17.. A are derived. dnu± cnu K Thus one can construct solutions Ec(u).K)=A(u. but solutions B. A second solution is written y = A(u) exp < / nsnudu >. (17. C which are valid for dnu±cnu <1.K). k' = Vl~ k\ \dmi±cnu/ 17. .Es(ii). by transforming the equations for A. B and C.e. more precisly for dnu = cnu F 1 » .q. ^ (17. i. / z J . one obtains again the same expansion for A. A. which are respectively even in u (or snu) or odd. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.q. dnu =F cnu .79) The very useful property of these solutions is that for the same value of A (which remains unchanged under the combined replacements q —>• —q.374 where CHAPTER 17.
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.38g2 .83) Then Ec(ti) Es(it) KB[fM^(u)r/2fc±C{zH] a uy n a n S/Ms Each of the solutions thus derived is associated with one and the same expansion of the eigenvalue A. who obtained this expansion for the eigenvalues of Lame's equation (i. Es(2f0 = Es(0) = 0. i.384n2k4(q2 + 1)} q r {(l + A. Miiller [205]. (17.^ ^ { ( 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£." 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) . W. Tricomi [87]. the ellipsoidal equation.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. Magnus.63)} + ••• . J. (17. 64. £2 = 0). W. a of _ _ B = aA. C = a~A. p. *A. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. as well as \duJ2K 11 (17. .e. 17.17. G.4.86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. (17.85) The first three terms of this expansion were first given by Ince [138].e. F.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . Oberhettinger and F.6(5g4 . Erdelyi.75).
2K and 4K respectively.89) one obtains the eigenvalues from which the level splitting can be deduced. J. W. One obtains with go an odd integer QQo =F2 2fl + ir\lkj 3(g02 k\K/k 8K qo/2 \lk ) 2 1 [i(g0l)]!L + l)(l + fc2) 25 K ^ 2 J J . 2K.j 26K2 (1 85) = A(g0) + (q .:J {3(1 + k2)2 (9q* .88g0 .i Here the upper sign refers to Ec*> l or Ec£°.376 CHAPTER 17.g0 is obtained by expansion around zeros.4k2(q2 + 2g0 + 5)} + • • • ].1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .Es^ 0+1 .g 0 ) ( ^ . .91) tt H.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)*. from which the difference q .78g^ .Ec£ 0_1 and Es2° of periods 4K. 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). and the lower to Es*>+1 or Es*5. go being an odd integer. Finally expanding A(g) ~ A(g0) + ( g .40g03 + 18g02 . Muller [205].136g0 + 9) n 3.90) + 3. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ . Periodic Potentials These conditions define respectively functions EcX°. (17.87) (17.go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .l n 2^+ i J + 0(«i) (17.2 /c +256fc2g0(g2 + 5)} (17. 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.
* Apart from the choice of notation. Thus in the case of the dominant contribution of Eq.Ah2 sin2 u}y = 0.74a). (17. Hence the conditions fi = 0.87) agree with those of Ince.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./2i and hence A^ A/ A AU I2 . in which they dubbed the classical configurations Lame instantons. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite.* M G . 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. Replacing u by x ± IT/2.4 h ( i e M 9 0 7 2 \. A . this equation becomes y" + {A . (17.90) one has in this limit: /l±±\ ~K/U = e . 17. that calculations with the Schrodinger equation are not only easier.2h2 = A. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. One can see. + E. Miiller [206]. (17. V. . A = 0.2h2 .93) (17. Dunne and K. Ince [138]. L.17.e.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. *H. it = x ± ^ .4. but also more easily generalizable. the conditions (17. J. K2 = h2 (say). W. however. K = 2/i. i.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation.J1 O. (17. Rao [79]. . then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . h .2h2 cos 2x}y = 0.86).f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«.
Quian [123]. D. Sukhatme [148]. J.g. 1 1 See Z. i.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). J. W. Khare and U. J . See Chapter 25.Y.W. de Veigy and S. 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.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. H. inverted double well and cubic potentials. Kan and K. Tchrakian [165]. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. A host of related elliptic equations has recently been discovered and studied. Thus. Rana [287]. H. A.378 CHAPTER 17. § . In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. Cho. Periodic Potentials 17. H. This is an interesting problem.^ The equations considered above also appear in diverse new problems of physics. See e. Gu and S. Ouvry [276] and Jianzu Zhang.Q . S. **Y. Shiraishi [49].e. A. double well.5 Concluding Remarks The potentials we considered above will reappear in later chapters. Ganguly [103]. Liang. W. A generalized associated Lame potential has been considered by A. for instance. all basic potentials. 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. M. N. . MiillerKirsten and J. MiillerKirsten and D.
demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. M.Q. Wu. f C .. Liang and H.Chapter 18 Anharmonic Oscillator Potentials 18. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Achuthan. MiillerKirsten and A. Wiedemann [4] and a revised version of parts of this reference by J. Bender and T. T.* The fact that a main part of their work was concerned with the calculation of the imaginary part of the eigenenergy in the nonselfadjoint case which permits tunneling.. MiillerKirsten [163]. [19]. D. T. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. J. Weinstein [281]. The recent work of R.". H. This seemingly simple problem revealed extremely rich internal structure . both of which make the calculation more difficult. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. 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. W. W. In particular the investigations of Bender and Wu.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. Bender and T. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. 379 . Wu [18]. ' T h u s A.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. J. Priedberg and T. In the cases treated most frequently in the literature the anharmonic *We follow here largely P.
described as the case of the double well potential. V(z) = \\h*\z2 I 2i 4 \cr\z . and thus lead to very different physical situations. (3) Complex eigenvalues with tunneling: In this case. These contributions may be given different signs. *~z (1) (2) (3) Fig. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . V(z) = \\h*\z* + \\<?\z*. 18.1 The three different types of anharmonic potentials. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation.380 CHAPTER 18. with the potential described as an inverted double well potential. 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. 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.
Since these exponentially small contributions are related to the behaviour of the late terms of the eigenvalue expansions (as we shall see in Chapter 20). The shift of the eigenvalues is best calculated with straightforward perturbation theory. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17. except for a change of sign of c 2 . The result is an expansion in descending powers of h?. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). . 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. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). The imaginary part will be rederived from path integrals in Chapter 26. (18. The eigenvalues of this case are given by Eq. Case (3) is seen to be very different from the first two cases. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. that the general applicability of the method becomes evident. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. however with decay as a consequence of tunneling. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. as is sometimes hoped.1 Introductory Remarks 381 normalisable wave functions.e.175) below. 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. since the potential decreases without limit on either side of the centre. We begin with the latter. it will not be possible to obtain the exponentially small contributions with convergent expansions.86) below. The eigenvalues of this case are given by Eq. Thus we are mainly concerned with the double well potential and its inverted form. The level splitting will be rederived from path integrals in Chapter 26. 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. In this case our aim is to obtain the aforementioned complex eigenvalue. (18. 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. i.18.
(18.2 The inverted double well potential with (hatched) oscillator potential. a point which has to be kept in mind in comparisons.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.1 and more specifically in Fig. 18. h 4 2 2h1> ~1/2 2c?. 18.3) "1/2 harmonic oscillator h8/25c2 . we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei. If suffixes 1/2.2 18.1 refer to the two cases. v(z) = ^h4z2 + ^c2z\ (18. Anharmonic Oscillator Potentials 18.1) for h4 and c 2 real and positive. . (18.2. 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 .382 CHAPTER 18.1 The Inverted Double Well Potential Defining the problem We consider the case of the inverted double well potential depicted as Case (3) in Fig. The potential in this case is given by V{z) = v(z). 18.z Fig.2. We take here h = 1 and the mass rriQ of the particle = 1/2.
and a variable w defined by setting E= b* 1 + A A and w = hz.2. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. \h?\ — oo and c finite. The result will be that derived originally by Bender and Wu. (18.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). 4 . /i).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. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. (18. (18. i. (18. y{w) oc Di. The problem here is to obtain the solutions in various domains of the variable.1. n = 0.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. (18. The positions z± of the maxima of V(z) on either side of z = 0 in the case c2 > 0 are obtained from h2 v'(z±) = 0 as z± = ± — with (18.5) becomes .18.d2 . to match these in domains of overlap.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite. 18.... _i\(w) and q = qo = 2n + l.2. the harmonic part of the > potential dominates over the quartic contribution and Eq.e.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.1/7  Vg(w)y(w) = o(J^j.4) we can rewrite Eq.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small..
Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. (18.11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) .10) E= qh2 ^ j. We observe — before touching the square roots in Eq.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18. 1 d r A!{z)±iA{z)±{ 0.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.2 We are concerned with the equation d2y(z dz2 where ^ 1 .9) we obtain 1 h4z2 c2z4 A L2 (18.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.4(2.14b) . be and A = A(q. This observation allows us to define the pair of solutions yA(z) = . as encountered and explained earlier. h) is obtained from the perturbation expansion of the eigenvalue. (18.10) into (18. In order to arrive at solutions we set in Eq.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^.9) z =0 these Thus these c2z4 1/2 + • 2 (18.+ ^ .2. these solutions are not valid around z = 0. Inserting (18. Before we return to solutions we derive a new pair which is valid in the adjoining domains. •A~2 hrz" 2^\ c^z V2 —— + (18.384 CHAPTER 18. as we shall see.2 A (18.14a) 1/2 VA( ) = A(z)exp (18.
we do not pursue their calculation. Since these higher order contributions are of little interest for our present considerations. 1 2 4 2 Aq(z) + 0 (18.16) For large h2 we can write the Eqs.18. Clearly Aq(z).34) below) — to be derived in detail in Example 18.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.Aq(z) approximate the solutions of Eqs.zh2 =()*2"(18.19) We see that one solution follows from the other by replacing z by — z.4 .1 and as a verification again in connection with the solution y# — as the other solutions. i.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials. (18.20b) .13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z). One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. (18. 1 (18. We take the square root by setting z2h^1/2 +.15) where A{z) is the solution of the upper of Eqs. / c z Aq{z)+0 1/2 . (18. (18.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. /H 4 Z h +~C Z 21.. (18.e. 385 (18.13) and A(z) that of the lower of these equations.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.
. * = — .1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. A = . (18.386 CHAPTER 18.18) we obtain A » = H^rL^z) 2 z . i.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q .2 .e. .f . ft = . c*i = 1 1 . A A _ 3 . z) ± yA(q. to obtain in leading order the eigenvalues E. (18. we write > V±(z) = .:=*£ +i(5l). 04 = 5 . h2 —> —h2). 03 = . h2.e.! ) ^ .20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.l)A(z) 1 = tfA"(z) A . h2.. (18. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs.e.[yA(q.17) and (18.20b). Anharmonic Oscillator Potentials i.20a). 2 . „ . As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z). procedure.. i.1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"^ The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z).^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl.2 ( * ) 2 and from this or separately A"(Z) : 1 (q .21) Example 18.• Using Eqs. away from the central minimum. W i t h proper care in selecting signs of square roots we can use the solutions (18. (18. X » / A q + 4 i = Mg+4i . Vq+ii = Vq + 2i.. (18.. 1 . zA'(z) where the expansion coefficients are given by ao = l.20a) and (18. z)). one observes that with £>. a2 = 2 1 . 128 8 16 A = i.v ' = o (« .
.22) is invariant under the combined substitutions q — —q. Rq '.5). (18. when i = 0. i.(q + 3)(q + l)Aqvri(q '^ ' .'2H9i) ^faDfr™) and . (18.23) ..Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. (18. give an equation from which A is determined.22) are parabolic cylinder functions D i . c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. +1J±iw) (observe that Eq. = •>— r. r ( CJw) = ^ .w — > > ±iw) or functions R B w q( ) . Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ ..18.—77 [i(9l)]. and the coefficient of terms with i = 0. of course.e. In this way we obtain the next order contribution A^1' to A^°>..2 The Inverted Double Well Potential Thus a term /j. ^f( g + i)(±H2^ + 1 > \ — 1. those in Aq(z). 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). 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. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . [1(9 + 1)]! . _1J±w) and D_\. (18. The solutions yq(z) of the equation Vq(w)yq(w) = 0. We return to Eq. w = hz.
As an alternative to Bq(w) in Eq.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 + .^ For higher even powers of w we write i w2l V<i = Y. and for j ± 0 : [q. Vq+4j = Vq + 4j.3.29) ' A . and so Vqyq+Aj(w) = Ajyq+Aj(w). (18.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.24) The extra factors in Eq.3)y g _ 4 .Aj)(18. (18.0) = = = ±(q±3)(q±7). (q±2)(q±S). S2i(q.±8) 5 4 (g.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.28) (A + c w )yq(w) E 2 4 1  2 ^ [q.388 CHAPTER 18.( < ? . (18.q + 4j] = c2SA{q.l)2/?4.±4) S4(9. Actually these factors can also be extracted from W K B solutions for large values of q.q + 4j}yq+4j. Magnus. pp. 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 . G. .4j)yq+Aj. See Example 18. (18. F.0). W. since Vqyq = 0.27) Now. (18.q}=A + c2S4(q. l(q2 + l).123.25) where in the case i = 2: SA{q. (18. Comparison with our notation is easier if this reference is used. j=i (18. Tricomi [86]. we also have Vq+^yq+tj = 0. Erdelyi. Oberhettinger and F. 115 .
. h2 —> h2. (18.3 °) For the sum y{w) = y(°)(«. ( 18 .) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.q]=0.10) we obtain E(q.e.[rp ) Y.e.1.—v*+*j ( 18 .) + j. We can now write the solution y(w) in the form y{w)=yq{w) + Y. p i(9»9 + 4 J'WK7». (18. so that the entire expansion is invariant under the interchanges q .35 ) . .^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.32) Evaluating this expansion and inserting the result for A into Eq.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j. . (18. i. Thus the next order contribution to yq is y H = ie 2^ —z.• q.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. h ) = qh .) ty^^w) to be a solution to that order we must also have to that order [q.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. n — 0.31) Proceeding in this way we obtain the solution y = yW („. and even powers of q in combination with even powers of 1/h2.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. 2 . This type of invariance is a property of a very large class of eigenvalue problems. Equation (18. (18. (18.18.—A(q + 1) .(!) («. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. A = ^(q2 + l)c 2 + o(J^. and c 2  replaced by — c 2 . i.
argz=0 (18.g±4] ±4 ±4 [g. Wiedemann [4] .37) For further details concerning these coefficients. W. q + 4j) = 0. q) = = 1. g ± 4 ] T4 ±4 and so on. their recurrence relations and the solutions of the latter we refer to the literature.390 where for instance v J CHAPTER 18.—. H. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. We thus have the following pair of solutions VB(Z) VB(Z) = B w . ? T 4 ] [ g T 4 ." Since our starting equation (18. We observed earlier that these are obtained by making the replacements q —> —q. ^VO(<L<?) ^i(?. > "p. w — ±iw. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. Achuthan. o. J. Again we can write down a recurrence relation for the coefficients Pi{Q.i 1 + 4j) in complete analogy to other applications of the method. and for j ^ 0 all other P0(q.)]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.11) is invariant under a change of sign of z.36) with the boundary conditions p o(q.38) [yB(z)]argz=7T = [?/s(2. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q.e. we may infer that given one solution y(z). there is another solution y(—z). i. MiillerKirsten and A. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.q±A) = MM!i±MM+[M±8][g±8. (18. 2 4tPi{q. q + At] (18.
39) are with the same coefficients Pi(q.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored. see e.z2 .40) "Variables like those we use here for expansion about the minimum of a potential (e.g. the solutions of type A being valid away from the minimum.g.2. Mauss [192].5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. (18.Vc suitable asymptotically decreasing expansions in one of the domains \z\ < O 7T argz We emphasize again that all three pairs of solutions are associated with the same expansion of the eigenvalue E(q. h2) in which odd powers of q are associated with odd powers of h?. Integrating and expanding as follows since h? is assumed to be large. J. with normalization since the normalization constants are also asymptotic expansions.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. (18.g.h^ih. In order to be able to extract the proportionality factor between two solutions.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . . 18. Thus in the transition region some become proportional. in fact. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. 4 . we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0.h^ih 391 (18. Vc(Z) = [V!B(Z)}q^~q. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. Such contributions arise. one has to stretch each by appropriate expansion to the limit of its domain of validity. like w of Eqs.4)). In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. Aq(z)) does not appear in any of the higher order terms.q + 4j) as in ys.18.The solutions yc.
eh* /12c* e\z*h? (18. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . The function Di.(«. (18.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18.0x ! [ £ ( g .7 r . > n.w yB(z) ~ B.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.l ) ] ! ( .42) H W2 (ql)e$W I „ . Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.42) we obtain for the solution ys(z). 2 ) i ' ir! i iHim 1 3 j argw < 7T. . Comparing the solution VA{Z) of Eq.4 i .2 « . We do not require this at present. with z in ys(z) replaced by — z. we would have to substitute correspondingly the expression (18.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\.41) with the solution ys(z) of Eq. 2 ^i![i(g4il)]!(2«. 4/T 4 Prom (18. 41) The cases of the solutions of types B and C require a careful look at the parabolic cylinder functions since these differ in different regions of the argument of the variable z.z). y^(z) we see that in the direction of z = 0 (of course. (18. Anharmonic Oscillator Potentials Considering the pair of solutions ju(.392 CHAPTER 18.45) .. (18.43) W + 1)]! w^+V A .43) (since z — — 2 implies > argz = ±7r).44) In the solution y~g(z).
50) and y+(0) ^ 0.46) However. (18.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18. Thus the boundary conditions to be imposed there are the same as in the case of the harmonic oscillator for alternately even and odd wave functions.50) will be seen to imply qo = 2n + 1 = 1. and the second qo = 3. hence we have to .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 h2.7.9. The first of the conditions (18.. (18.yc{z)pansion (18.41). Recalling the solutions y±{z) which we defined with Eq. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18. the ratio of yA(z)iVB(z) i s n ° t a constant.e. (18.49) d yc{z) Again there is no such simple relation between yA(z) 18. We proceed similarly with the solutions yc(z).y'_(0) ^ 0.42) into y~c(z) w e obtain Vc(z) = (h*. At z = 0 the solutions of type A are invalid.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .5.18. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. we see that in their common domain of validity yA(z) where a = =Vc(z).1 ) e 12c 393 Kgl)]^"1) l + O h? (18.11. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.. Looking at the potential we are considering here — as depicted in Fig. i.21) as even and odd about z = 0. large probability for the particle to be found thereabouts.2..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. 18.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.
y c ( 0 ) . with the reflection formula (—z)\{z — 1)! = IT/ simrz. Example 18. M.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.i ( g + 3)]l (18..^{(9+1)}.F ? .fc(0) Thus we obtain the equations (18.2: Evaluation of yB(0). Expressions for C 9 (0).s i n { . + 1)]!' D ' j .55) Solution: Prom the literature.l)(°) Thus with the help of the reflection formula cited above: D B.54) dio q(w) . [C. Abramowitz and I.(«. [ . one finds that [ 1 IT (q + 1)1! = .56) D 4(.( g + 3)}. Stegun [1]./o^Tpif (q+i) . We leave the detailed calculations to Example 18.2.l ) ] ! [ .(0) = and htoi)(°) ^—'.(0) (18. e.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.(iu)]..394 CHAPTER 18.(0) = . ( 9 + 3)]! s i n { .(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}.5 ( 9 + !!)(*"')• i ( 1) .SxW) = \ i»fl(0) . = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1. C.' '.57) [i( 9 _l)] ! 2 i(9i) B.g. y'B(0).58) diu [  ( g . IT 4 dio In fact.oi(. A. (18. — C.i ) ."*(«!) ^ ' ~ [ .. 1 [7(93)]! n «(°) = . Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.! ( . Z (18.) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.i)H = e.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.1)('») + ^ r + l)]! ^.( g + 3)}. we obtain i V7T24( g . .51) and 0 = V(0) = Jim i»xM .(0) + ^ ( 0 ) l z—>0 2..
7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.sin{^( 9 .49). (18. ~ ^ 1 5(9+1) 2i(''.3)}. (18.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18. A __f7r \4 'J We rewrite the right hand side of the derivative equation again with the help of the inversion and duplication formulas and obtain _ (^)g/4(_l)j(g+l) e /> 8 /6^ _ p (/i4)g/4(_1)(g+l) _^_ . isin{f(g + 3 .2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.i ( « + i ) [ _ i ( . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .56).i . W .+1iH]«=o = . Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.6 ) } ^ — i t a n < — (q + 3) >.(0) = *>i(„+i)(0> 2 . _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„. we obtain 2 i(»+i)[_i(. (18. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.18.+i)(°) [ _ I ( ._l) From this we derive C.46).)[i(g3)]! and ^i(. 4 r T (18.53) in dominant order and insert the appropriate expressions for a and a from Eqs.62) .1/2)!.sin{^(g+ 1)} = J .
_ e =*. .64) for q = g0 = 3. the vanishing of the wave functions at infinity). ..64) about go — 3. 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. Thus we have to determine these conditions first. With a Taylor expansion about go 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.5.1. (18. 7.5... 7 .64) the right hand side is an exponentially small quantity. we again obtain (18. (18. + 3)} = .. Recall > the original Schrodinger equation (18.go) ^ (1) It follows that we obtain for the even function with q — go = 1.53) becomes 8 m(I(. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.( g 0 + 3)  + • • • ~ (g . i.53)..^ .. equivalently. (18.63) becomes (? . we obtain cos((g + 3)) = .. . ^ .^ >^_>m e J*.L ^ i . (gg 0 )^±^ y e5?. 18.65) Expanding similarly the left hand side of Eq. and the left hand side of (18. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or.2.5 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.2) with potential (18.i . This is Case (1) of Fig..9..63) and (18. (18. In fact the left hand side of (18. the energy E is real.1). by the rotations E > einE = E. . z2 > z2.e.396 CHAPTER 18. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.63) Proceeding similarly with the second of relations (18.. 1 1 . (18. (18..65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0.11. 18.63) vanishes for q = go = 1.9o) ^ cos  . . z^ ein/2z.9.64) In each of Eqs.
Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. Now.e.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations.cos 36.e. y{z)~exv\i[) 7T dV'V. i. the resulting wave functions vanish at infinity and thus are square integrable. in the domain — TT/2 < 36 < > 7r/2. i. —}. r / c 2x 1/2^3 = exp{ ± i — —\.68) . replacing sin 30 by " 0 + ^) 1 IT = . we set z w r = \z\eie. i.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. 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. 1/2.18. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. 2 c2>0. i.e. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . 6 6 In the case of the inverted double well potential under consideration here (i.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. if TT/3 < 6 < 0. (18. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. (18. or — < 6 < .67) Rotating z by vr/2. the case of complex E).e.66) z3 = \z\3em. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s.e.
3 The inverted double well potential with turning points ZQ.21)) were defined in terms of solutions of type A which have a wide domain of validity. Our procedure now is to continue the even and odd solutions (18. Our even and odd solutions y±(z) (cf.68) for c 2 > 0. We therefore explain our procedure first.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.e. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. (18. we match the WKB solutions .Z\. i. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. We have to remember that we have various branches of the solutions y(z) in different domains of z.Z\. and there impose the boundary condition (18. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. We do this extension with the help of WKB solutions.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). 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.21) to + infinity and to demand that they satisfy the condition (18. This is the boundary condition also used by Bender and Wu [18]. Anharmonic Oscillator Potentials V(z) Fig. Looking at Fig. (18.398 CHAPTER 18. 18. Eq.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. for z > Too. 18.
h2 (18. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.2. *R. i.^ T rz\ 1 + yh* . Dingle [70]. I [195].3 is given by (using Eq. (18.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . The distant turning point at z\ as indicated in Fig. 18.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. ^ 2 4 •K + —(r.70). We begin with the exponential factors occurring in Eqs.18. A J24 1 ? 2 • "^ •z h \—c z ~ — q h . z2h4/A > c2z4/2. B. (22) or A. 6. equations (21).4 h + c z 2 1/4 < .^ 4 1/2 + icV 7T Jz + (18. f R . Messiah. (18.71) We now come to the algebra of evaluating the integrals in the above solutions.. p. Dingle [70].*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . 291. ^2? 1 2qc2 /2? .zi)/ \ 1/4 ywKB\ ) z — g/i* . p.B. 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\. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (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.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.e. 291. Vol.70) where z < z\. z\ ~ .4) for E and ignoring nondominant terms) t o . . Sec.69) The W K B solutions have been discussed in Chapter 14.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp .e. t To the right of the turning point at z\ these solutions match on to (r. 4 2 T ' i.4.
i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). Anharmonic Oscillator Potentials E± = exp = exp . CHAPTER 18. Thus the above factor yields '2c2\l/2 h4 so that (cf.^ I 2 2 1/2 zV = F 8c2 Jz [i .2..2 \ 1/2 .40)) E± = exp _/^_2 f ± In A 2c2 . (18.69)) rzi J<z z dz Z[&Z*]W = + ^Y /2 m '"U4 1/2 2d2 + 2c2 Since we are interested in determining the proportionality of two solutions in their common domain of validity we require only the dominant 2dependence contained in this expression.2. (18.211/2 1 \ /j.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c. Eq. (18.^ i ^ ^ c 2 . zi 2q 1 c 2 [1 .±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. (18.73) .40)). .\qh2 + \z2h± .4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq. Eq.400 i.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.e.
78) Inserting this into the solutions (18.18.74) Comparing these solutions now with the solutions (18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .20a) and (18.*l) (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q. > » Returning to the even and odd solutions defined by Eqs.(J*i) VA 0 ) = PVwKB (Z C.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. (18.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.\z2h4 + J Z\ nl/2 \M fZ .76b) apart from factors [ l + 0 ( l / / i 2 ) ] . h2 — — h2 is maintained.20b).77) Now in the domain 2 — oo we have > \qh2 .*l) (3y^(z) (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.76a) P P r 2/1 2 " 9 ( .77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) .Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.79) +S_(±)exp . CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.71) and these into (18.21 1/2 .1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.
85) + . (18. E(q. (18.. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .68).2. i.e.^ § ( 4 g 2 + 29) + O ( ^ ) .81) Inserting expressions (18.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.82b) This is our second condition along with Eq. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J ./ ? = 0. (18. we see that we have to demand that 5 + ( ± ) = 0.h2) + (qqQ)h + (18.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq. h2) = \qh2 . (18. (18..e. i / 3 ± . (18.. . 18.h2) + {qqo)(^pj = E(q0.6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.. (18.84) Expanding about q = qo we obtain £(g.402 CHAPTER 18./i2) = E{qQ. (18.76a) for (3 and /3.^(q2 + 1) . i.3.34).65). Eq.5..^ j .
4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z. ft6 ^ = e.18. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1.g. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18.~p{q.3. h2).h6/6c2 E = E(q0.yA{z). h2) of the relations KWKB (Z) =PVA(Z).36) of their Ref.g ( . * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. above) (1. i. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0. where the superscript {r.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. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z). Example 18. !*WKB(Z) ~ 2 .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o . 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. Since this is not without interest (e.e. o 2 + l)^(4. zo} means "to the right of the turning point ZQ" . S W K B ( 2 ) = PVA(z).2 .ZQ) .86) will be dealt with in Chapter 20.2 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 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .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.l ) ] ! > + = ^ 2 .
zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . N. 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./ ^ 2 X 1 / 2 exp(ifi. 2 z 2 ) / 2 e \ ." / 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. 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.g In other words.404 CHAPTER 18.ic224 2 4 qh2 + i^fe 4 .] { hz combined with the particular constants contained in y ^ ' ^ g . Anharmonic Oscillator Potentials where zo > z. 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. the (approximate) solutions* 1/2 exp • ± i ( . cf. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .O. f x Zn 4) h2 2.zo)/ \ qh2 2 ZO + z2hA 4 2 ..1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation . . Schwind [247]. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20.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.
But there are significant differences. as in the case of the Mathieu equation. 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.5 "l _ o f .kyil(.18. Then . in Eq. (18. We can reexpress these relations with the help of Stirling's formula.(r. i. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .13) of Bender and Wu [19].^ 1 .88) + [EV(z))y{z)=0 (18. we shall employ basically the same procedure as above or.J ' i J .DfJ»v we see Comparing now the expressions (18.e.yA{z) with expressions 27WKB i J'WKB ' that the factors p. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. (18.g.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.z ± ) = .89) . „ / M 1 / 2 i . (j ._ . h4 > 0.3." / 4 ( ? « / 4 .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.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. in fact.1)! = ( 2 7 r ) 1 / 2 e .. 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.23)) appear quite naturally. 4" ) (h?z2)^q+1) for q j£ odd integers. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.41) for yA(Z).zo). (18.3 18.
(18. (18. (18.V(z±) l{z z±fhA + 0[(z . vW(z±) = ±Qch2.94) (18. (18. . In order to obtain a rough approximation of the eigenvalues we expand the potential about the minima at z± and obtain y dz2 We set + E .93) w± = h±(z .z±). 18.90) V(i)(^)=0.95a) .i>5.4 The double well potential.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^.406 CHAPTER 18.91) (18. the previous equation (18.z±) y = 0. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.z Fig.91) becomes T>q±(w±)y(w±) = o(^)y.
Eq.i 4 f W » = o.93) for E into Eq.. n = 0. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].3. 18.100) . Inserting the expression (18..2 T h r e e pairs of solutions We define our first pair of solutions y(z) as solutions with the proportionality y(z) = exp \h\ f uV2{z)dz (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.. we conclude that in the dominant approximation q± is an odd integer. is again seen to be invariant under a change of sign of z.e.18.96).95a) and (18. given one solution. qo = 2n + l.2.95b) with the equation of parabolic cylinder functions u(z) = D„(z). (18. (18.. (18. we obtain another by replacing z by —z.97b) Our basic equation.1.87). Thus again. (18.99) V2\ZZ The equation for A(z) is given by the following equation with upper signs and the equation for A(z) by the following equation with lower signs: A'\z)TV2Lz2I^\A'(z)TV2czA(z)+(^q±h ±+^r)A(z) = 0.97a) U(z) = (zz±)2 + 0[(zz±)3}. d2u{z dz2 + v+ 2~r u(z) = 0. i. (18. (18.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.96) dz2 + 5 « 4 +  . we obtain d*y (18.95b) By comparison of Eqs.
(18.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18.yA(z) are associated with one and the same expansion for A and hence E.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .e. (18. Aq(z) = Aq(—z) = Aq(z). ~h2. (18. 8hw (18.101b). h2 h4 2c' with g+ = q. Vq = {z2+z2)— + (qz+z). The result is given by A and E(q.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. Integration of Eq. z) = yA{q.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. h2. h2.103) Looking at Eqs. z) may be obtained from the solution yA{q.106) . (18. (18.101b) To a first approximation for large h2 we can neglect the right hand sides. h2.104) Both solutions yA{z).c W + 1) . i.h2. z).408 Since CHAPTER 18.101a). we observe that the solution y^iq.^ A"{z) + 2c V2 *A(z) (18.h2) .^q(l7q2 6 4/i + 19) + (18. z) = yA{~q.105) "8 + 4 ^ .4. but the solution is a different one.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). these equations can be rewritten as + (qz+z)A{z) = .1 6 ) . i. We leave the calculation to Example 18. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0.e. yA(q. n± = V2h2. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.
Solution: The structure of the solution (18. The first such step would be to reexpress the right hand side of Eq. First.4zz+q+ z\{q2 + 1)].2 z\) • Aq$4 £Aq + Aq—4.qz+)2 .*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. however.e. + Aq+2 + Aq Aq+2 . (18.jf2i(2) We also note at this stage the derivative of the entire solution yA(z) taking into account only the dominant contribution: yA(*) =* A*2 V2 . <4q+2 — z — z+ Aq+2iAq+2j (18.101a) with A replaced by A .{z2 .102). = Aq+2i+2jAq. .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.110) Aq+2+Aq2 9+2 A. (18.3 The Double Well Potential 409 Example 18.+ 2 ^ ± l + 2^±i + . (18.112) From these we obtain.z\) + 2z(z .Aq 1 + 2^± 2 Aq+2 (^M^f (18. We know the first derivative of Aq from Eq.109) We observe some properties of the function Aq{z) given by Eq. Thus it is natural to explore analogous steps.(l + + 2  z y* .. A Differentiation yields » = ^^w (18.qz+) [2z2 . Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.108) We wish to rewrite this expression as a sum y coefficientiv4.103): . (18. (18. such as periodic potentials.106). .103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. . ± 1 . .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z . ± 2 . i. . as a linear combination of terms Aq+2i. 92 ZZJf (18. . (18.z2 ) N t.114a) .105) and (18.18.111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . for instance by componendo z z et dividendo. J = 0. we reexpress A'q in terms of functions of z multiplied by Aq. z • .113) 2 .
Since VqAq = 0. (q.e.108).4]j. A(°~) = Aq.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.4)A„_ 4 + (q.117) 4(q + l)2Aq+2 Here the first approximation of A.101a) the contribution R .(0) V2 ' 2c K + h4+Ao .q2)Aq2 + (q. leaves uncompensated on the right hand side of Eq. (18. i. .2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.34). _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4. .4Aq+2 + Aq+A Inserting (18. q) = 0.119) It is now clear how the calculation of higher order contributions proceeds in our standard way. (18. 1_4^±^+8^±±12^±^ +.2A g + A g _ 4 ] . we obtain 5 K = ( ^ f ) [ ( 9 .l ) 2 A .q + 6)A<.2\2 (z* .3 ) A .q + 2) = 4{q + l ) 2 . •••]. (q.q)Aq + (q. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18. cf.115) (z*z%) 2 i2 (z 2 .<? +{q.118) 2s/2c ~^j[(?.33) and (17.116) into Eq. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.410 and from this = Hence with Eq. _ 4 . (18. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .114b) . In particular the dominant approximation of A is obtained by setting (q.q + 2)Aq+2 (18.aAg+S 12+ 16 (18.112). QA?+4 _ 10 A + 6 . (17. (18.4 ( g . q + 4)Aq+4 + (q.4A ? _2 + 6Aq .115) and (18. (18.9=F4) = (g=Fl)(?=F3).l ) ( g .112) CHAPTER 18.+6 + • • • ] . Anharmonic Oscillator Potentials A 1 . where the lowest coefficients have been determined above as (9. (18. Eqs.4 4+* + .116) [Ag4 .2Aq + Aq4] 1.
g) 4z+ {q q)+ 2r (9.g) Az.123) ^[yA(q.h2.h2. y'_(q.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.122) The solutions yA(z). (18. y'+(q.g ( 1 7 g 2 + 19).Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9. i. A' (0) = q/z\) y+(q. z)\ = 2^A^ h2 '>z)± VA(Q. g .e. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.h2.4 ) ( 0 ) ( g . 9 + 2) Aq+2Az+ (9.(0).e. thus yielding the next approximation of A as given in Eq.124) .0)=0.g2)(g2.105). where Rq is the sum of terms left uncompensated by AW.h2.z)±yA(q. for i = 0. h2 2y^ Age? h4 (18.{^ h2. *)] (18. (g.h2.94)(g4.94) 8z+ (9. g .g + 2)(g + 2.0)=0. of course. h2. Terms jj. g + 4)(g + 4.101a) the contribution Rq .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.g) 8z+ (g. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. (i) _ / V2c\ [ ( g .121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g. in the domains away from the minima. 9 + 4) ^+48z+ (18. g) 8z 4 .yA(z) derived above are valid around z — 0. (18. (18.2 ) (0) (18.0) = y^(q. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.0) = ^ . i. h2.28.
yc(z) with complex variables and Cq(w) given by Eq. but w±(z±) = 0.126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. and another VB(Z) ± 25^ch3wl + c2wi . 18. Inserting (18.91). (18. yB(z). (18.2A y(w±). (18.1 ) D (18.125b) It is clear that correspondingly we have solutions yc(z). in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum.126b) These are solutions again around a minimum and with yB(z).412 CHAPTER 18.3.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z). Anharmonic Oscillator Potentials Our second pair of solutions.We draw attention to two additional points. (18.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). 2(9+) .23) with appropriate change of parameters to those of the present case.yB(z) is obtained around a minimum of the potential. Moreover. Bq[w±{ ^ .99) for z — z±. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).^ % [i(?l)]!24(«. we > . We see already from Eqs. Since w±(—z) = —h±(z + z±). + 0(hl2). we have w±(—z±) = —2h±z±.Evaluating the exponential factor contained in VA{Z) of Eq.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). This means.95b) that the solution there is of parabolic cylinder type. (18.95a) and (18.93) and setting w± = h±(z — z±). (18. the equation is — with differential operator T>q as defined by Eq. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq.
and comparing with Eq.147)). . .^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. (18.172) obtained later.z±)dz .ZI and find that these are given by z+ + 0(l/h) for finite q (cf..172)) we add here comments on this approximation. .3 The Double Well Potential (cf. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . 24(9D[i(gi)]! hj (18. 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. Later we calculate the coordinates of turning points ZQ. .frW.146). (18.• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. Allowing z to approach z+ in the solution IJA(Z).^/w + ± 1 .• • " . ein+z+e in+(z 22+(<?+i) .164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.18. (18. .129) . a= ^ .125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).« + ) 2 (9+1) e (18.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e.i*l(«^ 4'"+^+ 4"'+v ( « . the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4.97b) ~ exp . (18. Here z± = ±h?/2c are the positions of the minima of the potential.127) Recalling that around arg w± ~ 0. (18. 2 An+z+e 4w+ (18. Eqs. Thus for h very large these turning points are very close to the minimum at z+. (18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.
Thus 1 E . Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. Bhattacharya [23].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. The next aspect to be considered is that of boundary conditions. Furry [102] .3. 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. even above the turning points. they assume large quantum numbers.7r and the solutions of type C around argz = ±7r/2. g . we naturally expect the wave function there to be similar to that of the harmonic oscillator. 18. and this means at both minima.414 and CHAPTER 18.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . K. K. de Deus [66] and W. basically. since. We have to impose boundary conditions at the minima and at the origin. R. The solutions in terms of parabolic cylinder functions have a wide range of validity. D. Various investigations^ therefore struggle to overcome this to a good approximation. P. Bhattacharya and A.z+)]fr+V ' (18. S.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. Since it is more probable to find a particle in the region of a minimum than elsewhere. Rau [22]. The involvement of these WKB solutions leads to problems. but it is clear that none of the above solutions can be used at the top of the central barrier.(18. S.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. H. J. The pair of solutions of type B is defined around argz = 0. 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.
(18. y{z±) £ 0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.18. as indicated in Fig. The odd wave function then exhibits a correspondingly opposite behaviour.133) y+(z±)^0. and y+(z±) = 0.136) . y'_(z±) = 0. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. ^(*±)#0 (18. 18. y(z±)=0. as indicated in Fig.5. However.132).3 The Double Well Potential 415 the most basic solution would be even with maxima at z±.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. 18. (18.134) y'+{z±) ± 0.5 Behaviour of fundamental wave functions.5. an even wave function can also pass through zero at these points. We have y±(z±) = ^VA{z)±yA{z)]z^.135) Hence the conditions (18.Zii yB{z±) a ± a yciz±) (18. and y'±(z±) V'B(Z±) ± =y'c(z±) (18. as indicated there. 18.132) Fig.
(18. (18.141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i.142) (<Z9o)J(l)^°+1)(l)1/2 . Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. ..416 CHAPTER 18. ..m Using Eq..9. Eqs. (18. (18. 1 1 .2 we can rewrite the second of (18.«*{=(.5. 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. 7 .138) Using formulae derived in Example 18.134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 .„*{!<. . .136) as  M . also (18.136) the dominant approximations we obtain (cf. +1)} = 4 m. _ 8)} s . (1.
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. y+(0)^0.3 The Double Well Potential Thus altogether we obtain 417 ^ • . y'+(0) = 0. Since the type A solutions are not valid at the minima. We emphasize: We needed the solutions of type A for the definition of even and odd solutions.6 Turning points z$.» ^ B ^ ' . (18. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. We deduce . we must also demand this behaviour here along with a nonvanishing Wronskian. 18.z\. Hence we have to impose at z — 0 the conditions j/_(0) = 0.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin. V(z) — • z E=V Fig.3.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0.18. y'_(0)^0. We do this with the help of WKB solutions.1 " <813 1 4)  In Example 18. 18.
\qh\ + \h%U{z) (18.e.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.145) \qh\ + \zW Using z+ = h2 /2c.144).^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.e.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. i. + £\h%U(z)=Q.418 CHAPTER 18. In the domain 0 < z < ZQ. (18. h8 25c2 (18. one finds that . in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. As a consequence. We also note that the height of the potential at the turning points is h8 qh V(z) 20.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18.147) 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. to the left of ZQ where V > E.148) ("f * "The superscript (I. ZQ) means "to the left of 20" . Anharmonic Oscillator Potentials from Eq. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .
148) and (18. we have to match yA(z).^ e therefore have to consider the exponential factors occurring in (18. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+. (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) .* 4+ . 20 dz \qh\ + \h%U{z) .151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing. )+^*+)2lr} rnD h\z'2)}z Q )\ = .18.149) In order to be able to extend the even and odd solutions to z = 0.^ j O e .y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . .i (2q ln (18. (18.150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .^ + ) 2 + igln2(zz+).
We see therefore. 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. 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. In a corresponding manner — i.125a) and (18.126a). This is the aspect investigated by Furry [102]. 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. l/2 ~h\U{z \(zz+?h\ (18.154) .420 CHAPTER 18. of course.125a).153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. 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. the results necessarily require adjustment or normalization there in the q—dependence. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. using Stirling's formula (and not the inversion relation) — we have —(1. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18. However. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. t h a t the Stirling approximation is amazingly good even for small values of the argument).e.ZQ) .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.
linearly matched) W K B solutions.K?+1). (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.e. 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. Comparing for z —• z+ the W K B solutions (18.157) Since the factorials [^(q — 1)]!. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.157) is used in Example 18.103) and (18. as the results also seem to support.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.n('>2o) 3/WKB(Z) =1VA{Z)I (18.155) (2z.104).18.153).154) with the typeA solutions (18. .5 for the calculation of the tunneling deviation q — go by using the usual (i. .99).156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4. this result is somewhat imprecise. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18. \hl. 27T1/2 9/4 1VA{Z).159) 7 i(«i) r(9^3) ^2~ j(9l) .e.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. we obtain in leading order (i. (18. and is shown to reproduce correctly the result (18. 7 of t h e matching relations y^NKB\z) i. Returning to the even and odd solutions (18.e.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. (18. (18.143) T^^*)) 27T (18. The relation (18.158) Applying the boundary conditions (18.
160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. D. 60 and 361.19. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18. . p.422 CHAPTER 18.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.149) near z = 0. (18.k'/2.. 3G2 12V2c ^See P. (18. formulae 220.05. 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.. n + uY 2c „ ^6 8V2c2 (18.161) we can rewrite the integral as r I2{z) = = dz^{a2 . Friedman [40].3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. Byrd and M..163b) The integral hi?) . F. G2 K [i 4.z2)(b2  z2).164) / 2 (0) z ca [(1 + k2)E(k) . K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\.iV2 ! i .163a) (18. 213. . p.148).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.
r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .3 )l ! .1 " e " O T (18 169)  Y^V / 4 tfo. [4] misses an n—dependent power of 2 in the result.e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref.. It then follows that (again in each case in leading order) '2_hl)9/4 ^ ^ * > l * = ° ^ 23g/S(_g:)t/4(!)g/4eg/4 "1^Using Stirling's formula we can write this e ( 18 . . .± ' 3G 2 ' T ( ^ (18.. (G\ . . ± . where the result is shown to be (with q ~ q0 = 2 n + l .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n . The nontrivial expansions are derived in Example 5.( ft 7. .) = 2 \.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o .18. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 ..3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. n = 0.1. ^ ^ . which implies small G2. 1. [167]. /2n + l 2 ± — .2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .165) in agreement with Ref.2. n.1)]' Gfo..
Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.3.157). (18.^ W ^ / 4 t o . Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec. (18. (i*™) We obtain the energy E(q. 18. 18. and obtain ( * . 2 . we obtain for E(q0.424 CHAPTER 18.172) into Eq.143) with qo — 2n + l. 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).143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. (18. (18. .173). Inserting the expressions for h+ and z+ in terms of h..175) ~E0(q0.1.106).3. we can therefore impose the boundary conditions at z = 0 by making in Eq.174) .n = 0. h2) the split expressions 29o+1/i2(—W2 h6 (18.159) — on using once again the duplication formula in the same form as above — 2"/ 2 (2h% 1 V/2 7 Comparing this result with that of Eq.h2) and hence the splitting of asymptotically degenerate energy levels.. .i ) ] ! e ~ s A ' . (18.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.h2) + (qq0)^=. ..6 Eigenvalues and level splitting We now insert the replacement (18. (18. (18.»W. . Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18.3 concerning the exponentials.. with the help of Eq.
h2)E+(qQ.1.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. for finite h2.106).174) and (18. (18.n = 0.fr2) is given by Eq.2.e.h2) = E_(q0. i.. (18.. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . Combining Eqs.q2 7.3 The Double Well Potential where £0(<7o.. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency. i. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .e.143) with (18.e. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. T \ *n ! 1 ' (18.e. 1 7 425 2 . the pure WKB result of Banerjee and Bhatnagar [14] (i. K.7 =( e N n+1 .^ 2^ h* .178) "In S.e.V2^\ \n+ i . 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 . the following expression for large q ~ 2n + 1.176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.18.. . i n U n / ! Thus AE(q0.157). which evidently supplies some correction terms.2 = 2 . the level splitting. The result (18. 2h4' i. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.
This. We see therefore.h2 <> — h2. K.6 \9o/2 e~h&l^\ with (m 0 = l) (18. Then the splitting A ^ W B of Eq.176) together with Eq. Furry [102]. H. (18. however. the Furry factor corrected WKB result follows automatically. 2/z4 and 2c2 respectively (see Eq. (18. we identity the parameter k there with ft.^ 2 + ^h2 If in addition the potential is written in the form A / .. Bhattacharya [23]. . these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». 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. (18.181) "W. h2)\mo=i = .2X2 V(z) = \z2tL) . (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. ** Comparing the present work with that of S.4/4 here. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.180) h8 E0(qo.6) to yield an improvement of WKB results for small values of n.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).178). that if this symmetry is taken into account from the very beginning.426 CHAPTER 18. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. h4 and c2 replaced by 2E. (66). and A there with c 2 / 2 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. as we do here. we would have obtained the result with E. as is also explained by Bhattacharya [23].h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. A>0.2 / 7. h = 1. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation." Of course. Eq.3)). Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. (18.
M. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions.^ *See E. ZinnJustin and U. M. ^To help the comparison note that in J. See R.182) implies AlE(l. Mansour and H.^ Thus for the case of the ground state Eq. (18. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. Lee [98]. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. so that by comparison with Eq.h>) * 2V2^^^y/2e^3/3A. . Banerjee and S.c2 = A/2.15).180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. The comparison with Eq. (18. H.3. (2. Liang and H. J. P.11). Gildener and A. MiillerKirsten [186]. D.—Q. K. ZinnJustin and U. Eqs.18. Equation (18. We considered above only the symmetric twominima potential. Friedberg and T.34) and (E.§ 18. these are presumably not of much interest in physics (for reasons explained in Chapter 20). D.3 The Double Well Potential 427 a form frequently used in field theoretic applications. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. and g2 is given as g2 = r]2/m3. the level splitting is (with h = 1) This result agrees with the ground state (go = 1) result of Gildener and Patrascioiu [110] using the path integral method for the evaluation of pseudoparticle (instanton) contributions. formula (4. (18. The asymmetric case can presumably also be dealt with in a similar way since various references point out that the asymmetric case can be transformed into a symmetric one. J. Bhatnagar [14]. m 2 <• p 2 / 2 .W.88) h4 = 2/x2.* A similar correspondence is also found in the case of the cosine potential. §J. J.181) therefore implies the correspondence rj1 <> A/2. however. W. Patrasciociu [110]. D. first paper. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. Jentschura [293]. Jentschura [293].
Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions.428 CHAPTER 18. Perturbation theoretical aspects are "See e. P. . Ashbaugh and Harrell [12] and Harrell [130].g. has also been the subject of numerous numerical studies. The double well potential. Very illuminating discussions of double wells and periodic potentials. Pradhan [182]. can be found in an article by Coleman [54]. and numerical studies are presented to support this claim J It will become clear in Chapter 20 — as is also demonstrated by the work of Bender and Wu [18]. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. though not exclusively numerical. 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. B. like those for anharmonic oscillators. 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]. [132] and Loeffel. as he discusses. The immense amount of literature meanwhile accumulated for instance in the case of the Mathieu equation can indicate what else can be achieved along parallel lines in the case of special types of Schrodinger equations. Santi and N. There is no reason to view the anharmonic oscillator differently.[131]. [19] — that the expansions considered above are asymptotic. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. Martin. 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. for instance. as some relevant references with their view we cite papers of Harrell and Simon [129]. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. A reasonable. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. N. As references in this direction. Mahapatra. B. we cite papers of the Uppsala group of Froman and Froman [100]. In fact. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. in both symmetric and asymmetric form. mostly in connection with instantons. Simon and Wightman [178].
^ + • (18..f + • • •) 2u\ . T h e r e f o r e it is convenient t o deal w i t h this here. (18.183) S o l u t i o n : We have k2=1^.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. Example 18.186) (18. Wave functions of symmetric and asymmetric double well potentials have been considered in the following reference in which it is demonstrated that actual physical tunneling takes place only into those states which have significant overlap with the false vacuum eigenfunction: Nieto.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. We obtain the expansions from Ref.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 . which is assumed to be small. Saxena. + (G 3G2 3^ + 2lnU 2 + i(2n + l)ln(f)+I(2n ~~ 3G 2 ' 2 l)ln(^±i q 2 1 ^ln q/ 4 2n + l (18. Bender. 2 a. X +2 + ' ^ (18.2G^2q + 4G2q . 1+ u : Gy/2q.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' ./4 E{k) = 1 + • l n In (18. Thus k' and hence k' = v ^ ( l .188) T h e following e x p r e s s i o n a p p e a r s f r e q u e n t l y in t h e e x p a n s i o n s of elliptic i n t e g r a l s .G2q + . We have 4 ~k' 2u(l . Datta.18. k! (18.\G\ + • and 1k2 =2Gv/2q4.189) =2u 2u2 (18. Gutschick. (18. Cooper and Strottman [221].192) 'F 12 16 . [122] as 13 1. Srivastava and Varma [24] and Bhatacharya and Rau [23]. fc ~ 1 . Biswas.u)1'2 = V2u{ 1 .185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.187) We now reexpress various quantities in terms of u.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. q ~ 2n + 1.
3 y .13)}.u){24 . (18. 1 3 « 2 = .2{«(1 .«) + 2} + V [ .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .1)(1 .u)} .1)(1 .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .u){4 + 3u(l .u) 2 {8 .{ M 2 ( 1 . 8 + 9u 2 .430 and CHAPTER 18.3 u .u ) (^){« + 5^ 1 )} + 5.3) .197) .« ) ^ { 2 4 . . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .195) we obtain E(k) 1 + ln uK(k) .3u(6u .193) 2u(l .18u 2 + 39M} + 12u 3 (u .u 3 ( u .3) .3).3) = = ~ .3 « .a = (l«)(jl 2u/ 2 ) .u ( l . up to and including u2): u(u .1)(1 .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .— ) .13)} + ±u3(u .u) + 2} + i u ( u .6w2 + 13u} + 12u 3 (« . (18.196) Now consider the last line here without the factor 2.u ( l .195) With (18. 8 .4.u) 2 {24 . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3u(6u .3u(6u .u){24 .1)(1 . ) J 4 (18.u) + 2}] 1 H 1 o u(u .^ L ) H[(i _ U ){4 + 3u(l .194) ln '^J + 2 In i/2u )+H l + 2u(l .u)} + — u ( u .6 in the denominator and in the last step pick out the lowest order terms in u (i. (18.13)} + 12u 3 (u .3u(6« .u 13 \ 1 ln(± k> fc'2 +  (18.e.{l u ( l u) / W 2 + 2} . 8 + 6 u 2 .u)(u .13)} + ~u3(u ..3 .194) and (18.3 u ( l .3) 1 + In ^ .u) + 2u} + i u 2 [ 2 + (1 .u){24 .« 2 + 3u] In [ .u)} u{u(l .
199) We now return t o the expression on the left of Eq.195) with (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. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .zo) HWKB normalized _ ( h* y^eM.197) we now obtain 4 1 • E(fc) .3) .. (18.4u . + iu^Zu2)^~u+ — +0(u3) l . (18.«){4 + 3u(l . Eqs.148) (for particle mass 1/2 and in dominant order) is given by H.it)} . (18.3 The Double Well Potential Now consider the bracket [.196) and (18. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.173) t o be evaluated.3).^HfHGH Example 18.e. Inserting here from the above expansions the contributions up to and including those of order u2..uK(fe) ~ 1 + In I — = Mu2(3u. From (18.e. .2{w(l u) + 2}] = u(3u2 . where w±(z) — h±(z — z±).6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. ' 8G 2 q = 4 2 q ln f 26/2 \ o . V^LJ 4 2 16 ^ . l2 = ^ ( l + uf/2[E(k)uK(k)}. we have. 2 .128) and (18.200) Remembering that u = Gy/2q. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .18.128) and (18.2 (18.198) 9 «. i. (18..] in (18.196).155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z). [(1 .4u . i.2 l_ln(1 JL\ V {2. / to + 3GH~ \ 8 . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .u H I .2 /o 2 431 (18.
Eq. W. (18. oo .158).3)]!23<«D [5(91)]' >A.e.** oo / dwD\. (18.*o)/ ^WWKBOO (18. normalized = We see that the normalization constant is independent of a quantum number.e.hA/2 = h^/4 this implies . (18. pp. 93.zo).l)]![i(0 . 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 .143). and obtained as Eq. are given by Eqs.146) and (18. by using the periodic WKB solutions below the turning points. Oberhettinger [181].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+. Magnus and F. (26. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e. i. (18. We start from Eq.204) ./ . 1 7(l. Then ((r. Example 18.Aw) = s/2i 2("!) V (41) and [(9 .g.432 CHAPTER 18.35)). i. 82.f hi\ VWKB. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. 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 [ . Solution: The turning points at zo and z\ on either side of the minimum at 24. ° \Edzy/\EV(z) V(z)V4 VWKB.147).
iK h\U{z) + / 4 (18.r + ^=co S [r.. .207) (18. of course.3.e. from z\. from ZQ.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 .2.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. i.132) and (18.e. A.( . 6.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18. i.205) We now apply the boundary conditions (18.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .g. this implies sin cos sin cos Thus e.211) 7 z l . Then at any point z € (zo.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.. We could.J^Wj +(18. Sec. \h\U{z) 1/2 ..6. (18.212) tt See e.e. 7T\ + ^ 4 7r\ dz .210) Choosing the point z to be z+.. JV = 1. Messiah [195].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.g. approach the minimum also from the right.2..209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left. 1/2 .. i.^r + + .18. z{) we expect tt (18.
In Eq. 2IO/K.214) {«+"i} 0 for q = go = 3. ._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ .215). S.3. The prefactor 2\/2h2/n is. Bhatnagar [14]. M. (18.213) The integral on the left can be approximated by 1. (18. 5.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 .143)..27 9 .exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. P. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18.2 . . Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). 9. the level splitting is given by AE(q0. D. The formula thus agrees with that in the literature.e. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients). 1 1 . (18. where u> is the oscillator frequency of the wells. i. . W "Cf. K. (18. K.5. they become c o t { ( .Zi / r.209) and (18. 7r"y in agreement with Eq.210) assume a form as in our perturbation theory.148) and (18.146)).215) we can insert for 7 / 7 the expression given by Eqs. (18.217) exp ( V [ ° dz^/E JZQ + V{z) This expression represents the WKB result for the level splitting and may therefore be called the WKB level splitting formula. in fact (with our use of qo). Lifshitz [156].434 CHAPTER 18. Banerjee and S. (18. and L.216) Inserting this into Eq.215) we can expand the left hand sides about these points and thus obtain . We now see that Eqs.149).. .205). 7 . and cot< (q + 1) '!}» for q = 50 = 1.90 — =F— for q0 = 1.215) which results from the factor of 2 in front of the sine in the WKB formula (18.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. ' /^i . Landau and E. Bhattacharya [23]. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. + l ) j } ^ 1_ 27' (18. (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.
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.* However. H. also in P. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. M. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. MiillerKirsten [9]. 'For a review from this perspective with numerous references see H. Feshbach [198]. 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. f K . Giittinger and H. Nonetheless it was thought that a certain formal analogy could be seen if the field theory is supplied with Euclidean spacetime concepts at the expense of sacrificing the interpretation of the interaction in terms of particle exchange. 435 . The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. n > 2. Aly.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Morse and H.Chapter 19 Singular Potentials 19. J. 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. are not as troublesome as one might expect. However. M. Case [45]. W. in an early investigation Case^ showed that potentials of the form r~~n. W.
In the concluding section we refer to additional related literature and applications. W.Q. Liang and Yunbo Zhang [189]. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. N. It is therefore natural that we study this case here in detail. Recently. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. J.e. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. MiillerKirsten and Jianzu Zhang [226]. K. MiillerKirsten. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. Hashimoto [124]. W. H. .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. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. § There one could visualize this scattering off the spherically symmetric potential as a spacetime curvature effect or — with black hole event horizon zero — as that of a potential barrier surrounding the horizon (shrunk to zero at the origin). Manvelyan. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. Gubser and A. Singular Potentials theories in mind. J. J. Park. that with the hyperbolic cosine replacing the trigonometric cosine). S.2 19. Maldacena [41]. R. H. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. S. Callan and J. (19.436 CHAPTER 19. 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. In fact. S. 19. Tamaryan. for instance. D.2) § C . Singular potentials arise in various other contexts.2. some also permitting further research. D referring to Dirichlet boundary conditions) in 10dimensional string theory.
( I + 0.g. 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. and as we have in mind in Sec. It may be noticed that since I is an integer.19. Eq. (19. Manvelyan. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. . W. MiillerKirsten. H. This is rarely possible and therefore this case deserves particular attention. (19.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. r = 7 e z .mo = 1/2 and h = c = 1. using the method of Dingle and Miiller of Sec.3a) In the case of the attractive potential (as in the string theory context referred to above.3b) 7 = % j. R. J. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. (19. It is clear that we draw analogies to our earlier considerations of the periodic potential. Liang and Yunbo Zhang [189]. (19. h = ei^y/k^. The following substitutions are advantageous: y = r1/2cf>. 19. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. J. In the following we study the solutions of Eq.28) below). and so =4 5 and n ez = r = ~ = e^4fir.3 below) g2 is negative and hence h? is real for E > 0.4) first for small values of h2 and then for large values of h2.2 The Potential 1/r4 — Case of Small h2 437 where E = k2.7 that we employed also in previous chapters. 8. 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. h2 = ikg.Q. ig h \j g (19. We observe that with the replacements p2 z^iz. \[\ in the correspondence is nonintegral.
Aly. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. dwA w dw { wA ) (lg.5) so that Eq.7. (19. 8.2 A / i 2 . . where these are the Bessel function of the first kind.11) We follow here largely H. Proceeding along the lines of the perturbation method of Sec. —Zu — (Z„_i + Zu+i). H. First we make the additional substitution w = 2/icosh z. (19. w 2 .2 CHAPTER 19. and H„ (w). where £>„ : = —_ + i _ + J l . (19. W.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .1. The zerothorder approximation = Zv[w) (19. Eq.. J. (19.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation.Nv(w).6) terms amounting to 1 „ 2 d2Zu A V (19.I .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.i l .e. VahediFaridi [10].6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. MullerKirsten and N.438 19. H.Hv (w). i. The solutions Zv(w) are written Jv(w). — Zv+i). (19.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2].9) leaves uncompensated on the right hand side of 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. (19.
(19.18) Now </>(°) = Zv left uncompensated Rv . z . ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.12) (19.10) in terms of functions G„ defined by Gv+a = —nZ v+a . v + 2)*Z„ +2 (19. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). = §.u + 4fZu+A}. v . (i/.2)G„_ 2 + (y. . wz The expression (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except. The terms (19. v + 2)Gv+2].. where the starred coefficients are defined by (19.19.2 ) * i ^ 2 + (!/. where We observe that DUZV = Q.6). of course.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19. I / ± 2 ) = 1. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19. However. v ._2 + {v.13) therefore lead to the first order contribution jfU = h2[{v.16) ( I / . and this means in (19.13) (19. it is more convenient to use these relations in order to rewrite Eq. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ .10) as a linear combination of various Zv.2 ) * ( z / .u + 2)*Z„ +2 ]. (19.10) is now particularly simple: i?i°) = h2[(v. v + 2)*(v + 2.17) Thus a term fiGu+a on the right hand side of Eq.15) (19.13).2.i/) = 2A.14) (19. when a or 2v + a = 0.i/4)*Z 1 / _ 4 +(u. (19.u (19. v2)*{y2. .21) +(i/.v + 2)*{u + 2. v)Gv + (y.!/+ 2)*4°J 2 ].2)*Zi. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.20) 2)*ZV_2 + {y.
v2)*{v2. Adding these terms and setting the coefficient of Gv equal t o zero. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.1169)/i 12 . (I/. i/ + 2j).23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . . v ( 4 5 a .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. . p 0 (0) = 1.26) iy — 2.v + 2)*(i/ + 2.i/±2)*. in Eq. P2*o(0) = 0. subject to the boundary conditions P2i(2j) = 0 for  j  > i.2 + 2 9 ) ^ 64(^ 2 .j^0 (19.4)(»/2 .24) P2i2(2j + 2){v + 2j + 2.v) + hA[{v. (19.v) + {v.v)] (19. j=i.l)(z/ 2 . . we obtain 0 = h2{v. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. . we obtain t h e expansion . (19. Evaluating t h e first few terms.v + 2y{v + 2. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + . p4(±2) = (i/. v + 2)*{v + 2. i/)] + • • • . ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .455a + 1291a . i/ ± 2)*. .440 CHAPTER 19.25)/i 8 32(al)3(a4) . v — 2)*(v — 2.. we obtain 9 h4 = a 2(al) 3 2 (13a .2)(v + 2j . . 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . v + 2j) (19. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. and p 0 ( 2 j ^ 0) = 0.g.2. (19. . v) +(i/.22) where p2(±2) p4(±4) = = (i/. Ri.17)).4 + 58. i/ ± 2)* (i/ ± 2.I/±4)*.25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv .I/±2)*(I/±2. etc.
cosh 2z)4>v = = 2h2A(/>„ .2.19. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1.32) (19. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . . we can rewrite the latter as j ~ . The solutions 4> of the modified Mathieu equation are now completely determined.i . 19.31) It follows that Dv+2n4>v+2n = 0. (19.4). we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . ±2. so that J2 (19.^ + e±(v~2)z. (19. £>„ = — .2)<t>„2 + (v.33) = = cosh(^ + 2)z + cosh(^ — 2)z. (19. (19. v .v + 2)<t>v+2].h2[(j)u+2 + 4>v_2] h2 [{u. 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 . Thus to O(0) in h we have (jr ' = 4>v — cosh vz.30) £>„&. Substituting (19.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.34) . (19. and is therefore real for both cases g2 > 0 and g2 < 0.v2(j> = 2/i 2 (A . 2 . = 0. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z.7) into Eq.cosh 2z)4>. Thus v is real for small values of  h21.29) sinh vz or e±uz. v)(f>v + {v.3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17.An(u + n).
h). Schafke [193].I/±2)*. > (f> with </v = sinh vz — Seu(z. W. we prefer to adhere to one equation with different solutions.4 Notation and properties of solutions We now introduce standard notation as in established literature. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. . M.I/) = 2 A . Meixner and F.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter. 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.38) (f> with (\>v = exp(vz) — Meu(z. The solutions in terms of cylindrical functions are written *J. Arscott [11]. In order to avoid confusion arising from the use of different equations. (19. > These solutions correspond in the periodic case respectively to the solutions and me^.J7t0 p2i(2j)<f>u+2j. (19. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. CHAPTER 19.37) where p M (±2) = (I/.7 — we refer to Meixner and Schafke [193].442 where (I/.2.J/±2) = .h).h).35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. For rigorous convergence and validity discussions of any of these solutions — our's here differ only in the method of derivation with the perturbation method of Sec. v ± 2) etc. Singular Potentials (I/. The use of the symbols (v. F. • (19.1 .h) = <(>„ +Y. 19. 8. etc.h2i *=1 J2 j=i. in the present context should not be confused with the same symbols having a different meaning in the case of solutions in terms of cylindrical functions since it is generally clear which type of solutions and hence coefficients is being considered.
Me„(z.42). (19.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant.40) Cev{z. as Mev{z. Schafke [193]. Also from the expansion of Jv{z) in rising powers of z.h). h).h) = exp(vz) + Y. 443 (19. For the relation of Hankel functions used below.43) and hence similarly M^\z + inir. Oberhettinger [181]. see this reference p.42) The solutions Meu(z.g.45) T Below we frequently write the normalized solution as in J. 16. ' (z. h) = exp{vniri)Me„(z.h). h) = exp(inwn)MP(z. whereas the — as yet — unnormalized solution (19. As emphasized in Chapter 8. {2h cosh z) • MJ?\z. h) are therefore proportional to each other as a comparison of Eqs. h). 17. which. as we remarked earlier. (19. h).MP(z. (19.e.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) . .h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. (19. p. 'See e. h) = av{h)M^ (z.h) and oo i M^\z. h).44) implies. 4>v = Hi. <f)v = N„(2hcosh z) MJ?\z. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. 4>v = Hv (2h cosh 2) * M^\z. W. Meixner and F.h2) oo = Y. (19. i. Mi. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv.19.h).h)±Sey{z.h). — oo cv2r{h2)e^+2r>. this normalized solution possesses contributions exp(i/z) in higher order terms.(2hcosh z).e. Magnus and F.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z).44) (19.41) i=l We see immediately that Meu(z + niri. i. (19.40) does not. W.
g.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4.h) Using further properties like (19.h2) = ] T cv2r(h2)e{y+2r)z. the first relation is obtained as follows: 2M<£1 = M^l+M^l.49) Inserting this expansion into the modified Mathieu equation (19. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19.h2): oo Mev(z. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. r=—oo (19.exp(TM"r)MW(. .e. oiv{h) = (19. Meixner and Schafke [193]. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). (19.46) H^z) = e^H^(z).43). h). For later essential requirements (i.50) E. H™(z) = e^HfXz).\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. p. i.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. Singular Potentials where clearly MeJQ. i. 130) to converge uniformly for all finite complex values of z.h) on the other hand can be shown (cf. 2M™ = eiuvMi3) + e~ivlrM^\ . h) .e. These points are important in our derivation of the S'matrix below. MJ>>(0. Mi 3 .h) .48) The series expansions of the associated Mathieu functions M„ (z. we have the following for a change from v to — v. s (19. Meixner and Schafke [193].8. p.h) can be shown (cf. ^ . h) = M™(z.e. As mentioned earlier. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions.444 CHAPTER 19. The functions Me±v{z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z.
+ 2 r ) 2 ] .* For 'Actually.54) 2 C2r 4 ^ h. h~2[\ . (19. . Eq.^ C 2r+2 (19.2 [ A . W.[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.52) in agreement with Meixner and Schafke [193].^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 .53) /i2[A(^ + 2 r ) ] . + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. Thus now c 2r . 2r (19>51) 2r For ease of reading we give the steps in rewriting this./ i 2 C ^ .19.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . Alternatively taking the inverse of Eq.^ + 2r + 2 ) 2 ] .^ p 2 2 ^ or c 2r+2 2r2 1 (19. .(/^.(v + 2r)2} / z . Schafke [193].51) we have 2r (19. see J. eg = c~v = 1. \) /. p. Meixner and F.T22r[\ A . 122. \ . p. (39). 117.
i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . 128(i/+ l) 2 (i/ + 2 ) ( ! / . Aly.Q .5 D e r i v a t i o n of t h e S . H. . J. (19. J. + 0(/. W. J . Liang and Yunbo Zhang [189].C + l) 3 (i/ . but follow our earlier reference H. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19.^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). Here we insert for A the expansion (19.[v^. MiillerKirsten. and also R.+4)!ll_.+0^"> < 19 .2. Spector [257]. Solution: We put r = 1 in Eq. H.( " + 2)2] ~ h2 h. W. H.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy.55).. M.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..27) and truncate the continued fraction after the second step. (19.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. Then di 2 1 t" + 2(^1) + • • • .446 CHAPTER 19. + i).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.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. MiillerKirsten and N. . VahediFaridi [10] (this paper contains several misprints which we correct here).l ) 19. Manvelyan.1 for the evaluation of a typical case): hb + 0(hw). 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ ._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(.55 ' Example 19. t For this purpose *We do this in the manner of the original derivation of R.
(19. we have W r = ip + ut — 7r/4' .h) 1/2 w) (19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. In the case of the repulsive singular potential here. 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).19.59) which tends to zero with r. I / + 2 2 l\TT +E ^ i\ E j=i. W . Oberhettinger [181]. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19. and then the continuation of this to infinity. Then y reg = = r rV2M®(z.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. p.3a) and (19. (19. and considering the propagation of this wavefront. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. Magnus and F. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const.58) The behaviour of y reg near r sa 0 is therefore given by 2 1/2 2/reg "\ 5 exp 1\TT r\ —r exp i .56) where by Eqs.. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19.j7L0 P2i(2j)exp i[v + 2j + 2 2 .57) ^ 1 . 22.
h) through the entire range of !&{z).A1"—2 z '+"2 A~~ n or rA . (19. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z. Singular Potentials so that when t — oo : r — 0. 2h and the square of this expression implies 2 2 ~ ' A. (19. p. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo.448 CHAPTER 19./i). The series M^ (z. 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.j¥=o Ki(2j)exp(Tmj) (19.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. p. '2"4 . We require therefore > > the continuation of Mi.V + § 9 2 > 4hh* ./l ~ 2 ~ 2z + " 4 > 0. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.h) can be shown (cf. Magnus and F. Oberhettinger [181]. '(z. 178) to be convergent for  cosh z\ > 1.3a) and (19. 17.62) Also with Eq. Meixner and Schafke [193].h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.A14A.3b) ln • 7T Q"*4' r fg ° Vk (19.h) = exp(Mr^)il4 4 ) (z.60) Using the above asymptotic expressions for the Hankel functions.j^Q p2. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. Now from Eqs.#. . 4%r g >0 4r^r ' rg See W. AV . This means that the origin of coordinates acts • > as a sink. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).61) In fact.(2j)<gH • (19.
449 This condition. Meixner and Schafke [193]. Originally we chose the sign of m / 4 as in Eqs.38) or (19. Thus we choose .e. 130). Since the real part of z changes sign at r = ro. is satisfied only for r < r_ < r__ or r_ < r + < r.62).ijt/4 Rez . z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo.rl)(r2 . i. we have to assign different signs to the imaginary part of z in the two distant regions.r2_) > 0 with r£ = rg(2 ± >/3). which has to be bridged by using another set of solutions. 19. 19.49). The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. p. This interchange in the > solution My interchanges y reg with y+.3a).3b) and (19. (19. (19. and in order to maintain this symmetry. These solutions converge uniformly for all finite complex values of z (cf. (r — r+)(r + r+)(r — r_)(r + r_) > 0.63) Thus there is a gap between the two regions of validity — i.e. the domain r_ < r < r+ — as illustrated in Fig.19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 .1 The domains of solutions.1. A suitable set is the pair of fundamental solutions Me±v defined by (19. i Im z z = In r/r0+i7t/4 . r 7r for 0 < r < ro. . (19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig.
Singular Potentials Then starting from the region r ~ 0. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3. as just explained.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_. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.z=—in/4. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}.450 CHAPTER 19.66) a = (3 .[ r l/2 M.e. . (19. At r = ro the real part of z vanishes and then switches from negative to positive. Hence we match the right hand sides of Eqs. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. to r^2[AM^ + BM^\ A. p.^)] dr z=in/4 dr (19. (19.v(z)\ —m/4) + ^[r^Me. i. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. i.(2)]z=w/4 + 0Me.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.B^0. Thus.l/2Me_i/(T. h) (as one can check or look up in Meixner and Schafke [193]. 131).v{z)\ .)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. 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. at this junction we require also the imaginary part of z to change sign. (3 = a . h) = Me^u(—z.
(19. Then.g]:=f(z)*&g(z)WZ) dz dz Thus. (19. ^ ] = ^ + ^ ^ . the nonderivative parts (from the first term on the right of Eq.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri. multiplying the first of Eqs. (19. Thus we obtain from Eqs. (19. A and B.68) From Eqs. (19.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. (19.71) We now determine the coefficients a.64) the relations M&Hz) jz[M^\z)] and from Eqs.70) by Me'_u (the prime meaning derivative).70) by MeU and subtracting . with the replacements of Eq. = aMev(z) + 0Mev(z). we use the notation W\f.19.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . and we are left with relations expressible only in terms of z. (3.69)) cancel out in view of the nonderivative relations. (19. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.3a) and (19. (19. (19.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.3b) we infer that (z = l n r + const.65). for instance.68) the derivatives with respect to r by this relation. dr BM^(r)].70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}.2 The Potential 1/r4 — Case of Small h2 451 This solution can be continued into the domain below r = r+ by matching to the right hand side of Eq. (19. and the second of Eqs.65) and (19. (19. (19.69) Thus if we replace in the derivative relations of Eqs. (19.
M^} W[Me„. (19..48) (or cf. . h)/M^\o. M^4)] + W[MJ>3).Mev] = = = av.47) we obtain (av defined by Eq. = IT <*„.(0. (19. TV W[Mev.Me^} (19. 7T [1.4] = . B = J 2 L _ e . (19.e.\ \ ( av 2iwKav ^ v . 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.* « * r r z H .j] given in Meixner and Schafke [193] (p. .46). and the second is obtained similarly: W[M?\Mev\ W[Mev. or obtainable by substituting e.. we obtain immediately W[Z. i. p.^ ^ a „ Q _ . 7T (19.M^] J™aV.3] = [1. (19. the leading terms of the respective cylindrical functions. 7T W[Mev.73) We now use Eq. i.M^]= IT e~iv*av.76) .4] = .M^] W[Mev. using Eqs.57) and (19. (19. we use the following circuit relation which can be derived like Eq. Me.58) and considering large values of \z\. W[M^\Mlf)]W[Meu.Mev]' P l ' From Eqs. h). 170/171).u}W[Me. W[MJ?\ M^] W[Mev. (19.Mevy W[M?\Mev] W[Mev. we obtain the first of the following two equations. (19. (19. Singular Potentials the second resulting equation from the first resulting equation.e. av{h) = Me.77) av I 2ism VK \ ctv av ^Thus.74) Moreover.71) we obtain in a similar way A B = = W\MP\ Me.452 CHAPTER 19.46)) W[Mev. Meixner and Schafke [193]..MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. and Wronskians W[M^\MIJ)] = [i.75) With these relations and Eq.g.] W[Mev.^ 47 9? MJ>4)] MeV] [3.u. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.A] = W[Mi3\z).
(19. B=tfD'"".78) we obtain Ay reg « _ ./ ( .19.h).79) where the second expression follows from Eq. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq.46) together with the relation Mev(z. Z ) e ^ ] . (19.85a) . (19. Then f' . 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 . in Chapter 16)._ [ / ( f c .h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM).l) ' We note already here that with the substitution R = ei7r7. f c r . Z ) e .2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z.(ite™)" 1 { 8 J ' f(k.f c . (19.80) 2^fc (1981) where \7r/ i t sin VK /(*.h) = Me_u(z. (19.(l)le^}Pi(cos 9). as we recall from earlier considerations (e./) = (l)1/2R2^2i^)ei^)U^.h) + BM^\z.83) The Smatrix element is therefore given by ifc*"* . eii/7r/2 (19. (19.g.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).
We can rewrite Eq. Singular Potentials r(i+k) sin 7r(7 + u) (19.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). R 2i sin TXV. and with the left hand side following from Eqs. (19. Thus in terms of the (3) variable z.Smatrix in the form s i n 7T7 Si = CHAPTER 19. and recalling that yreg is proportional to a function Mc> {z. (19.56).58)) 1/2 2r e~ 2hir cosh z. (19.2. Fig.86) . V(r)=g2/r Fig.56) to (19.58)) in the limit r — 0. h) (see Eqs. cf. R — — = 2i sin ivy.454 we can rewrite the . we can define respectively as amplitudes of the incident wave. 19. we can write this (multiplied > by 2ismvTT. 9?. (19.2 The repulsive potential and the various waves.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.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. (19. 19.
( i J e ^ ) " 1 ! 2 = R2 + ^ .h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz). " S e e S. 178). (p. 180).m a t r i x Our next task is to evaluate the expression (19. H. i. Jv+2r. i. this expansion is given as 1 oo MJ> Hz. J. Eq. (19.In Meixner and Schafke [193] (p. in which v is complex. Manvelyan. l=—oo 1 1 lite*"* .. Gubser and A. S.e. unity minus reflection probability = transmission probability. and then exploit it for the evaluation of the quantity R. MullerKirsten." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* .e™*/R\2 ' ' . y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. W.Q. (19.** 19. (19. This implies basically the evaluation of the expression R of Eq. J. / i ) = Y. The latter is what happens in the S'wave case of the attractive potential. (105) to (108).h) The function Mo (0. (19.28)) and R = ey real. therefore we cite it from Meixner and Schafke [193]. Liang and Yunbo Zhang [189]. (19.19. i. Eqs.e.2.79). h) serves the use of the same coefficients as in the other expansions. 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. (19. but in the 10dimensional string theory context. Hashimoto [124] and R.2COS2VTT = (R.88) where the factor Me„(0.h) is (cf.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. i.84) of the ^matrix.±)2 + (2sinj/7r) 2 . The expansion is: 4 r ( / i 2 ) M S ( z . however not here. (^)l4lu{h2)Jir(hez)J±u+i+r(hez).6 E v a l u a t i o n of t h e S . By inserting in Eq.e. A derivation is beyond the scope of our objectives here.e. if R becomes a pure phase factor and eil/7r a real exponential. unitarity is preserved.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.89) . M^l)(0.
Next we evaluate M^1}(0.. (19.3 9 ^ .90). by allocating different values to r. setting r = 0 in Eq.456 so that in particular for z = 0 oo CHAPTER 19./i) with the help of Eq. r = 2 as a check).e. e. /n ^ .90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. Eq. (19. Singular Potentials 4r (h )M$(0. The formula (19. ^ (19./>) = Co> 2 ) £ 7 " = — OO j ± > .49). + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ .28)).55).h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. /=—oo (19. u»^. we obtain M&(0. (19. h) in many different ways.g.91) Inserting here the coefficients given in (19.2_4)(l/2_9)iv2y) + <AA. (19. i.49) from which we obtain for z = 0: Me„(0.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. 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. h2 is complex but v2 is real (cf. Here in the case of the repulsive potential. r = 0 and 2.6 2 ) m 6 (^_i)3(z. (19. one obtains *. As a matter of introduction we recall the definition of the coefficients with Eq. (19.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).90) and choose r = 0 (one can choose e.94) .g.
] 2 \v i + {v2 . _2_(h\2 .1 ) 2 ( I / + 1 ) ( V 2 ..l) 5 (19.2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19. one obtains M£!(O.96) With this explicit expansion we can evaluate the S'matrix.7) F " (i/±l)2(^^l)(i/24)V2.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .. 4(z/ T H ^ 3 .32i/ .IV 2 2 2(v =F 3i/ . .2 _ ^ 4 ( ^ 2 _ 4)2 (19.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu . + # "+" ( I / .111) / / i 5 4 3 1+ 4i/ 2 + +' (z.. From expansion (19.9) 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.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .1) 2u(4u + 15i/ .4)(i/ .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy./I) h u .2^ 2 ± 59^ . We observe that this approximation involves v and h4 and hence is real.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).93) and the coefficient expansions (19.12i/2 + 64u .95) These expansions imply for the quantity R: I/! 2v ri1 i .19.
28)). 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. and sin 2TT/3 = ft/. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2). This is the case of real . (19.103) Then SiSt (1+5) i . (19. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. (19. + ( i + s w (E 4 .101) is > zero.9 / . Thus we set v = n + i(ot + i(3) = (n — 0) + ia.eiun/R2)(eiu*n ~ (19. In this case R — R* and so 1/i?2 ~ Ofji4).99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19. (e iun e~iv*^/R*2)' (19.^ i 4 2+^ # 1 / •. 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.98) (cf.28) that v is real.2.104) We now consider the case of a — 0 implying that g of Eq.458 19.102) (19.84): Si ST 1 (11/R2)(11/R*2) . (19. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3.7 CHAPTER 19.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. Then cos 2TT/? = 1 .R*2 ^ * #2 \ 11 tf2 R* (19. (19. Eq.101) where / is complex and g is real.
102) we obtain Sis. 1 29/ ^ (1 . With t h e help of Eq.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. H.107b) = Ah2[l + 0(h4)}. i. Manvelyan.^ ) ~ 2 2 459 1+ 9 / • & ( ! . J.28). At = 2i[l + 0(h2)}. "This is the result effectively contained in R.19.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 . W.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)]. S. = 1 .106) and evidently violates the unitarity of S.3.e. We can also evaluate now the amplitudes (19. .61n/i} + ~9~ 0(h8) (19.1)\} 2 h \2 iv 1 + Au {y . Liang and Yunbo Zhang [189] (there Eq.3b)) h2 is real for E = k2 > 0. (19. Hence with Eq. MullerKirsten. Here in t h e case of the attractive potential (as remarked after Eq. Hashimoto [124].SiSf » 4 sin7rz/ R (19. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .l ) 2 2 \) +0{h«) (19. ( I l l ) ) in agreement with a result in S.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 .107c) Here h2 is actually Vh4.97) this can be written A 4TT2 [v\{v . (19.105) sin7r/3 (19.Q. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the threedimensional case we are considering here. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z).6C . Magnus and F. f W .1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. Oberhettinger [181]. (19. Gubser and A.61n2 . J. (19. p.577.
(19. h) for large h and upwards I = 0.109) ^We follow here mainly D.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. . S. MiillerKirsten and Jianzu Zhang [226]. We are again concerned with the associated Mathieu equation given by Eq. 2 2 a = [l + 2 (19. K. H. VahediFaridi [8] and G. Park. This is therefore a highly interesting case which found its complete solution only recently.3 19.3.460 CHAPTER 19. 2. where the part of interest here is based on H. Wannier [278].e. H.108) For the large values of h2 that we wish to consider here. Aly. N.H.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . i. completely different solutions. W.1.26) and hence set a1 = 2hl + 2hq + A{q. 19.4 Fig. 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. Tamaryan.3. we again make use of the replacement (17.3 The function q(l. J.$ q(l. J. Singular Potentials 19. H. Miiller and N. W.h) 10.h) (19.
5) that this is given by the relation cos iris = ^ M .3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively. however. i.§ a W[y+. with an essential mathematical step.113) 'We are considering nonintegral values of v. i.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. to which we have to return in this subsection.e.3. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.yJ\ — 1 and cos TTV .y} where VF[y+.h*y (19. we defined the Floquet exponent v and observed with Eq.63).110) a where y+(7r) is the solution evaluated at z = TV.19. (17.111) we have W[y+. J4(0)=0. 19.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. We begin.e. and a is its normalization constant. (19. We also found the boundary condition (17.h*y_(l. y± oc A{z)e2h sm * ± A(z)e2n sm z . See also Eq. (17.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively.3. y'_(0) = 1. Therefore the second term on the right is nonzero. which is even about z = 0. 19. y(0) = 0. The replacement (19. (19. With the boundary conditions y+(0) = l. The behaviour of q as a function of h for some values of / is shown in Fig.21b). which therefore cancels out or can be taken to be 1.21b).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.l + 2y+(l. (17. (19. .
Hence we require (obtained from Eq.43) we see that z = 7r/2 implies W(TT/2) = 0.117) a a Referring back to Eq. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. 1 1 Or see R. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . K [ (z)}2hsinz (19.32). (19.A(z)e2h sin 2 sin z sin z (19.118a) "We emphasize again: For integral values of v the normalization constants are different. W. B.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. This paper contains the normalization constants of large/i 2 periodic Mathieu functions.e. Dingle and H. We now obtain the expressions needed in Eq.112) from evaluation of Eqs. (19. W. J. Dingle and H.114) at z = 7r/2.114) a a Our first step is therefore the determination of the normalization constants No. which is \/2 at z = 0. 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 . (19. in the second expression 2h is obtained in addition from differentiation of the exponential factor). See R.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq.39 2 14 /i 2 + (19. Miiller [73]. .115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).46) together with the expansion of the Hermite function from Tables of Special Functions" ) B[w = 0] 7T 4 [lfol)]![l(g+l)]! 2 /i 5 82q2 . J. i. (17. (17. (19. Miiller [73].462 CHAPTER 19. (17. and hence with Eqs. B.
MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. (19. p.! ) ] ! [ .117) are known from the matching of solutions in Eqs.87 2 14 /i 2 [ i ( g . Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .119a) This formula — derived and rediscovered in Park. (19. a in Eq.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .1Q11QM cos irv + 1 ~ •==— e .118b) The factors a.2g + 3) 27/i ( 3 g 2 . 210.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v. (19. .62b) (see also the explicit calculation in Example 17. (17.19.3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .i ( g +3)]! 1 24/i {2hl>2).3). Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 .! ) ] ! [ . (19. Tamaryan.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19.62a) and (17.h .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].
? s m h z) Vcosh z (19.h)iP(q.z) i/>(q. . z) are TP(q.h.120) The resulting equation for the function A(q. With the solutions as they stand. z) = ip(q.z).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/\ (19.464 CHAPTER 19.3. In the following we require solutions H e ^ ( z ) .3. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . 17. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.3.123) We again make the important observation by looking at Eqs.h). z) can then be written as .z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. .109) into the equation (19.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i.122) Correspondingly the various solutions ip(q. —z). tp(q. z) = ip(—q. for h — oo). h(q. dA 1. h.h. Vcosh z \ 1 . h.h. the expression (19.h. h. of course. —z). z). h. V— 2in (19. We define these solutions in terms of the function Ke(q. it is still a solution but not with the symmetry property ip(q.122) that given one solution ip(q. we can obtain the linearly independent one either as ip(—q. h.120) to (19.z) = A(q.h.h. —h. —h. h.2 we insert the expression (19. i = 1.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. .z)exp[±2hismh z\. Then straightforward integration yields > inq 4 „.2. (19.124) Since this function differs from a solution ip by a factor k(q. each with a specific asymptotic behaviour.109) remaining unchanged.108) and set ip{q.h) =?==.z). (19.4.3 Following the procedure of Sec. h.z) or as tp(q.. „ 1 A „ —A cosh z——H .e.
.e. in performing this cycle of replacements the function picks up a factor. i. we h a v e 77 = 0—> 6 = 0. i. [ \ . (0.<5±1) (19 125)  (the expression on the far right in leading order for h2 large). : 7r/2.3 The Potential 1/r4 — Case of Large h2 465 Instead. Me„(z. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr.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 ./h). / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . achieved with the help of the Floquet solutions Me±v(z. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. (44) of W a n n i e r . (19.fi2) can be reexpressed in terms of Hankel functions. IT a n d 7 = 00 — * 7 > : 2fcsinh y + i<j7T h Aw 26/? T h e factor e " r ( ? + 1 ) / 2 o n t h e r i g h t of E q . This is. In order to be able to obtain the Smatrix. . av{h2) = ^(0J ^ Mi i . 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].127) sin z T /'1 M TT\ — <>{ 4 .h2) = OvMPfrh2).^ s i n h 7?)]g.e./i 2 ) M (19. the Bessel functions contained in the expansion of the associated Mathieu function M{. as we saw earlier. It follows t h a t S e t t i n g s i n h 77 = t a n 0 . or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . (z. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO.h2). (19. >• * F $ 0 * e".45).e. i.126) For large values of the argument 2/icosh 2.19. w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a. We observed that these satisfy the relation (19.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]. 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 . 2j<. * ** = > rojKe( ".
/i) Be^(z.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H.q.h) He (z.h) (3) = = = = Ke(q.g.h.eW(z. 3te»0. Singular Potentials Note that the sine part is nonleading! Retaining only the dominant term of this expansion we have (with z — 2h cosh z) > Me±l/(z.h) = e(z) exp ihez . KeW(z. He (z. z): Re(2\z. (19.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T .q.466 CHAPTER 19.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.q.g.h) Be^(z.q. HeW(z.h).130a) Re^(z.q. In this way we obtain in the domain . 3te»0.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.3.q. (19.130a). (19.q.130b) r_0 rll2exp\g/r (ig) ' 19.h).130b). exp[—ikr — in/4\ kr Re{2)(z.q. (19. (2) (19.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.4 T h e c o n n e c t i o n formulas We now return to Eq.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.vr .z). (19. h./i).
q. Changing the sign of z. (19. (19.134) in the region where the function He^(z. q.q. h. / i ) + sin7r(7^)He(4)(2. (19./i). (19. q. h) = Ke(q.li 2 ) Oil. (19. q.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e .h) is asymptotically small (cf.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. z) and He (3) (z. h) — sin 7T7 He^(z.132) into Eqs. g .131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.g. — sin 777 H e ( 3 ) ( 2 .130b)). h.h2) Me„(z. (19. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .131) where the second relation was obtained by changing the sign of v in the first./i) — sin 7T7 Re^(z. and thus can be matched in this region. z) are proportional there. and He(*> exponentially decreasing.h2) a. h) = sin 7r(7 + v) He (1) (z.q.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19.q.h) sin irv = sin7r(7 + ^)He (3) (2. t With Eq.134) Considering now the first of relations (19.(z. these are exponentially increasing there. Eqs.132) These relations are now valid over the entire range of z. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e . we obtain in the domain tfcz <C 0 the relations: Me.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z.i ^ / 2 H e ( l ) (z> ?) (19.h).130a). (19./i).135) sin irv 'This means.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. h) = Ke(q. .129)) He (2) (z.19./i 2 ) = Me_j.g. (19.g. Substituting the Eqs. we see that the solutions (compare with Eqs.(z.133) In a similar way one obtains the connection formulas — sin nv HeW(z.
119a) or (19. (19. which in our case is the solution He^4^. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . i.119a) and (19. and therefore remarks after his Eq. (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".3.Sin TXV. Wannier did not have the relation (19.135).119b) as „4/i 1 + COS TXV ~ rrl ( g .85b) obtained earlier. Squaring Eq. H.(19. cos TXV + sin TXV.e.l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. sin TXJ. . of course. (19. The quantity 7 is now to be determined from Eq.133) we can now deduce the 5matrix Si — e2t51.137) We see that this expression agrees with that of Eq. The latter is defined by the following larger behaviour of the solution chosen at r = 0.e'i7ri 1.139) ''Equation (19.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . where 8i is the phase shift. (19.1 .135). h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . (60) in the work of G.133)) — sin TXV He^4' (z.l ) (l„—ikr e From this we can deduce that Si (19.119b). Wannier [278].138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. (19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. q.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.( .135) is the equation corresponding to Eq. Thus here the S'matrix is defined by (using Eq.m a t r i x From Eq. (19.5 D e r i v a t i o n of t h e ^ . Singular Potentials The factor on the left. (19. is only the dominant term for large h2} 19.468 CHAPTER 19.
e. since cos TVU is large. h) of the Zth partial wave.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19.4 The absorptivity of the S wave (attractive case). (19.28) shows. imaginary part uj (so that 1 on the left hand side of Eq. as a comparison with Eq. Si xe ~ te ilir COS KV V cos2 irv — 1 — e (COS^Q COS ITU ig7r 1 + e'iqn ielln cos TXV 2 2 C O S TTU (lp/2)n (19.8h COS From Eq.19.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. .109). for the attractive potential.140) From this we obtain the absorptivity A(l. 9/2 (91) with the approximation q ~ h obtainable from Eq. (19. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. (19.§ Since the right hand side grows exponentially with increasing h the Floquet exponent must have a large 1=0 1 large h approximation small h approximation Fig.141) COS TTU irvj — i s i n WR s i n h TTUJ.e. T h i s is r e a l for UR a n integer. i.137) together with (19.e. the real part of v must be an integer. (19.139) could actually be neglected) implying  cos m>\ = cosh nvi. i. This is different from the behaviour for small values of \h2\. 19. we can approximate the equation for q ~ h (i.138) we obtain. A{l.
Eden [46]. Y. Cvetic.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. Apart from the papers already referred to.107b) and (19.4 (there are tiny fluctuations in the rapid approach to 1). L. F. one may expect corresponding results for other singular power potentials. 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. The diminishing fluctuations of A(l. N. E. small h2) case have been considered by some other authors. Shabad [248]. Bertocchi. 11 J. Jones [78].139) 2Tr(16h)q /32\ f e Thus with Eqs. H. S. Challifour and R. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . such as phase shifts. Partial or other aspects of the weak coupling (i. **A. J. in further contexts beyond those already mentioned.4 . 19. D. Treiman [271].e.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r .** 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. Lii. and G. C. A. in fact.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. Paliov and S. Lew [128]. N. A. H. H.P. YuanBen [285]. Cvetic. Limic [177]. Furlan [20]. 19. With the exploitation of perturbation solutions of both the periodic Mathieu equation and its associated hyperbolic form for both weak and strong coupling (more precisely h2) together with corresponding expansions of the Floquet exponent it was possible to go as far as the explicit calculation of the S'matrix and the absorptivity. N. (19. W Some of these sources can be helpful in further investigations. B. Lii and J. VazquezPoritz [61]. Tiktopoulos and S.k) in the approach to unity are too small to become evident here. D. Apparently this is one of the very rare cases which permits such complete treatment. **R.470 CHAPTER 19. Esposito [88]. "G. Fubini and G. Dombey and R. A. see also M. Pope and T. This is sketched schematically in Fig. In essence." as well as other aspects. S . Handelsman. I r a n [61] and M.142) we can see the behaviour of the S'wave absorptivity as a function of h. and. Rosendorf [225]. (19. Masson [191]. Pao and J. .
T.g. W. Wu [18]. 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.* In fact. *R. This is indeed a remarkable connection. J. 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.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. B. Dingle [71]. B. We concentrate in this chapter on the large order behaviour of asymptotic *C.* However. M. also convergent expansions are known and a lot of literature exists on the equation. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. We shall also see explicitly. e. the asymptotic solutions in different domains all have the same coefficients. Dingle and H. f R . Bender and T. 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. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. Muller [74]. 471 .1 Introductory Remarks The subject of the large order behaviour of perturbation expansions — meaning the study of the late terms of the asymptotic expansion of some function with a view to extracting information about the exact properties of the function — received wide publicity with the publication of the anharmonic oscillator studies of Bender and Wu. [19].
2 and below. as evident from Example 17. 20.2.7. 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.472 CHAPTER 20.1 and cases in Sec. Presumably this has never been done so far. Fig. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.2. r for p = 4. 10 12 14 16 18 20 Fig.6. 19.6. 8. Since the method is also applicable to convergent expansions. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.8. r for p = 4. This approach leads .1 The function ur = pr jr\ vs. 8. 5. 20.2 The function ur = r\/pr vs.
whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. in the power expansion of the exponential function). The formal perturbation expansion of a quantity E(g ) . For an authoritative discussion of these aspects we refer to Dingle [70].2. . 20.3) n=0 Thus for any given value of g . The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. (20. and the integral E(g2) is called the Borel sum. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term. i.1 and 20.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20.2) or. and the expansion is said to be Borel summable.e. It can be seen from these.J2 2n h E (20. for instance. A considerable amount of work has recently been devoted to the question whether it is possible to reconstruct a function exactly from its asymptotic expansion. The behaviour of such terms is illustrated in Figs. 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. this term represents the size of the error. equivalently for g = l/h. In some cases the answer is affirmative.e. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. as we discussed earlier in Chapter 8. The traditional method of using an asymptotic expansion is to truncate the series at the least term.1 Introductory Remarks 473 automatically to asymptotic series. 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. even if each En is known. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). T V lim h2N 2 n EQ?) . i.20.
In fact.474 CHAPTER 20. (20. . The function F = 1/(1 + z) is an analytic function which can be continued beyond this circle to the entire positive zaxis as required by Eq.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V .5) then with the integral representation of the gamma function F(z) = (z — 1)!. Thus if CO F(z) = aiS(z) + J2 anzn. .4). 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. n=0 (20. (and E0 = 0). that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). . . (20. Large Order Behaviour of Perturbation Expansions Eq. . In general this is not so easy. (20.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. (20. . We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: .6)).1)? Asymptotic expansions originate from integrating a series (i. n=0 (20. 1 we obtain oo F(z)£(*)" = — .n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. In any case.n _ En+\ n! En+1 ^ (!)"«!. Q. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0.e. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. El . In this ideal case the integral of Eq.
for < ^ 0. we have a quantum mechanical problem where probability can leak away. 20.2. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. The potential (20.2. S... This is the case if the real potential has a hump of some kind.20.. i.9) The question is therefore: How can we determine 9 E(h). (20. Popov and V. (20. e. $s E can be calculated by the WKB procedure.10) is considered in detail in this work. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).1 Introductory Remarks 475 except for possible subtractions.3 and g < 0. if the potential is V(r) = +grN. § where 7(ff) = To exp  . ? When ?s E ^ 0.g.. Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) I jQO Ar = 7T JO tir~l%E{ti)dti. 20.3 The potential V(r) gr».10) r as shown in Fig..11) V..N = l. M.e. (20. ... or can we determine Ar independently? Fig. N = l./ \p\dr\. Weinberg [228]. and E represents an eigenenergy..
Eletsky. Popov and Weinberg. the BreitWigner formula (10. Here n is the principal quantum number of the Coulomb problem (see Eq. However. Solution: The solution can be found in a paper of Eletsky.11). (20. as we expect on the basis of Eq. Example 20.76) (see also the comment on the comparison with the Kepler problem before Eq. Popov and V. Thus in the following we apply these methods to our typical examples.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. (11. (14. of course. 20. M. Equation (20. 8.113)). V. (11. as was (and still is) not the case for t h e equation 1 V .73) in the W K B or semiclassical approximation.11) is. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations.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.9) for Ar. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. S. 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 . . (20. L. the perturbation method of Sec.134)). In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. An interesting application is provided by Example 20. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients. the eigenvalue expansions of the cosine potential and of anharmonic oscillators.1. the cosine potential being t h e most completely investigated will also be seen in the following.476 CHAPTER 20.7 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. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. Weinberg [83].^ The logarithmic factor in Er is a novel feature of this case.
are given by i . obey the recurrence relation jPi(q.26) and (17.20. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. (20. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.12) (20.e.l(q. (17. Eqs.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. B. The recurrence relation (20.13) In the special case j = i the boundary conditions stated after Eq. M2i = 2j2 J[Pi(q.l ) ( g + 4 j . % = Aq.h) = 2h2 + 2hq+j. j . Miiller [74]. i.Cq.j) of functions ipq+4j. h). arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.58) and (17.34)) that the coefficients Pi(q.14) Thus these coefficients can be obtained once the coefficients Pi(q. (cf. Eqs.J) = (g + 4 j . J.M2i+1 for large i. Dingle and H. (17. (17.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.2 A = Mx + ~M2 + 7^WM3 + •• • . 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.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.3 ) P i _ 1 ( g .l ) +2[(q + Aj)2 + l + A]Pl. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. (17.Bq.j + 1). W. (20.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.13) represents effectively a partial difference equation. (17. 3)? . .55)) X = E(q. Eqs.j) a n d M2i can be deduced.60).j) are known. or equivalently of the quantity A.
Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.18.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . also written lCj.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign. Using this relation and the duplication formula lw('\W±^.l (20. Hence exact Borel summation is not possible.478 CHAPTER 20. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.*.91 l)]![i(g3)]!} 2 { [ .16a) (20. .17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods. see also below) .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. (20.•91 {[l(? + i)]![(? + 3)]!}2 (20.
J. The potential we consider here then has the shape shown in Fig. when this is g0 = 2n + l . f H . 2 . \ 1 \ \ \ \ \ 0. 20.3. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. the difference q — q$.* However. n = 0 .2 solutions obtained earlier. this difference was found to be real. Stone and J. . Reeve [262]. 20. remains unchanged.3 Cosine Potential: Complex Eigenvalues 479 20. . What changes as a result of different boundary conditions is the deviation of q from the odd integer qo.3 20. in the calculation of the level splitting as a result of tunneling. we follow here a formulation in the context of our earlier considerations of the Mathieu equation. but with different boundary conditions.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/?. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.20. . This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2.4.4 The cosine potential shifted by TT/2. W. 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.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. . MiillerKirsten [200]. At the risk of repeating what we explained earlier. we emphasize a few points *M. We retain the previous boundary condition at one minimum. Previously. l . i.5 \ \ / \A Fig. A means to achieve this was found by Stone and Reeve. .e.
20. X + 2h2 = 2hq+—.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . but achieves exactly what we want. 8 the equation can be approximated by d2ip(w) (20.4. as we found these for this potential.4. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear).T K ^ a D_i_{q+1){±iw).480 CHAPTER 20. Thus one chooses not the solution with real argument at the minimum. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.20a) As stated. a boundary condition which at the face looks somewhat abstract. i.2h2 cos 2z]ip{z) = 0.Ah2z2 • • • }^{z) = 0. we shift the argument by n/2.21) <u> + ( i « .e. Here we choose t h e second way. but t h a t with complex argument there. (20. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. Setting w = 2h1/2z. Effectively we consider the tunneling from the left well in Fig. 20. We can distort the potential there or alternatively impose a different behaviour there on the wave function. Now we change this situation at the minimum on the right. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . The physical wave functions there are therefore the solutions which are real and exponentially decreasing. 20.4 (around z — 0) to the one on the right (around z ~ ir). (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. Previously this situation was the same at the neighbouring minimum. . which means replacing z by z±ir/2. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. and allows its imaginary part to tend to infinity.
24) Considering t m O i n Eq. such that this deviation becomes imaginary. . A derivation is given subsequently. We consider here explicitly only the case of the ground state around z = 0.e. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. Thus. (20. we naturally constructed the even and odd solutions with these (cf. .23) .22). whereas here we construct these from oscillatortype solutions.4. i. whereas here we start off with a solution even or odd about a minimum. (20. A (which we can loosely dub solutions of WKBtype). The calculation for the general case is nontrivial and so will first only be written down and will then be verified by specialization to the case of the ground state.20. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions.> 0 for z > ±ioo. Since the large/i2 solutions valid away from the minima were the solutions of types A. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. 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. 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. 17. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. 2. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. . . 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. 20. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i.e. by a different method. that at z = 7r. we see that ips ~ cos \j2qhz for z ~ 0. . 3 .63).3 Cosine Potential: Complex Eigenvalues 481 As explained above. Eq. however.
g.g.28) Extracting Du{—w) we have A. 92. (20. We see this as follows.482 CHAPTER 20. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. W. 2. Oberhettinger [181]. 17.(.24) are linearly independent around z = 0 for q ^ qo. and then evaluation at u = 0.w) = e T Dv{w) 2TT n») AvViDv^iw). The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{. This is the feature which gives rise to an exponentially increasing contribution in the neighbourhood of z — 0 and hence provides the dominant contribution in going away from the origin.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing. §See e. we know from the solutions of types B and C in Sec. Thus the only nonvanishing contribution comes from differentiation of Y[—v). Magnus and F. tp(v).h(2h1/2 J £2=0 (20.25) qo) E2 0: (20. W. p. . However.27) differ by an angle IT.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. From Tables of Special Functions'.•w) + 2ir T{u) <"+!)% Dvi(iw). p.30) 'See e. (20.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. The arguments of the functions appearing in Eq.29) Equation (20.27) requires the differentiation of this expression with respect to the index v. Oberhettinger [181]. Magnus and F. this can be handled in a few lines.3. (20. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. (20. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq.
4.29) v=0 (in + In w)D0(w) + v2ire (m + lnu. W. p. which are valid around z ~ 7r/2. 5 7 7 .29). "See e.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.//n and ij>{y) = C .)e" w /4 2 l ^Dx(iw) (20. r(i/) We now return to Eq.e.. i.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. This result is » not easily generalized to <jo > 1.g. Magnus and F. Stone and J. . Thus calculations for qo ^ 1 are not given by M. i/^0 483 T(v) It follows that d dv [T(u)\ (20. 3 .+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. . Oberhettinger [181]. (20.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.33) = IV2~TT. and "Prom W. Differentiation of this equation and subsequently setting v — 0 yields. (20. . pp.27) exponentially decreasing contributions.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . Magnus and F. Oberhettinger [181].32) which we can use here since all other problems have been resolved: d D {w) dv v (20. including the derivative of DE2ufl(2h1/2z)).31). dropping from the derivative part of Eq. . For u — 0 one obtains the result (20.20. (20. We now go to the WKBtype solutions. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. Reeve [262]. those of type A. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. we obtain 4>s D0(2h^z) D0(2h1/2.e. 92. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20.
36) V'WKB cie~2hehz' 9h  . we obtain the required solutions from the earlier ones. Then V'WKB becomes (with cos z ~ — 1 + z' /2. To this end we set z' = z — ix. (20. near z = n the boundary condition (20. cos z/2 ~ (20. (20.29) and (17.sm I We now want to match this solution to that of Eq.40) From Magnus and Oberhettinger [181] (p.40).38) in its limiting domain towards 2 = 0. y COS c\ J (20. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). 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. (20. smf (20. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .41) 2TT . Eqs.cos z — 2/icos z ipA^ Tcos  and i n A —. Recalling that we have shifted the potential by 7r/2. i.36) cos  . (20. i.37) Comparing Eqs. where we can then match these to the solution ips.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. that in approaching z = ir.— 721 . 2 c2e2hehz UJ2 .39) Now.e. We now return to the WKBlike solution (20.35) We construct with constants c\. x .34) and (20. (17.36) and consider this at the other end of its domain of validity.e. (20.. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left. We observe that in the matching domain (i. by there replacing z by z + ir/2.34).484 CHAPTER 20. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. Thus in the present case and with q ~ qo = 1 we obtain „2ft.e.
39) and (20. i.e. We derive the general result in the next subsection using our own method. (20. .34) we obtain E2 = i%\ = ±ih22Qe~8h.38) can be written TAWKB (20. J Hence Eq. (20.^. ^'mliw <2a47) For qo = 1 we recover the result (20. with c\ = exp(—2h) from above. (20.e.46).43) we obtain.43) Comparing now Eqs.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2.1/V)_le^_ Z With this result Eq. (20.20.46) Repeating these calculations for the general case (i. The general result stated by Stone and Reeve [262] is therefore presumably guessed. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.45) Comparing this result with tps of Eq. 1 2 4 1 2 8h ^^2 2 (20. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l .3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real. (20. 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 .44) = DQ{2h / z) ± (2h7r) / e. (20.2/.
Ah2 ( z + . We have to impose two sets of boundary conditions. (20. MiillerKirsten and A. H.46) we consider a half period of the cosine potential as illustrated in Fig. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. (20. i.2 CHAPTER 20. and expanding cos 2z around the point z = — TT/2.e.z. •+(=)* V(f) 0. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . Wiedemann [4]. at z = TT/2. On the other side of the barrier at z = —TT/2 we consider different boundary conditions.e. Achuthan.*. 20. (20. i. Fig.E curre it \ j. i. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. W. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .5.20a). J. Eq.()#0.486 20. At the minimum to the right of the barrier shown in Fig. V(z) .e. the result is formula (334) there. Considering the original equation.5 The cosine potential from — TT/2 to n/2.) + 7T V> = o. 20. 20.48) These conditions provide the required quantum number qo = 2n + 1. 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. . •_(:)*<>.5.3.
51b) 2i(«1)[i(9l)]'.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum. Recapitulating these from Chapter 17.20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. as shown earlier.z).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.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .h.e. i. (20. For the continuation we use again the WKB formalism.r/2 ~ e i(2hq) V 2 . B and C.h.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. (20. Proceeding as in our previous cases and substituting the expression (20. we demand that ^(*)L<.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. we have IPA(Z) „2hsinz Aq{z) + O il>A(q. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. 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. three pairs of solutions of types named A. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2.z)=tpA(q.21)). (20.21) into the Mathieu equation we obtain. (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/2.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. Here the upper sign refers to the even states and the lower to the odd states. (20.e. 5.48) and proceeding as in Chapter 17.=F2 2\1/2{28h2y/4e~Ah l + O (20. Finally the third pair of solutions is Cq[w(z)}+0 Cq[w(z)}+0 Cq[w(z)} where Cq[w(z)} = 22(9+1) •((73) 1 h1/2 1 hW> (20.57) where go = 1) 3. We return to this condition after derivation of the .53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. we obtain our first transcendental equation. \vw).55) Extending these solutions to the domain around the minimum at z = 7r/2. the relation q . (20. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. and the second around z = 7r/2. we obtain 1>±W = \ a a (20.488 CHAPTER 20. (20.56) Applying the boundary conditions (20. i.52a) = Cq[w(z)}.^ i .52b) The first solution is valid asymptotically around z = —vr/2. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . We saw in Chapter 17 that some of these solutions can be matched in regions of common validity. ^ W H .qo .
r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .20. p. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A.2hq] / dz) We know that the solutions I/JA{Z). (20. i.67).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 .49) and we can apply our boundary condition. (20.[ [Ah cos z .58) Ignoring the correction A / 8 . Byrd and M. e x p ( . I^A(Z) overlap parts of the domains of validity of these solutions.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. as we shall see with the result of Eq. formula 280.Ah2 cos2 z ~ 0. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. . Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum. D.Ah2 cos2 z ip = 0.01. F. 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. Friedman [40]. Our procedure now is to match these to exponential WKB solutions. Thereafter we extend the latter across the turning point down to periodic WKB solutions. Jz+ ' T In principle the integral could be evaluated exactly — see P.e. These then provide complex exponentials like (20. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. we see that the turning points at z± are given by 2hq . and we obtain the proportionality by appropriate expansion. Some nontrivial manipulations are again required to uncover the generation of important factorials.61) [Ah2 cos2 z2hq]1/2dz.
) +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. (20.59) 2h COS Z+  ~ 0.490 CHAPTER 20.63) We handle the logarithmic terms in Eq. we write cos2 z = l . We have for the following expression contained in (20.sin2 z = l .62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.64) . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. (20.62) as follows.cos2 ( z+^) = (z+ .» _ i y v .
/ * [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\"  . 1/2 (20.62).60) by (4h2 cos2 z)1!4).63) and (20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .+ qj V2 4 A. (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I .67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .50b).3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . We ignore this here.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.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. (4^2)i/4 [ i( g _3)] ! V/ l (20. . .64) into Eq.66) The WKB solution therefore becomes (approximating the denominator of (20. .50a) and (20.20. (20. on using in the second step Eqs.
(14.. i 27r(8^/ 4 2/t /4 ' r(±) .i ( 2 / i g ) 1 / 2 ( z + . (14.55) 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+.492 CHAPTER 20.™) .49).z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i).—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { . S1H (2/ KZ ) / (z f 2 (2^). we obtain T _ ( ± ) = 0. Thus with IPA(Z) and TPA(Z) taken from Eqs.l^l !T ^^* T2 7 r ( ^^ /.KW4^8 4 .50)).?/ 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.U ^K^.69) l(8M9/4e2^ \2 [ i ( g . i. 1 2 1 ±2 . (20.z)}].2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T . we obtain l(8/i)g/ 4 e.1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .2 [ ? ) Imposing the boundary condition (20. where _ 1 + [±) *) + £ (20.49). Large Order Behaviour of Perturbation Expansions again using the Stirling formula. io(8/*)g/V^i6ft2)V4[±(g3)]! .67) and (20.+) + B^) HAh^/\ir. Eqs. (20.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.68) 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.T 7 77 r )V2 /4(2 . We now return to the even and odd wave functions defined by Eq.
in fact. Eqs. Equation (20. who derived it for qo = 1 and guessed the form for general values of qo.e.qo) El Ah' (20. (20.72) We observe that with the square root as on the right hand side both sides are complex. We now have the two equations which have to be satisfied.25)) E = A = Ei + E2.26). (20.72).20.57) and (20.h) = = E(q0. In both cases we wrote (cf.h)+2h(qq0) (16h)i°+le'Sh E(qQ.3 ) I = }(»« !7rl/22I(.47). Ex = 2h2 + 2hq0 + A 1 (<? . Eqs.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .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.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20. the result (20. (20. (20. i. Combining these equations by replacing (28/i2)<?/4 in Eq. h) of Eq.74) This result agrees with that of Stone and Reeve [262].h)+i%Ego.21) expanded about q = qo.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .75) . (17. 20.72) is our second transcendental equation.h) + (qq0)( BE — 90 E{qQ.57) by the left hand side of Eq.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. for q an integer both sides are imaginary. (20.3. (20.h) + i l + O '4{[^ol)]!}2 E(q0. we obtain E(q. (20.73) Inserting this expression into A = E(q.
I g  ± l ( 2 „. (20. 25Eqo=8h[ 2\1/2 . (17. 20. we can recalculate with the help of Eq. W.9) the coefficients Ai of its perturbation expansion.) {I6h)™/2e4h 1 + . Bogomol'nyi and V. i. Fateyev [26]. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. (20. MiillerKirsten and A. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions.74b). B.3.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. H. .4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE.80) *P. by others. A.e. f E . A = .* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. The same formula can be derived in the case of the double well potential. the discontinuity across its cut from zero to infinity.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. Setting ng AEqo:=2iSfEqo. Achuthan.W [£(901)]! < (20. Eq. Wiedemann [4] and Example 20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. its tunneling properties and the large order behaviour.494 CHAPTER 20.2 below. J. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. 7 8 ) i=l V . We had A= or _2ft2 + 2ft. cf.+  .
(20. MullerKirsten [200].20.1).l ) ( < ? + 4 j .13) out again: jPiiQj) = (g + 4 j .3 ) P i _ 1 ( c ? .81) We observe that this result agrees with the result of Eq. f H .74).j) satisfy the recurrence relation (20. We also observe that these terms do not alternate in sign. (20. J. (20.j + 1). Dingle [74]. (20. a = J + \<1> (2084) . j . and the method can be applied in many other cases.83) One now sets ftH = ( 2 % P * ( 9 . J). *R.j) + {q + 4j + l)(q + 4j + 3)Pi1(q. (20. the integration is seen to give (20. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. For convenience we write Eq. there is nothing special about this case. The integral is recognized to be of the type of the integral representation of the gamma function. But as Dingle remarks. B.. W.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.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo .13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq.14).19) for the large values of i under consideration here.l ) +2[(q + 4j)2 + l + A}Pi_1(q. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. so that an exact Borel summation is not possible. 20. Using again n = (go .
(20. Pi(a) i.h) = ^2pi(a) with p. (20.27h pi(a) a = 4j(4j + 2 g .1) + 2 (a — 1) a . Abramowitz and I.83) can be written j .l. a.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.z)WaA/2(z) Setting V. we regain Eq. (q + 4j) 2 ~ 4j(4j + 2g). d2WKh dz2 + 1 4 K z zM2n WKtll = 0. (20. e. we obtain (2a .85) Considering now the generating function of the perturbation coefficients.89) Replacing here K by a and \x by 1/2. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a . Stegun [1].^(Z).1/2(z) (a1)! (20.e. 4j(4j + 2g + 4) + / .91) + Wa+lil/2(z) = (1 .4 ) ^ .496 CHAPTER 20. Taking dominant terms. g < 4 j .a R . the recurrence relation (20. A.l/2l Wa.h) +p(a + l. .86) one can convince oneself that Eq. 1  1 / 2 (4 (20.e.90) See M.h) 2+ 8h a p(a.85) implies p(a .4.1) + 2 P i _i(a) + Pi_i(a + 1).87) Taking terms of 0(h~l+1) of this equation.g. i.h). Whittaker functions WKjfl. z2 (20.31. Now. p(a.ll{z)= L . nx pi_i(a + l).85). 0 4j(4j + 2g) pi_i(a .(a) oc . (20. i. (20. (20. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i. which are solutions of the differential equation known as Whittaker's equation. formula 13.K + ^ ] L + K^)WK.e.
^Observe that if one solution of Whittaker's equation (20. (20.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.20. .h) = \~\Pi(&) with pi oc h l. (20.86) have to be chosen such that p(a.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. we have to remove a factor (recalling a = j + (7/2 of Eq. (20. h —> —h. g(h)(20.4 Cosine Potential: Another Method of Calculation we obtain Vai.h) ~/(fe) ' (a1)! . h) is expressed in descending powers of h.(« . In the present case this means the replacements q — —q.93) (a1)! ' where f(h).(« 1) 2 ].(« .i/2(z). Oberhettinger [181]. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged.88) is known. such that p(a.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo.84)) {8h)q/2( Ah for j < 0. p.94) Hence to insure that p(a. > (20. v . We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20.e.\)V .g(h) are functions of h which in view of Eq. i. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).92) Comparing Eq. > > ^See W.. 89. a (20. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . (20. Magnus and F .i/2(z) + Va+lA/2(z) = 497 2+ Va.j — —j. Dingle discovered the solutions^ p(a. h) is expressed in descending powers of z = 8/i.87).92) with Eq.
83) and is given by 1 1 .8 / t ) for j > 0. Correspondingly we set for j > 0 / 1 A pla = j + q. W. Muller [74].). h\ oc (8h)q/2e~4h ( _ . n = 0. B. Dingle and H. This expression can be obtained from the recurrence relation (20. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .95b) Then for j > 0 we obtain from Eq.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. j — —j. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2..+ i f or 3 < 0 (20.94) here we have a series in descending powers of h. we obtain the result of Dingle: (20.. (20. ^ . i). J. (20. _ i s.96) and a similar expression for j < 0 with /i — —/i. 2. g — —q.95a) Inserting (20.1. (20.498 CHAPTER 20. Comparing > > > this result now with Eq.j _ g / 2  1 / 2 ( .l ) ] ! 11 R.. .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.1/2(8^) ( .
(20. formula (30).97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g .0) = 1 as desired (although not really enforcible for large values of i).l ] ! {[. note the misprint there: (r — s) ought to be (r — s)\. p.20.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. B. l{Q.3)]!2 2 '+^.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.100) **R.157(4). Thus ° 9 (i + \q) 2«(i+Igl)! p .l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.+i?_l]!}2 (20. M. Miiller [74]. A special case of the formula can be found in I.e.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. Ryzhik [122]. is M2l = 2j2j[Pi(<l.97c) We assume that in a leading approximation we can cancel the factors (i + \q). (20.{j + \<l).14).98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. J.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. W. E* J'=I i+M 2 [2i + 9 . formula 0. .97d) Then the coefficient M<n of the eigenvalue expansion. Dingle and H. S. (20.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. of Eq.l ] ! (20. Gradshteyn and I.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.e.j) (i + lg)[2i + i ( g . 5. i.
i.i ) ] ! [ ^ +  ( g . replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. i\ M i large 1 . [19]..o i • . since \{q — l) + \{q — 3) = \q.' 1 i • (20.105 ) the latter because we see from the eigenvalue expansion (18. the result of Bender and Wu [18]. (18.. 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. n = 0. (18.3 ) ] ! _ 1 Hence M or. [^ + i ( g . The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.e..86).5 20.e.1 Anharmonic Oscillators The inverted double well In Chapter 18 we obtained the imaginary part of the eigenenergy E of the Schrodinger equation for the inverted double well potential. .86) and set (cf.. i.e.81) for q — qo = 2 n + l .. (18.5. 20. i.17).101) This formula allows us to cancel all factorials [i.e.34) that £ is an expansion in ascending powers of h 6 /c 2 .1.l l ! with the large i behaviour in agreement with the result (20.Am i . Eq. i.. Eq. i.l.2.102) Thus the result is M m 1 2«Fi + g . We refer back to Eq.4)) £ = ^ = ^+ w and y= ^ > ( 20 .]\.e.500 CHAPTER 20.
107) we obtain (i)r r 71" JO yirZ£{y')dy'. (20.(20. in terms of K where go = 2ftT + 1. We leave this to the following Example.e.20.109) or. (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.4) of Bender and Wu [19]. (18.5. and then the relation (20. and hence inserting here the imaginary part from Eq. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . 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. i. the discontinuity across its cut. 20. in fact as one with Borel summability in view of the factor ( .106) With £= E r=0 Mr (20. .l ) r .2 The double well It is clear that one can now proceed and apply a nonselfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue. The result establishes the series unequivocally as an asymptotic series.77) can be verified.86). This can then be compared with the level splitting we obtained for this potential. i.110) This is the result in agreement with the large order formula (4.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.108) Proceeding in this way.e.
G. W. Solution: For the solution we refer to the literature. Harris. J. J. Grecchi. J . of International Workshop on Perturbation Theory at Large Order (Florida.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. Nakai. *R. Achuthan. Eqs. H. M. as in later chapters. W. Cizek. Wiedemann [4]. J. Propin. K. K. MiillerKirsten [164]. J. t The relation (20.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161].77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. M. Cizek. S. E. V. Silverstone [62]. from p a t h integrals with instanton methods) and the large order behaviour of the perturbation expansions which establishes the latter clearly as asymptotic expansions. R. Achuthan. R. W. R. MiillerKirsten and A. Paldus. 1982) [139]. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). J. Bender and T. Wiedemann [4]. *Some of the literature in this field following the work of C. T.502 CHAPTER 20.^ 20. S.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. Wu [18]. J. Harrell. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. Grecchi. Large Order Behaviour of Perturbation Expansions Example 20. S. J. + 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. Paldus and H.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. t p . Damburg. in t h a t when the quantity on one side is known. Harrell. it can be used to obtain the other.Q . M. J. J. J. . J. V. Propin and H. Gram. . 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. MiillerKirsten and A. [19] can be traced back from articles in the Proc. Damburg. The relation (20. E. Graffi. (357) to (371). " P . Its usefulness has also been demonstrated there. Silverstone [52].77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . Liang and H. A relation similar t o (20. J. H.
Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. The formalism is not so useful for boundstate problems (in this case the initial wave function is. Feynman and A. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. However. Feynman [94]. This formalism is particularly useful for evaluating scattering or transition amplitudes. can be found in the book of B. A further wellknown reference is the book of L. S. Probably the most readable introduction. Hibbs [95]. *R. A standard reference is the book of R. however. Coulomb problem) we refer to existing literature. Feynman [93]. For the application of path integrals to the hydrogen atom (i.e. We illustrate the method here by rederiving the Rutherford scattering formula with this method. P. Kleinert [151]. 503 .* 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. Schulman [245]. A brief introduction is also contained in R. Felsager [91]. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. R. Chapters 2 and 5. since in such cases the probability of this system choosing a path far away may be considered to be small. in general. 'See in particular H. The path integral is. P. P. 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.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. we digress later a little and consider the path integral also in simple contexts of scalar field theories which can be treated like models in quantum mechanics. ^ Our interest here focusses on quantum mechanics. unknown). However. which explains also difficulties.Chapter 21 The P a t h Integral Formalism 21.
ti\tp).ti)ip(x. =J dx(xf. 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). Obviously K(xf. i.504 CHAPTER 21. (21. (21.tf\x.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).5) Sn = X / fc=l mo (zfe . U. ti). (21. ti) = 5(xf .tf\ip) .2) where K is a Green's function.1) Let the wave function i/> at time t = ti. we can compute the action S1. we wish to know* ip(xf. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.ti). The Feynman Path Integral Formalism 21.tf.x).e.e—»0 Sn.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. the initial time.2 P a t h Integrals and Green's Functions The onedimensional Schrodinger equation for a particle of mass mo moving in a potential V(x) is (with mostly h = c = 1) i d2 + V(x) tff(x.ti){x.t). x. 2mo dx2 (21. be ip{x.tf) = / dxK(xf. */ = U + (n + l)e. (21. *In Dirac's notation we write this equation later (xf.x.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.
of which one is illustrated in Fig.tf).21. 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. 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.t).ti) to its final position at (xf.28)): f dr\ exp ia a) ' (21. as will be shown to be possible below.6a) . the Fresnel integral (13. The usefulness of the method therefore depends on the fact that some paths are more probable than others. 21. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi.1 A path in (x. Here and in the following we need one or the other of the following integrals (cf.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. 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. (21.tf. 2 + 2a.z f c _i) 2 } } .5)): G(xf. for example. ie f me\ 2a\a~) 1/2 (21.0 ^7 (x f c + i ..(x fc+ i .ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT. We then integrate and finally let /x go to zero. dx exp z—{a.x f c _ i ) 5 \m0 J (21.37)) dxk exp d(xk .xk+x oo —oo oo .m0 .s f c _i) + (xk+i .mQf Xk L ~2e^ ^ .506 CHAPTER 21.Xi. one has j ^ dr1eifa+i^ /£ = . dxneiSn/n.* 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.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) exp .t ~Xk+l> + xk+i 2 ) +{xk .Zfci) 2 } I +{xk+i x*..6b) —c ia o drjr] e x p—V € (21.6c) In the evaluation of these integrals we replace a by a + ifi.2 + (x + x f c + 1 .ti) '•= f g m ) (N(€\]n+i dxL. tf > U. n > 0. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2. J—oo dxneiS/h — • 0. One therefore removes the troublesome factor by defining (Sn being the sum in Eq.xk_i) dx exp i — { 2 a . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.8) Note t h a t in spite of the n integrations.
183. 183.tf. 21. and will therefore not be attempted here. p. Felsager [91].t In the following we demonstrate that the function constructed in this way is.Xi. (21. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. nor the limit of integration of some variable Xi.2 The Path Integral and Green's Functions 507 Eq. (21.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).U) = / JXi=x(ti) = K(xf.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration.e.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. However.5). the Green's function K of Eq.g. B. § See e. 11 For discussions see e.ti) for tf > U. (21.tf. and so on. however prefixed with a new factor. .tf.1.5).9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman.Xi. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. Different constructions of these families of paths usually lead to different measures. we arrive at a path integral without the measure factor N(e) above. It is neither a variable of integration in T){x} —> Ylndxn. summation over all possible paths.24).ti) For tf > ti we now write* G(xf. G(xf. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. The consideration of the denumerable number of possible paths connecting the particle's initial position with its final position may be handled in a number of different ways. (21. We shall see later (with Eq. See the square root in front of the exponential in Eq. (21. (21. (21. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims.2).7) for n — 1) to give a time interval. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. i.e. These integrations are indicated by the horizontal arrows in Fig. it may be noted that with a different philosophy concerning the summation over paths. which all vary from —oo to oo. as in our discretization in Eq.26)) that — ignoring the potential — the multiple integral (21. i. B.21. Felsager [91].Xi. in fact.g. p.
3 in the evaluation of I a specific path integral (see Eqs. 21. I We employ this method later in Example 23.72) to (23.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf. pp. (23.xi. the other with that of the action of the free particle SQ.tf. one with the full action S. It is for these reasons that — unless required — the measure is frequently ignored.1 and allowing n to go to infinity.xi. Felsager [91]. Expanding this in a Taylor series around x. we have See. . Then we have on the right the factor G{x+rj). however.ti).ti) = 1 j^p: f°° cte n +iexp } • fm0.Xi.74)).tf + e. the discussion in B. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. so that the piecewise linear paths fill the entire space of paths.tf + e. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.exp 2e n ieV(xn+2) G(xn+2 + V. we have G(xn+2.ti. Then the measure drops out. we obtain the same as before except that there is no factor N(e). 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. We have G(xn+2.508 CHAPTER 21.U) imo 2 G(xn+i W)Idr. (21.188. Each of these linear pieces is completely characterized by its endpoints.Then we can construct a recurrence relation as follows. Consider a time t = tf + e for position Xf = xn+2. 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.Xi. 187 .
xi.tf.Xi. tf.ti) +\if(^G{x.xi.tf + e.tf. U) = e^G + 0(e 2 ) dtf 1 d2G(x.x.tf.ti) for tf > ti (21.16) . tf + e.6c). It follows that ip(xf.ti)\ .ti) .G(x. . tf.tf.tf.Xi.Xi.xi.2 The Path Integral and Green's Functions so that G(x.U + 0(e ). tf.21.t Thus + 0(e2 (21.6b) and (21.ti) N(e) or G(x.Xi.— p G ( j .ti) = G(x.ti) +V(x)G(x.Xi. Retaining only terms up to those linear in e.tf.14) = G(x. we obtain: G(x.ti) = 1 d2 + V(x) G(x.Xi.6a).ti) = j^rr dnexp I ^rj1 . Xi.xi. U) + .tf.ie 1 d2 2nV0dx^G{X. U) . U) + Q(e2 = (lieV(x) + 0(e2)) G{x.U).tf + e.xi. ti) satisfies the timedependent Schrodinger equation.tf)= dxG(xf.15) We see that G(x.tf''Xi.ieV(x) + ri< —G(x.xi.xi. d and this means d i—G(x.13) We integrate with the help of the integrals (21.tf.Xi.tf + e.tf.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.tf. tf.Xi.U) +••• \ 509 G(x.tf. (21.ti) 2 —l€ — ^2 + V(x)G(x.xi.(21.Xi. 2mo dx2 (21.ti)ip(x.
ti) = = 9{tfti)—G + G5(tfti) . is the Fourier transform of the nonrelativistic propagator.tf.x.17) Differentiating K with respect to tf we obtain —K(xf.U) = G(xf.tf. (21.U). (21. (21.3.20) .8)) Go(xf.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. We have therefore established that the Feynman integral (21. (21.1)).tf.Xi. 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.x. The function G which satisfies the timedependent Schrodinger equation (21.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.U) = / T){x} eiS\ [fdt \mx2. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation. Hence our original function K is the same as G.U)6(tf . or K(xf. In a problem where V is small it is convenient to calculate K with a perturbation method.tf.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.16). 21.10) is indeed to be identified with the timedependent Green's function. 21. Eq. as this is also called. now called So.x. denoted by KQ.U)G0. (21.ti) = 5(xj — x). It follows therefore that {{W ~V+2m~0~^) K{Xh t/. K0 = 9(tf . and the Green's function KQ\ S0= Here (cf.510 CHAPTER 21.15) is a wave function (cf. Eq. The result.U. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. as may be verified by inserting this into Eq.x.
23) for Go and obtain Go(xf.tf.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.ti)= iX (o. exp where Xj = xo and Xf = xn+\ and there k = 1).Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq. (7. (21.Xi.tf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x. (21.24) ( s / .7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.25) T h e quantity Ko(x. We obtained the same expression previously — see Eq.37) — by explicit solution of the differential equation which the Green's function satisfies.ti) Ko(xf.21) fe=l and t = (n + l)e = tf — U and h = 1. f c .tf.i ) 2 .Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.Xi.tf. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.ti) = = G0(x.. (21.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.ti)9(t)./ . Then Go(xf.22) • \ n/2 . (21. 2t Go(xf..3 The Green's Function for Potential and hence Go{xf. In Example 21.Xi.tf.21.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function. But t = tf — ti = (n + l)e. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.tf. .8). exp 2 (21.Xi.z / c . we have Go(xf.Xi.1 we show t h a t oo /.23) We now insert our expression for In into Eq.t) K0(x.Xi.x 0 )" n + 1 (in agreement with Eq. n + i .
2y(xn+2 .22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:.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. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x0. For n = 1 (see below) h This result can be verified like Eq. Solution: The result can be obtained by cumbersome integration.)!} J — OO . (21. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() . as / JXi f r D{x{t)}eiS°lh ._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.1: Evaluation of a path integral Verify Eq. In our proof of induction we assume the result is correct for n.ti..7) by direct integration with ^ .^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.f_^y e. i.25) we can rewrite the Feynman path integral or kernel (21.tn+l•xn.512 CHAPTER 21.26) with n integrations and n + l factors KQ. Example 21.x „ + i ) 2 = ( x n + 2 Xi + Xi . (21. we have instead of 5*0 and KQ expressions S and K. If the potential V(x) of Eq.xi)2 = 21 xi ^2o(„.x ^ 2 + y2 .)+(1"+2. Eq.tn)••• K0(xi. 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.t0) (21. or else by induction.6a).10) in a product form.x „ + i ) 2 = (x„ + 2 .22). ^ (xi . ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l. (7.5) is carried along.54). the latter being a Green's function with interaction like that for the harmonic oscillator. Then oo 2 / (21. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .e. (21.xo)2 + (x2 .. (21.
29) Hence with KQ from Eq. (21.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.Q . n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 . The first. 21.Xi) + ( x n + 2 Xi)2.2.2 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.e. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.2 y ( a . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . (21. (21.x2rriQ 1/2 toij_.25): Ko{xJ) = = (S) e{t)(^X'2 y J . (21. ) H — (xn+2Xi) n+2 n+l n+2 y2 . Using Eq.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). i. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.27) we have (with a = t/2m0) . We pass over to momentum space with the help of the following two integrals. it . The second integral is an integral representation of the step function. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig.a .27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq.i) H —3/ .6a). n + 2 .3. e > 0 .i j ) H n +— z 1 .2y(xn+2 . (n+l)A n n j(a.. This quantity is also described as the nonrelativistic free particle propagator. and + 2 2 l n + 2 ( x n + 2 . A(n + 2) 21.n+2a.21.i) . a dp exp [ipx + iap2] = exp — i(21.
28) this becomes K0(x. ( 3 b) ° The expression E+ V — it 2m. 21. The generalization to three space dimensions (required below in Eq.*' where 2mo — r.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.2 Contours of integration.51)) is fairly selfevident. Thus in first . The Feynman Path Integral Formalism Im t \t>0 Rex Fig. This is the type of expansion most people describe as perturbation theory. an expansion in rising powers of the coupling constant (contained in the potential). 21.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£.e. The first order correction contains one factor of the potential V.Q (21. (21.514 CHAPTER 21. Replacing 9(t) by its integral representation (21. i. the second order two factors of V and so on.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion.
_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 ..33) xG0(xi. . dxr.U).20)) and with elS = exp i S i o / hi dtV(x. (21.ti. li f f 515 order (cf. .t)] n+l .Xi. Eqs. dxn exp i ^ —.^.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.32) Since — we recall this for convenience G(xf. .xi.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.t) JSo dtV(x.5) a n d (21.^ J_00 V(xk. x (n+l)/2 \ lim . (21.t)\.t) the first order correction is (G\ ~ G — Go) Gi(xf.xfc_ OO J— OO J—OO L J. (21. . dxi ( m0 V V27rie/ m0 \^He \ exp fe=.4 Including V to First Order: Perturbation Expansion h=l. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2. \2meJ n+l J. } e °° f l f •/ oo /"oo F ( a .t/...U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2.i V e / D { a .21.tf..ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.)} (21.tf.i.Xi.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.(xfc .i.^){iV(a.
t2)V(x2.e.t2) xKo(x2.36) {if •j.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'. t) Ju Thus the term containing n times V contains a factor 1/n!.t) (21. OO /"OO / oo dt / J—oo dxKo(xf.516 CHAPTER 21.e.t2.tf. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.x2. Xi.t)V(x.xi.U) xV(xi. = i dt2 / dx2Ki(xf.i / dtV(x.t) =li dtV(x.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). t2. in writing down the expansion exp i dtV(x.37) Jti The last two contributions do not describe the time evolution from ti to tf.ti)K0(x. (21.xi.t) ! Jti r*f dt'V(x. i.ti). The factors of "V" originate from the series expansion of the exponential exp ftf . It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti. dx2Ko(xf. t') + • • • . i.t. (21. Jti Consider the term of second order. This is cancelled by n! different time orderings.tf.x2. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.t2)V(x2. .tf.t)K0(x.I dtV(x. They must therefore be deleted.xi.ti). (21.ti) t2)K0(x2.35) We also observe here that the problem of timeordering arises only in perturbation theory.
tf.21. The diagrams represent mathematical quantities and hence are designed on the basis of rules.t.t)= f dpdEeipxiEtV(p. Here.tf. we are concerned with quantum mechanics.tf.x.xi.3 Feynman rules and representation of the Green's function. 21.Xi.Xi. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. E) which is given by (in onetime plus onespace dimensions) V(x.Xi.t) is V(p.t. For various reasons — such as illustration.40) The rules for the present case are given in Fig. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. U) (21. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. or in momentum space as mostly in field theory.Xi.t)K0(x.3 along with the diagramatic representation of Eq.t) ) K K K0 'V X K o Fig. of course. tf.ti).38) .tf. called Feynman rules.ti).x. x r l i 'V(x. The Fourier transform of V(x.39) dxK(xf.t2 K o (x 2'l2 .tf. 21.t)V(x. (21.ti) (21.ti) = K0(xf.E). (21. Such diagrams in the context of field theory are called Feynman diagrams.ti) i dt = K0(xf.i ] P I dt J dxKn(xf.39).t)Ko(x.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf. . These may be formulated in configuration space as below.t)V(x.Xi. x"^ X x"2.
i. . dn = (L/27r) 3 dk. with p = Kk. to one particle in all of space. for the allowed values of k: kx = 2imx/L. * k 1 • .tj)K{xf.ti). (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. We recall for convenience from **Note that from the wave function we obtain.Xi.e.41) We have to pass over to three space dimensions.7 = e i k ' x _ i : l i with f i/}*i/jdx. (3) The transition probability per second is = \A\2 . There the wave functions ipi. (21.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit threemomentum interval** = V/(2irh)3. mo V since ip is normalized to one particle in volume V. The plane wave function normalized to one over a box of volume V = L3 is.ti)il)i(xi.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.k = 2wn/L. and dn = dnxdnydnz. i. ^(x. 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).ipf have to be properly normalized.= l. (21. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. T = duration of time for which potential is switched on. The Feynman Path Integral Formalism 21. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf.e. here with E — k 2 /2mo.518 CHAPTER 21.t) = . 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.tf.
t) = e~^lT\ r where r = Ixl. P(*) = j r (2145) (cf.47) .— . u = ^\H'kfp(k).44) and f_oo can be replaced by fQ.k. V(x.4). The transition probability per unit time is . i.e. in which T is a large but finite time. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. (21.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential. # = ^an(i)VnM) n = a^(t) + \a^(t) + .21. Proceeding in this manner one obtains Fermi's "golden rule".. (21.44) = u(Ek. 47T 1 3 7 (21. H = H0 + \H'. Eq.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.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. «= — . Then H'ki can be taken out of the integrand in Eq. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. and with these the following perturbation ansatz: ih^* with an(t) = HV..117) and Example 10.= t7Ei4 1} wi 2 k It is frequently assumed that the only time dependence of H'(t) is that it is switsched on at t = 0 and switched off at time t > 0. (10.
ti)'ipi(xi.44) (here in the scattering problem with initial and final energies equal).50) Here x — xj — Xi and t = tf — ti.52) . Thus with the ansatz Ko(x.t) (21.44) by contracting the propagators to instantaneous point interactions.* / .i x (i)V(x. so that 0(t) can be dropped) K0(xf. T a large but finite time.f.ti).39) OO /"00 / dt / OO J —OO dxK0(xf. We can relate this amplitude to the amplitude ark of Eq.49) together with the following wave functions (of asymptotically free particles) 1>i(*i. This now has the form of the amplitude (21..t.tf)K0(xf.ti).tf)Ki(x. i. £.e. by threedimensional dxf and so on. dt dxip*f(xf. (x*. (21.tf)ipi(xf.>/>. Xj. ti) = co5(t — U)5(x — Xj) etc.t)V(x.Xi.520 CHAPTER 21..30a) K0(xf.tf.).49) i.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P . t)K0(x. (21.f. (21.x. we obtain A1 = clS(ti tf) / dxf(i)V(xf.tf. In our previous onedimensional considerations we had with Eq.x. (21.tfx.tf. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas. x.t)K0(x. t.xuti) = ^ J°° d p e i p x _ i ^ * .yLi. The lowest order approximation of A. (21. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x. we have to replace the onedimensional quantities dxf etc. also called Born approximation.tf.tf). t. Ai = dxf / dx. The threedimensional generalization is (taking tf » T and ti <C —T.48) where from Eq. (21.^ / . (21.tf)ip*(xf.51) Inserting this into Eq.t) = ^ J*ie*<*'*><<*'*>. (21.
~ . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .r) = = y . (21.).55) We evaluate this integral with the help of the following integral in which we complete the square in the argument of the exponential and set r = t . which is evaluated in Example 21.T) = JdtJdx(^\V(x.57) we have A(p.p 2. £). (21.k.59) .5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. (21. 2 „ x v~ m o (21.f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i 2mo (t — U) + ik • x 2mo .p ) .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.k. Then integrating over q and q' we obtain .54) Now inserting the expression for the potential V(x. this becomes A(p. k2 ipx + i—t + ikx — i 1 2mo 2mo (21.k.58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .T) = x exp L z/ q/2 k2 521 f dx. P2 .t)exp A(p.iaT2/8: L With 0 0 • . Then we require the following integral. .56) (21.53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.T) = ^JdXJdt%*t2/T2exp i ( k .21.k.i ( k 2 .
64) Inserting the result now into our expression for the total cross section. In our case this has not been done. <rtot. i. This expression assumes that the integration over time t is normalized to 1.63) we have: Probability per unit time = 77777.„ 2. e 8{k) = lim 7=*—^ (21. mo m (21 . (21. { I i _ F T l } .< ^ £ \M VTTT2.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .51 — 2 /w2 y ~ —.3. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T .62) We want to consider the limit T — oo.522 CHAPTER 21. k .T2f 8 V/ 2 . write down the probability per unit time which is ( 8 \1'2 _a2 16^ . e (21. / / P 2 _ . 66) o . dE = PjV = ^ E ± .60) Now. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).k. (21. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.P ) 2 + M2]2 4 VvrT2. ( 21 .\ 2?7lo / 1 (21. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. y 2 [ ( k .65) But E=*.e.^ . Using the result (21." + M ] p) ' "•"• ^2 .59).t o t= y dp ( 2 v r ) 3 "V 3 I W: 2mo y m0V k 32^a2S(^) y ( k y 22 [[ ( k _ p ))2 + M 2 ] 2 ' kP 2 (21. .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.T)2/T. T ) ~ . k2 . Using Eq. 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .63) We establish this result for T — 1/e in Example 21.„. otot contains the factor \A(p.a ( — ) ^ . we obtain .
(21. esfn (21.= e " f c a ^ o 2v^fa /4a = lim £^o —. Example 21.M £ M * = / m0kdn = m0^2m0E (k .^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.* 9 ' d r e .5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .e.p ) 2 = 2 k 2 ( l .67) and (21. we obtain f mlV2 dQ v / 2 m ^ 3 2 7 T 3 a 2 atot r) 3 fcV 2 4fc 4 4sin 4 (0/2) ' JW) da _ d& = i. (21. i. and inserting the expression thus obtained into the final result of Example 10.65) with the cutoff M > 0. —Mr 4w = .cos 9) = 4k 2 sin 2  . so t h a t (21. 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 .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 — .^ .68) Inserting (21. This result may also be obtained from lyukawa of Eq.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.21.e. 2mo / dft. giving the potential a coupling constant. Then (x = r) foo „2jj.j .69) This is the wellknown Rutherford formula.59) by replacing there q 2 by ( k — p ) 2 .. p 2 = k 2 .68) into Eq.67) T h e delta function here implies energy conservation.e . setting M2 = 0.70) .j .4. (21.^ . /Yukawa = ) Solution: We choose the 2axis along the vector q.
(21.oo J / )eikx. (21. We consider again nonrelativistic quantum mechanics.* . however. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. '<*> = .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 .70) and (21.. ' i a = lim e —0 pk 2 /e2 e^TT 21. (yx) = 6(y . We have x> = I J ^ l p X p l x ) .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.^0 a^C 2^/TTO. In configuration space the state of the particle is written x). (21. a simpler method suffices. In the present case. co oo dx co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). We write the state vector of a particle of threemomentum p in the momentum space representation: p).72) co From (21.q).x) (21.524 The function required is CHAPTER 21.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . Since dxe" 1 1 1 = « / .75) . a. 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 ) . 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. Va we obtain g(k) = Je~k V a = 0.6 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.71) with respect to k yields the same as °° 2 . Direct integration /4a . Differentiation of (21.
If '/'(x.£) = ( x V ^ ' . we must have d/dt{x.t) = +iH\x.t'x..81) Comments on signs in Eq. The time development is given by the Hamilton operator H.160.x. (21.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.t'.82) (21. i.i) = e iH4 x.e. Then the states {x.e. (px) = e~ipx. (21.We now introduce the Dirac bra and ket formalism.We define x. (with ft = 1) d/dt\ip) = —iH\tp).t) = —iHip(x. Hence for the infinitesimal transition (x'. which describes the propagation of the particle from its initial position x. d/dt\x.e. The equation 1 = J ^p)(p. so that \if>t=o)s = H>)H. (21.82): The general Schrodinger equation has the state vector to the right.t\ = iH{x. (21..t)} constitute a "moving reference frame" in the sense that ip{x.^ l x ) . at initial time ti to its final position xy at time tf. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.e.t> =iH\x. M^ip(x.e. i. x.t).t\.83) (21.t). See also L. Ryder [241].79) Feynman's principle is expressed in terms of the propagator or transition function (x'. pp. as had to be shown. i.i) = K(x'. so thattt d — x. H.t). 159 . i. Thus (qPp) = p(qp) = p(27r)3<5(q .76) These relations imply the completeness relations 1 = J dxx><x. t) = (x. t\ipo)s.i'x.80) (21.p).t) = e* H t x). t' — t. .21.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc .0) = e""x). x' — x infinitesimal. We consider first an infinitesimal transition. i. 4) is to satisfy the Schrodinger equation.77) (21.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). (21.t). Then ( X  ^ . ) H = (x</>o)s. Similarly we have a position operator X with Xx) = xx).
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. Pq) = J 7^3Pp)(pq) = J dppp>*(p . (21.* .85) the amplitude or expectation value or position space representation (x'#x) = .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).i p '. 2mo Since (21.p').. (2187) (21. (21. "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.90) (P'IVIP) = y<&.88) Thus P has the operator representation P= /dxx)V(x.P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.x') + V{x)5(x' .91) .y<&(p'ix')(x'ivix)<Xp) = f dxe.526 CHAPTER 21.V 2 < 5 ( x .89) Using this and previous relations we obtain for the Hamiltonian (21.x = V(p . i.e.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).x).
e. But e'ipx + 0[{t .92).p'). i. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj. the signs are reversed when these operators are applied to the ket vector \x. EQ — Hi . 527 (21. (x'.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.j.fx.t) (cf.82)). (21.84).p') + V(p . for the amplitude of an infinitesimal transition.tj).92) We now return to the expression (21.t'f / w? eip''x'^(p ~p/) = / l r ip''x' / "x"e~JP'x"v{x")e" = = Hence + y(x') 2771Q d /dx"<5(x'x")F(x"KP'x' y(x')e i p ' x '. We now consider the transition of the particle from Xj.p) + i{t .93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.p') • • • j W < 5 ( p ' .21. ti to x j . t).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 . tf subdivided into infinitesimal transitions (xj+i. We have <x'. i.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .i'x.95) when applied to the bra vector (x.t') \^(27rf5(p 2mo +V(pp' where we used (21.tj+i\x. (21.
i { f l " ( p i ..ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa .x i ) H h pn • ( x n .+ V(x f c ).xk (21. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.94) and thus obtain a phase space expression. . .. d x „ _ i .tfei) fe=i .99) ..t 0 ) + H(p2.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .a — —e/2mo and divide b o t h > sides by 2TT.. We now insert for each of the infinitesimal intermediate transitions the amplitude (21.528 CHAPTER 21. . then corresponds to the sum over all paths.98) tk — *fcl # ( p f c . zmo T h e n t h e integration over all p ^ can be performed by using Eq. x f c ) = ^ . x exp [ .27) in which we make the replacements x —> eifc.97) H hH(pn. (Xf.xn)(tn in_i)]. i..e.2.xic) by fc — Xfej H(pk.ti.tixo. T h e integration over the n—1 intermediate positions. d x i . x i ) ( t i . = 1 X X<jPfc We now replace H(pfc... d x n . (21.tf\Xi. .tj) = / ••• / d x „ _ i d x n _ 2 ..ti) rx/=x(i/) d x i dx2 .i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) .t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . t n _ i ) t„2><xi. .tn respectively. dx. x n which the particle passes at times tQ. t n  x n _ i .X n _i)] h) (21.i/xj. . d x i ( x n .tf\Xi. x i .e. i.x 0 ) + p 2 • (x 2 . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. . .
(21.6 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. A.e.H(x.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .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. (21. G. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. G.103) is another form of the pathintegral for which the integration with the help of Eq.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21. to / m0 \ 3 / 2 las. i.21. Truman [254]. Smolyanov. (21.. Tokarev and A.98). > Thus the phase space representation of the pathintegral* (21.102) We see t h a t the factor (with h = 1) m0 2m(tk 3/2 3/2 [N(e)}< = tki) \2irie) is precisely the normalization factor inserted earlier for the onedimensional case in Eq.101) Then Eq.e.tf\xi.8) as N(e) so t h a t the integral does not vanish for e — 0.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.] ex H ^I (.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. .p)]\. (21. / {xf.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. i.
e. We have now obtained the path integral representation of the matrix element ( X / . £ J ) with no operator in between. pfc) = Pfc • xfc . xfc). pk) = L(xk. this agrees with the corresponding discrepancy in the onedimensional case of Eq. i.tf\xi.iJ(x fc . (21. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. nl (xf.104) where for the particularly simple Lagrangian L under consideration here m0±l .106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t).x) (21. (21. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. Thus finally we have Feynman's principle: (Xf.H(xfe.t'}. Consider the following expectation value with primed and doubledprimed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'. Garrod [106].tf\xi.8).V(xk) = m0±l .530 CHAPTER 21. t to x. operator (21.ti) where nl D{x(£)}exp '. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. x. (21.8). £ /  X J .x) . .105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. I hi dtL(x. t. p .
21.5 l^ t 'lp / > e . x fdxG  ^ P <p"x3)<x3e^V) x (21. (21.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x"./ ^ I < « .*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.t'}.i p '• x . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.* * . (px).0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.t"\x\x'.t"\x5){x5\x'.107). With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. .76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). (21. .109) . Thus <xV'xx'.6 Path Integrals in Dirac's Notation For the verification we use Eq.* * f dp = j (2^ p) ^ d . With the wave functions (21. .v ' x ' and thus {x".77) and thus have 531 *!*') = *M = * 7 ^ I P > « .
x / / e . (21.tj) at equal time intervals e. tn2) • • • (xi.£j to Xf.110) Recalling Eq.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .111) 1 . ( x " .iit!' 1') 2mo + vtf) + <5(x"x'). we obtain with (21.t') dp .ti = to as before.f ) i f e i p . We now consider the transition of the expectation value of operator x from Xj. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK. = x"6{x" .0 dp x„ / (27T): 1 .i t " ( x "  #  x 5 ) ( x V ) +. t0). (21. (21.tf subdivided into n transitions (xj+i. Thus with x^ = xn. tixx 0 .112) where x" is a cnumber.• dx1 (x n .94) for the case with no operator in between.74) of the delta function. . This is the expression corresponding to that of Eq. .£j) = / ••• / dx n _idx n _2 .x') .t n _i) x ( x n _ i .532 CHAPTER 21. £ n _ixx n _ 2 .£/xxj. (21.i ( t " ..t'){x"\H\x') = X + •••} (21.tj+i\x\xj.x') . We now proceed from there along parallel lines.90) we can rewrite this expression as (x".i(t" .t') (xV'xx'.x ' ) (2^ i^+vM + D «p(x"x') (21.tf — tn and x$ = XQ.85) (xVxx'.*/<x"x5)(x5tfx'> + .£ n xx n _i.i{t" . Inserting here the integral representation (21.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .£'> = x"[«5(x" . we obtain (here and in the following the limit n —> oo at the end being understood) (x/. ] .t"\x\x'.113) Inserting for each of the infinitesimal intermediate transitions the amplitude .
.89) is consistent with the conjugate expression dL (2L116 * . Xn _i X„ xi dx.x n _ i ) ] .and hence may be put in front of the integrals to yield (xf.x2)(t2 ~ h) + • • • + H(p ••• + p „ .21. tj) = xf (x/. For simplicity we consider the case of one spatial dimension...t') (21..x . (21. tj  xj1 x». (21.ti) 533 rxf=x(tf) = / • • • / dXidx. tj \ xt. .t') sider the variation S(x".112). dyi l xi x 2 . we obtain* (xf.3 " n d x f c .( x „ .\Xi. U).t"\x'. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 . (21..105).114). the string of factors xi X2 . x (21. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.x 1 )(t 1 Z & ) +H(p2. x n _ i x n reduces to the single factor x^.2 .t') = (x" + 5x". We demonstrate first that the operator representation of P given by Eq.y (27F (27F •" (2^(2^)3 exp[. .115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/.t"\x'..tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.7 Canonical Quantization from Path Integrals (21.7 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.iyxxj..{^( P i. dxnn _i Xi X2 .114) x exp [i{pi • (xi . n—>oo *•»• 21.t"\x'.tf\x.t"\x'.« • > We reconsider therefore the transition amplitude (x". .t') = and con 5x"Vx„{x".t') .t"\x'.{x"..117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq.10) or (21.x 0 ) + p 2 • (x 2 .2 .
x). with N = N(t" .122a) = 17) 5x"Vx/. t". Then. we separate the metric factors from T){x} in Eq. and we demand the validity of the equation of motion. " (t). The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.t') =  6— / 1 V°{x}t .10) and set D{x} = D°{x}/JV(i/ .118) JI(x)/h .tt). For the purpose of enabling some formulations below.t"\x'. dx dt \ dx _ (21.l&cdi. (21.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). <9L' dt —dx + — 5 x ox ox f . (21. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. Hence w « = (t). dL r /•*" .118) — = (21 5x"(x".t'): 1 fx" 6(x".(x".t i_(dL_ x't' h\dx J t„ (21.t'). x) Jt Jt dL.t"\x'. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. jt/ 37 ( £T.i/(x)/fi ) j ^ ) dtL(x. Thus in this case of classical mechanics dL_ d fdV. (21. .534 CHAPTER 21. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.
x' (21. 535 (21.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization.tf\xi. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.1 (A)iV. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.125b) and the measure property (3) N~\A + A') = A^..125a) (2) the completeness relation holds.1 (A / ). (21.126) R.7 Canonical Quantization from Path Integrals or JL(x»y\x. ^2\x'.125c) Corresponding to our relation (21. F.21. / Jxi V°{x} / Jx' V°{x}.t'\ = l.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. Wightman [264]. Streater and A. (21.124) where with A = tf — ti.1>) = ^{x".Xf = x(tf).t').122b) By comparison with Eq. (21. 6 (21.t')(x'.x). § The evolution of the system is expressed by the Feynman functional integral (xf. S. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.t"\p(t")\x'. I(x) = / JA dtL(x.115) in three space dimensions. .
. (2) and (3) above.133) = ih{x". We have (cf.tf\T(Xl • • • xn)\xi. Then the relation follows with the help of the properties (1).127)) (x".t') = ih(x..t') + x"^J^{x".t>).ti) = x' 'Y^{xf..122b) we can deduce the nontrivial canonical commutation relation arising here.t').tf\x(ti)\xi. Eq.t"\x'.ti). (21.t"\X'.t').129) = ^ y With Eq. tn lie correspondingly in consecutive order between U and tf.t'){x\t'\x2\xi.t')=x"(x". (21.tf\xi.131) = ih(x".t"\x'. we have the relation (xf.x](x.tf\xi\x'.t"\x"\x'.t') (21.t\x'..ti) = Xi(xf. We can rewrite this expression as the commutator relation (ITU*" x"^](x".t').132) The generalization to higher spatial dimensions proceeds along parallel lines.127) We can now write down an expression for a timeordered product of operators.536 CHAPTER 21.128) where tf is later than ti and t\. (21. . (21. (21. (21.t\x'. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.t') \ i ox" i ox"J or in commutator form [P. (21. We consider the case n = 2. Thus (xf. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.ti) f ' V0{x}xlX2eu^lh.ti) = —!— p V0{x}xx • • • xneu^h.130) ~i5x7jX"{x".114) (xf.t"\x'.t"\x'.tf\T(xix2)\xi. (21. Thus if T denotes such ordering. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.t"\x'.
e. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. Watson [283]. The vast majority of perturbation expansions discussed e. integrated over) beyond its radius of convergence. nonzero radius of convergence in the domain of the coupling parameter. z = a + i/3 with a — oo or f3 —> oo or both. i. although the first few terms indicate a convergent behaviour. W = e x p ( l / z ) . Consider e. 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. Besides. It was therefore natural to search for alternative methods of expansion.g. i.g. divergent.e. If the expansion is mathematically well defined. p. 102.g. > 537 . Such expansions are strictly speaking. Whittaker and G. T. the series converges. An essential singularity is really an ambiguity rather than an infinity. in the context of quantum mechanics are not convergent.Chapter 22 Classical Field Configurations 22. N. in the sense of rigorous convergence tests. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. for sufficiently large r. Expansions of this type arise. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. as we have seen in Chapter 8. Such an expansion parameter is usually known to be or assumed to be small in some sense.e. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. In physical applications such cases are rare. such as the fine structure constant of quantum electrodynamics.g. see e. i. it has some finite.e. so that successive contributions to the first approximation do not invalidate this approximation. E. but asymptotic.
h2..) l/h2.. On the contrary.e.h°. This type of expansion is decsribed as "semiclassicar. higher order corrections are of orders ±.h. In searching for other means of expansion. One therefore faces the problem of quantizing a system which is subject to constraints. i. Many of these aspects are interesting by themselves . nature of the dominant approximation does not imply that this describes classical motion. 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). 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. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential.. The study of expansions of this type has turned out to be extremely fruitful.e.538 CHAPTER 22. the latter's topological properties if any. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). i. and the corresponding expansion is called the "loop expansion". and in general this change is accompanied by certain constraints.e. A method which achieves this is in particular Feynman's path integral procedure. such as soliton theory. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. In the following we consider various typical examples. cnumber. and has led to insights which previously seemed unimaginable.h3. such > a series begins with a contribution proportional to (e. 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. 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 classical. cnumber first approximations. A further challenging task was the development of methods of quantization of theories which incorporate classical. There.g. However. although this terminology is not precise. The fundamental extension of the method of canonical quantization (i. the consideration of conditions which insure the existence of Green's functions of this new type of expansion procedure (which again leads to asymptotic expansions) and so on. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative".. These methods are often described as methods of collective coordinates.
It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. vortices and kinks. Tong [272]. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking".22. Thus references to field theory will be exceptional and hopefully do not irritate the reader. or even a solution to a modified field equation (e. 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. it is important for a better understanding of the considerations which follow. to keep this case in mind. (22. 22. monopoles.2 The Constant Classical Field 539 and in any case deserve detailed study. a static (time independent but spacedependent) solution of the classical equations of motion. modified by going to imaginary time). and also to see these in a somewhat broader context (by comparison with higher dimensional cases). However. may like to consult the article of D.1) ' T h e reader who wants a highly advanced overview of instantons.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. d^] = d^d^ .g. This means theories with field densities which are therefore infinitedimensional. 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).2 The Constant Classical Field We consider first the case of a constant classical field in a scalar field model called the complex Higgs model or Goldstone theory.V(4>). 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>. V{</>) = m§0V + A0(<^)2. 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. We shall not be concerned so much with the case of a constant first approximation. .
For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. An even more familiar example is the GinzburgLandau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape). £)) is bounded from below we must have Ao > 0.e.2) We ask: Are there classical (i. 22. i.1.540 CHAPTER 22.3) See e. . (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. and hence here [d^ + m20}(l) = Xo(f (22. V[<t>] m 2>o o Fig. Felsager [91]. m l + \04>*4> )$ = <). To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. cnumber) solutions <j)(x) = 4>c = const. B. 433. Classical Field Configurations where d^ — (—d/d(ct).1 Different potentials for different signs of m§. § (22.g. as illustrated in Fig. p. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0.e.S£/<90(x. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. 22. of the equation of motion? For these we must have all derivatives of <j) zero.£). V ) . The EulerLagrange equation is d„ dC d(d.
not the solution x(t)). the only such solution is the case we wish to consider.2 The spontaneously chosen phase. 22. like the solution x(t) of the simple Newton equation mx(i) — —V'(x) violates the invariance of this equation under time translations t —> t + 5t (i.22.5) . the equation has this invariance. we have 541 0. The Hamiltonian density H is defined by the Legendre transform H[<f>. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig.2 The Constant Classical Field For Ao > 0. 22. Fig. mg < 0.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. But for Ao > 0. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. with 0 = 0. H = —C = V. (22. and so ft[<M=m6:(^* 1 + Ao(0*0)2. TTIQ > 0.e. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. 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>. w A iP 3ml 5* "XT" (22.ir]=7r<fiC. We can convince ourselves that <fic is the field associated with a state of minimized energy. TT = —J.e. aq> For (b = const. i.</>2)plane.
3) obtained above. 22. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 . This is the condition (22. We examine this in more detail by setting <f>(x) equal to the classical cnumber configuration plus a fluctuation field n(x) which is again complex like <f>(x).e.3ml = 2ml = ml + TA 0 A 2 = 0. and we wish to investigate the behaviour of the field cf> in its neighbourhood. .7a) (22. 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.Then m\ m\ = ml + A0A2 = m2. >° for m 0 < °> . V?) (22.e. i. the one for (3 = 0. (22. Suppose we choose one such configuration.2. i. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi.2) and separating real and imaginary parts. in a transverse direction. &H/d(fi = 0) if [m20 + ^>]</> = 0. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.g.e.7a) vanish on account of Eq.6) Inserting this into Eq. Classical Field Configurations the density Ti is minimized (i. along the trough of V (which we can call a longitudinal direction). For every value of the phase (3 we have a different constant cnumber field configuration 4>c. (22. and partly by climbing up the parabolic wall on either side.7b) In Eq.7b) the constant on the left and the coefficient of A on the right of Eq.542 CHAPTER 22. (22. i. we set cf>(x) = <f)c + ri(x). as indicated in Fig. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. (22.tp2.3). e.
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. tangential to the classical path. ^ = h eip / \ ^2 l . We can see the problem here by looking at Eq.3 The spontaneously chosen phase seen from above. of course. t') = d(r .7b). (22. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r.22. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. the existence of individual perturbation contributions).r')S(t .e.e. (3=0 Fig. t. which. 22. Goldstone's theorem applies to fully relativistic field theories. The elimination of the . has to be removed in order to allow a well defined perturbation procedure (i. irrespective of the question whether the perturbation series as such converges or not. i.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. Thus the Green's function is similar to that in electrodynamics. 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 £. then a massless boson exists. r'. There the wave function with an associated vanishing eigenvalue is called a "zero mode". 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. though not identical phenomenon. 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.t'). whereas ipi(x) is y ' " " " " " ^ . In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar.
instead we shall consider static and timevarying solutions in important models of one spatial dimension. Zimmerschied and H. t Potentials of this type have been discussed by M. Lohe [179].t In particular the sextic potential can be considered along parallel lines.£) in one spatial dimension. and we choose the metric ?7oo = + 1 . see e. we shall need only at a later stage. Soliani [237]. To insure that the energy (see below) is positive definite we require V[$] > 0. 22. Rebbi and G.544 CHAPTER 22. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. *S. Khare [17]. however. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. C. equates to zero the coefficient of this wouldbedivergence. 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. N.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories.* We consider the theory (later: quantum theory) of a real.* Potentials of higher polynomial order in $ than $ 4 can also be considered.d^} = ~d^d^V[^) = ^2^(^j V[$]. 7711 = — 1 ( 1 6 Minkowski manifold). Classical Field Configurations other components results from the vanishing mass and the constraint called the gaugefixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution. A.g.e. (22. . Behera and A. F. We make one more assumption concerning V which. The Lagrangian density is taken to be C[^. J. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. MullerKirsten [291]. Later we wish to develop a perturbation series about a classical configuration of $.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). W. i. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. spinless field $(x.
In a quantum theory the field $ is an operator defined on a space of states.9) with quartic potential is defined by ^V\g$\ = ^V\g*. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. (22.t) has an explicit timedependence it is a Heisenbergpicture operator.11) (22. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. The first such condition to be imposed is that the energy of the solution of interest must be finite. i. (22.e. — m$ 1 — cos Im 1 I ^ $ (22. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. (22.«{x) = V'(<p).g] = \ v m For instance. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. Since $(x. at least in the present case. (22. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. i.22. t) is an observable (which in general — in field theory as compared with quantum mechanics — does not have to be hermitian although we assume this in the present example).e. the socalled ^theory = ^V[g$. (22.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.l]. ^ ^ * + V'[*]=0. The other condition which we impose (and study in detail) is that of stability.12) we see that they satisfy the Newtonianlike equation <j.13a) We write such fields <j>(x). From Eq. the reason for this condition being that we visualize the classical solution as representative of a lump of energy.§ Unstable classical configurations are also important and will be considered in later chapters.12) where the prime denotes differentiation with respect to $ .l}. The field $(x. such that J > — — — = 0. . 1].13b) We are not interested in just any solution of this equation.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.
15) The zerozero component is equivalent to the Hamiltonian density H. i.4 The wellknown soliton of $4theory.546 CHAPTER 22.2\dx 2 ' 1 fd& + V[$]. dC • Too = r?oo£ + =r$ = C + vr$ = H.14) where the overdot implies differentiation with respect to time t. (22. HI tanh m(x>k) 9 Fig. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($. <9$ i. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. (22.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. . n = & . 22.T^ = 0.17) Since we have to integrate over all space.e.7r) where IT is the momentum conjugate to $ defined by dC ? r = . We observe here already that the integrand of the integral in Eq. d.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>.e.c = <£z + .T = <I>. the classical solution will therefore have to be such that this integral is finite. (22. (22.
22. Again XQ is an integration constant. (22. — XQ). . Rubinstein [240].11). In fact.5 The soliton of sineGordon theory. 4m .18) Here XQ is an integration constant. (22.13b) for the specific potentials (22.22. O±m(xxo) (22. See J.5. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n".777. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. the $*theory. < = — tan [e >  ° ] 9 2 rrm/g — Fig.1 . around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . > . one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2.3 Soliton Theories in One Spatial Dimension 547 As we shall see below. 22. i.20) (22. it is not difficult to find classical cnumber solutions 4>c to Eq.x n ). m ( x .e. "The name sineGordon is derived from the analogous KleinGordon equation or theory.4.18) is depicted in Fig.21) This sineGordon configuration is depicted in Fig. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . 22. in the case of the first. (22.10) and (22.
13b) which we can write d \W? = !"(•».9 2 4>{x) = i. ± 1[tanh.20) are obtained from Eq.* ( m' m 2 r)] ^2" .18) and (22. Classical Field Configurations The solutions (22. Inserting the potential (22.o) =m7r. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ.548 CHAPTER 22. and hence 4/71 9<t>{x. 4m In tan tan 5 l/e±m(*xo)tan^£^)U. <^>(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 . or \(<P'? = V(4>)+ const.1 "»JtKxo) 9 1 .e. . 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . With we obtain g(/>(a. Here the constant is zero for V(<p) and <j>' zero at x = ± o o . (22. In the case of the sineGordon theory defined by the potential (22.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ).
(22.17). The expressions (22. we obtain these expressions later (cf.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) . The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as. (22. However. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. however.23) See for instance R.21) for the energies can also be obtained from Eq. (22. SE((f>) 5cb(y) 0. the curves rising monotonically from negative to positive values are known as kinks. the discussion of Bogomol'nyi equations) in a simpler way. Then we have to demand that the functional derivative of the energy E(<p) be zero. 22. in the onedimensional Ising model. Jackiw [141]. 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. for instance. A state of stability is therefore associated with minimized action and/or energy. .19). the others as antikinks. this does not require the action to be minimized. Suppose now that we demand stability in the sense of minimized energy. i. The relation (22. (22.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". instead of doing this now.22) and that its second variational derivative be positive semidefinite.e.4 Stability of Classical Configurations The concept of stability can be complex.22.
i. The original Lagrangian was. a very general phenomenon which can be established as follows.26) Comparing this equation with Eq. 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. namely (p'c. Since it arises from the violation of translational invariance by 4>c.13b). if we apply the generator of spatial translations d/dx to the classical equation. (22. with eigenvalue w2.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. i. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). in fact. This is. (22. the function 4>c(x + a). in fact ip0(x) const. dx (22.25). written down in Lorentzinvariant form.28) . its eigenvalues w\ > 0.e.e. Eq.13b) retains the invariance although the solution (f>c does not. dx2 + V"{4>) (22. of course.e. and the EulerLagrange equation retains this property. Classical Field Configurations Since y ^ ±00. Thus the vanishing of the first variational derivative yields again Eq.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. (22. (22. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. In fact. i.13b) with classical solution 4>c. if 4>c{x) is a solution. (22.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite.550 CHAPTER 22.A. The Newtonlike equation (22. = 0. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. the integrated contribution vanishes.a / 0 is not the same. The eigenfunction 4>'c{x) is for this reason called a "zero mode". 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. in particular the invariance under spatial translations. it is also called a "translational mode".
33) x—>±oo . In the onedimensional case under discussion we can rewrite Eq.32) = V'(</>). / oo Thus for E(4>c) to be finite. As stated above. (22. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). in fact. We return to our considerations of stability. i. Thus TUC is a zero mode. We can choose the potential V such that the constant is zero. 0 = S'[UC] = S'[gUc}.27).18) and (22. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. = 0. this integral has to be finite.31) Inserting this into Eq. (22. lim [V(<f>c) + const. We see therefore that the occurrence of these zero modes is a very general phenomenon.17) we obtain oo dx[V(4>c) + const. and that S[UC] is finite (or equivalently the energy). i.20) are such configurations in two particular models.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. Then the vanishing of the first functional variation implies that 5S = 0. (22. (22. We shall see later that these zero modes lead to undefined Green's functions for the semiclassical perturbation expansion unless one or more suitable constraints are imposed.] = 0 or x—>±oo lim V{(/)c) = 0 for const. (22. we are interested in classical cnumber field configurations of finite energy. 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.].13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. (22. (22. it is an expression like (22. and we have seen that Eqs.22.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0.e.e.
» In the <fr4theory defined by Eq. 22. . Thus the classical energy or vacuum configuration is not unique. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory.1). V{4>) = 0 at different values of </>). ± 2 .6.6 The minima of V in $4theory at x — ±oo. Classical Field Configurations V Fig. 22. n = 0 .35) and it is seen that in this case q can be ± 1 or 0. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. . . (22.11) we must have correspondingly for x — ±oo: » 771 9 2vrn.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. the potentials have degenerate minima (i. oo J—oo y The topological quantum number q is defined by 0. In the case of the 3>4theory we can define a conserved current k^ by (22. This number is defined like a charge in field theory by the spatial integral of the time component of a current. The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22.552 CHAPTER 22.c o and another for x — oo.34) jfc" = e^dv(l since d^k^ = 0.e. in view of the antisymmetry of eM^. . (22. the classical field (bc must approach one minimum for x — . In the sineGordon theory > • defined by Eq. (22. For a configuration to have finite energy. . ± l .
A property is.22.11) are given by ( so that for x — ±oo: > — J. 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. . . and we call this "topological stability.cos(27rnA)].37) for the onedimensional case here.38) 4 V(\<t>c) >m ~2[l . In the case of the sineGordon theory the minima of the potential (22. .7T (22. one may ask how topology comes into the picture. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. in general. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. zero is not a meaningful topological number. We can see this as follows.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. Again we see that E(\<f>c) is finite only for A = ± 1 . The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An .7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). First. continuous values. ± l ± 2 . For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. Tfl \ (22. termed topological.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . . n = 0 . x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . and exp(ix(0))<pc reduces to 4>C when x(#) = 0. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy.A2)2 + 0 except for A2 = 1.
33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22.9 the foregoing arguments in a higher dimensional context. (22.31) to (22. (22.17)) f 00 > 0. Classical Field Configurations involves the difference An = rioo — n.g.43) **See e. the energy is minimized.13b) or ^ ' ( * ) ] 2 = V(</>). (22.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. 22.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion). From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.41) dxy/2V{$)4>'{x). 00 (22. We observe (a) the topological current k^ of Eq. its integral is nonzero only on account of nontrivial boundary conditions. 00 (22. 22. i.554 CHAPTER 22. 425 . (d) the topological quantum number iV arises in a completely classical context. (c) the time component k° is a spatial divergence. Thus with one space dimension higher there. if oo / dxy/2V{<l>)<f>'{x). Eqs. W.42) Thus the energy has a lower bound given by the right hand side of this inequality. Eq. . (22. H. MiillerKirsten [215].e.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (For a broader view we consider briefly in Sec. the factor A = exp[ix(#)] varies over a complete circle).^ = N. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. pp.428.40) (22. J. The inequality is saturated.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents).
We discuss this in the following Example. 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— .40)) this occurs when 4>'(x) T V2V(4>) = 0. (22. (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.21).39).22. (22.n > 1. Bogomol'nyi [27]. 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.^ Clearly it solves the second order equation or (22. Here (V(c/)) > 0. . B.19). the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. 555 (22. n E . Eq. for instance. In the sineGordon theory we can consider.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.44) This first order equation which saturates the inequality (22.5 Bogomol'nyi Equations and Bounds We observe that (cf. 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.
Chapter XI.e. Assume that the pendulums can move (i. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. ka —> K. . Fig. all being identical. rotate) only in the (y.t) measured e. 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.t)\2 +mgr{\ .556 CHAPTER 22.7 The pendulum analogy.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . Solution: Consider a string along the xaxis as described.cos</>(x„.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. } n= — oo roc —> / L J . H.46) **See e. We can now write down an expression for the energy of such a system of (say) n pendulums.g.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. Construct an expression for the energy of such a classical system and its Lagrangian. so that oo > u. rotate) only in the (y. 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.2 V dt + k[(f>(xn+i. Goldstein [114]. from the yaxis as shown in Fig. 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 ''( "' ) .g.7.e. Assuming that the pendulums can move (i.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji.t) <j>(xn. Initially the pendulums are suspended freely in the gravitational field. 22. xn+\ — xn = a) 00 a —> dx. Classical Field Configurations Example 22. t)). 22. z)plane and only such that their elastic strings remain taut.t) 4>{xn. with mass m and connected with short strings to neighbouring bobs. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.
50) This equation is identical with Eq.47) into (22. If some w\ were negative. of the continuous curve in Fig. (22.49) exp(iwkt)ipk(x). The equation is called stability equation or equation of small fluctuations.49) would imply an exponential growth in the future t > 0 or in the past t < 0. which represents an expansion in terms of normal modes. 22.7 linking the pendulum bobs). each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. wk would be imaginary and so the factor exp(—iwkt) in rj(x.47) where n(x.13b) we have for small fluctuations rj(x. Hence we set ${x. (22. ' Equating coefficients of exp(—iw k t).t): {w With the ansatz r)(x. Inserting (22.t). (22. Eq.t) = ° (22'48) (22. and so would invalidate the procedure which assumes that rj(x. We can understand what stability means in the present case.25) obtained earlier.e. t) of Eq.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. is therefore also called "winding numbed. This number. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). (22.13a) and (22.t) is a small . (22. the topological quantum number.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x).22. (22. It is instructive to obtain this equation by yet another method.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. 22. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.t) = <l>c(x) + r}(x. t) is the fluctuation (field).
20).1. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. i. The equation of small fluctuations (22.50) has the general form of a timeindependent onedimensional Schrodinger equation.e.558 CHAPTER 22. Inserting (j)c of (22. V"(<b) = 2m2 + 6g2(f>2. V"((/)c(x)). (22. In fact. 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£^)*. Thus for stability w\ must not be negative. In fact. If that is necessary. and need 9 = tan" 1 j e±m^Xo) 1.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . cos46» = 2cos2(2(9) . approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. For x — ±oo the quan> tity acting as "potential".A \ Inserting <f)c of Eq. . the zero mode.2 m 2 + 6m 2 tanh 2 m(x . This means that the stability is not universal but only local in a certain sense. Classical Field Configurations fluctuation.e.52) tan 6 := e±m^xo). we obtain V"(<t>c) . i. we shall see later that the fluctuations n(x.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. 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.18).
we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. A  *(* + !) cosh 2 z tp(z) = 0. .* For such a potential the spectrum can be b o t h discrete and continuous.53) = m(x — (22.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.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.e.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. even and odd solutions. ^+(0) d 0. ip+. Those of the other case are complex and periodic at infinity. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.ijj^ respectively. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.tan26> l + tan 2 i9 / 2 (cosh 2 m(x — XQ) 1) — cosh m(x — XQ) cosh m{x — XQ) 1cosh m{x — XQ) 2m 2 cosh m(x — XQ) Thus V'fa) = m2 xo)) d2 .22. (22. We can therefore construct solutions of definite parity. which satisfy the boundary conditions V>(o) = o. (22. i. We can investigate the spectrum as follows. Consider the equation fd2 dz2 . — XQ)] cosh[±m(x — XQ)]: 1 .55) The differential operator is even in z.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .2 (22.
e.( 2 n .e.60) terminate after a finite number of terms.Z)2. 2n < Z. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. A ^ An = .57) Then the equation for x{z) 1S (22. when in the even case n ( n .Z ) 2 .e.Z ) + ^(A + Z2) = 0. (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. and in the odd case n+\)U+\l)+\(\ i.61) Discrete eigenvalues are obtained when the series (22. i.4 n ( n . i.l2 = .Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. + l2) = 0.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.( 2 n + l .560 CHAPTER 22.I) . i.58) = (1 + sinh2 z)~l/2x(z). A * An = . + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . . (22.59) Setting (22.
3m 2 .N<1. (22.. (a) ^theory: the equation Inserting (22. i.22.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0. °2 i. (22.6 The Small Fluctuation Equation 561 and the function ip.( 2 .\=Kl2 From (22. for N = 0.65a) _dz2 cosh 2 z l=2..10) and (22.j < I . d w = tanhz.u2)—^V 2 dio and ip(z) = elkzV{to). (22.62) where even (odd) iV are associated with even (odd) eigenfunctions.63) we see that the continuum starts at A= 2 2^ = 0.50) we obtain with z = m(x — xo) tp(z) = 0.2.e. m i. m N<2. (22.54) by setting X = k2 Then 1 . (22. w2 = Am2.1: w2 = 0.1. the continuum is w > Am2.is normalizable provided 2(n + .. N = 0. We can see the continuous spectrum of Eq. From (22. 2 .Ny.2Z = .e.62) we see that the equation possesses discrete eigenvalues w —2 .51) into (22. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .11).64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz..e. We can now derive the spectra of the stability equation for the potentials (22.
.65b) dz2 cosh^J. 22. and — as expected — this eigenfunction is even. %(x) Fig. N<1.27)) .66) We see that this is a nonvanishing function for any finite value of x. m d . .53) into (22. also be deduced from d(j)c/dx. of course. Thus 4>'c(x) has no node. for N = 0: to2 = 0 and the continuum starts at w2 = m2.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. Eq. 22. d . m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. The zero mode is depicted in Fig. Thus either spectrum contains the expected eigenvalue zero. .8 The zero mode as typical ground state.562 CHAPTER 22. In the $ 4 theory we saw that (cf.8 and we see it has the shape of a typical ground state wave function.. . so that its derivative is nowhere zero (except at ±oo).. Classical Field Configurations (b) SineGordon theory: Inserting (22.50) we obtain with z = m(x — XQ) the equation (22. .e. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. We know that 4>c is a monotonic function.=1A=^_. i. It is now particularly interesting to look at their wave functions. This property can. (22. the zero modes.
Finally we consider the normalization of the zero modes.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. . Similarly in the sine—Gordon theory we write so t h a t 4 /oo * " / > & • ) W171 dx oo cosh m(x — XQ) TYl = 4 — [ t a n h m O r . (22.z 0 )]oo = 8 . 1 o rrr tanh x — . Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.3 .22.6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to .70) . . d .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).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. — XQ) OC Hence the conclusion is similar to that in the previous case.= E(<f>c). .t a n h x . without quantum corrections). (22. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. i.e. . (22.oc (22.e.
to underline the fundamentally different nature of topological and nontopological finite energy configurations. ddx {V4>f + v{4>) (22. however. for if such configurations exist in a realistic theory with three spatial dimensions.T(4>). This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0).41) for the energy of the classical configuration but now with respect to a ddimensional position space. P. Derrick [68]. This question has important consequences.72) with T{$) = I' ddx]{V(j))2 > 0.2d. .g. One can have different types of scale transformations — see e.g.564 CHAPTER 22.> 4>a{x) = 4>(ax). *The considerations of this section extend in part beyond quantum mechanics and into field theory.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . J As a first example we consider the expression (22.e.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. A. ^For basic aspects of scale transformations see e. H. They may help.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . I.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. Goddard and D. I. *See e. (2. i. (22. P(</>) = f ddxV{(f)) > 0. f G .4) <j>(x) —• a<f>(ax).g. H. Classical Field Configurations 22. Kastrup [145]. Olive [113]. we consider E(<P) which we also write as (22. Affleck [5] who uses in his Eq. it is important to understand their physical implications. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here.
e.e.74) From (22. (22. a=l (22..79) In order to emphasize that (f> — (p^ in any direction. Consider first the Lagrangian density £[&M = \{d^)(d^) . (22. we set 0 dE(<pa) da J a = i (22. we see that the stationarization of E with respect to a is possible only for d = 1 as in the examples we have been considering so far. to be finite the field <P ~^ 4>oo £ Mo for x — oo. i.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}.xd). the second derivative is positive) only for d = 1. In fact.e..d)T(<f>) = dP{4>).75) Since both T and P are > 0 and d is integral. x = (x1.\ eMo(22.x2.73) we see that this implies (2 .V(<P) (22. we write • = lim J r .71).7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1. (22. d = 1 minimizes the energy.78) For the energy of a static configuration. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. since d2E(<pa) da2 2(2d)T(<f>).. 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}..80) .22. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).76) we see that E(<f>) can be minimized (i. (22. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. Next we consider more complicated cases. i.
These observations are summarized in Table 22. — 1). —1. of • course. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. i. are the same.e.e. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. . the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00. in fact. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . that we have a set of disconnected classical vacua (</>oo><^oo. i. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. We see. i. 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 .81) and (since d = 1) S = ± 1 . therefore. is a connected region. and. x = 00.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. one set can be mapped into the other.566 CHAPTER 22. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively).1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . as is the case for the constant solutions. Then each point at infinity is characterized by a different value of the polar angle 9.and sineGordon theories we assumed $ to be real.1. If we go to two spatial dimensions (d = 2). Table 22. —00 or equivalently S = + 1 . (ptx = <f>oo. ( with — = lim <pc(x) 9J *>±oo TYl \ (22.38). S1.e. Thus in this case S^1 is a disconnected set consisting of two points. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. —00.
that if the problem is to be topologically nontrivial. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. (22. assuming invariance of C under the transformations of the group SO(3) in internal space. the one point S = + 1 . if cj> were real and position space twodimensional. This will be "nontrivial" in the above sense only if 0oo is independent of 9. i. i. If position space were onedimensional. . 00. —00 or S = 1 .82) Now Mo := \ 4>(oc) ^24>2 = a2\. therefore. cj)(r. Thus. On the other hand. to have topologically distinct classical configurations. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.3 (^real) v{4>) = / (22. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. Similarly. VMx)) with (0(oo) = linvxx) (p(r.g.e.• M0. This can be achieved e. we would have only two points at infinity. the field must map position space into field space at infinity. We see. for instance. i. * n reach any other <^(oo) by simple phase shifts. i. $(r) : Sl . i.e. phase transformations 6Q ^> 9Q + 59.83) .2. map into. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. by allowing <fi to have three components in some "internal" (isospin) space.1 . if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}.e.22. 9) would depend only on r = x and hence the problem would be effectively onedimensional.e. we have w e ca {Mx)}> and 1 3 1.e. if <p is real.e.7 Existence of Finite Energy Solutions Suppose. i. Thus.
so that details of these models are irrelevant. Olesen [220]. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. a = 1. . going once around S%~1 (i.e. through each point at infinity). Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). If. circle.e. Example 22. % . $ — $exp(ia) in a U(l) symmetric theory. the appropriate rotation. the subsequent discussion requires only the general form of the Lagrangians. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. Georgi and S. B. A) = T(4>. the winding number is n. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit.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) £(</>. it is effectively a surface (and so a continuous set) S2. we cover <S^_1 n times.e.</>] = ~G^Gai. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved.a M J . $ — $ + 27 > • T in sineGordon theory. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. Glashow [108]. i.V{4>a].Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. i.568 CHAPTER 22. Nielsen and P. + i ( D " f t W ) a . invariance under replacements $ — —$ in $ 4 theory. L. The symmetry of the potential (and Lagrangian) — i.e. 11 H. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. $ — > > $ exp{i Xlo=i Pa.. 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. sphere.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3.„ where G r = 9 M £ .. A) + T{A) + P{<j>).2. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo.3. .
= 0. (22.e.dP(4>) = 0.d T ( A ) + AdP(</>). Thus we set dE(<fix.<^[x) transforms like a vector. J 4 ) + A 4 .d)T(A) .Ax) dX Since T((f>.i T(AX) = X4~dT(A). (i)^ 2d T Considering the contributions of (22. <0 >• <£A(X) = <KAx). P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.84) is positive semidefinite and so must be finite by itself. negative.A).A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. i. A A ) = A2dT(^.22.87a) C?M„(x) — \2G^{\x). Solution: Each contribution in (22.85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>. and so B ( ^ A . (22. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.d T ( ^ . A. T(A) = f ddxV(4>) > 0.T{A).d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). (22. (22.89) P{4>) are positive semidefinite and not all zero. > Since T>IJ. (22.7 Existence of Finite Energy Solutions where T{<j>..e. some contribution must be .86) (22.84) and investigate the existence of finite energy classical field configurations.d)T(4>. l j ddxGijGai] > 0. We now consider scale transformations with A £ R. Mx') = E with a scale transformation x — x' = Ax. we have {^/»)A^ (22. A). >• A„ x ( x ) = AAM(Ax). ^ A ) = A 2 . 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd .87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d.84) in this way one finds that the quantities involved transform T ( ^ . (2 . A) + (4 . i.
We then have a situation of the potential like that described in the context **T.570 CHAPTER 22. a = a(T) = a T —T c . (e) For d > 5: In this case it is not possible to stabilize Eq. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. This case is. In the following we consider a related theory. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. H. 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. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). From this development one knows that in the superconducting state of the metal the electrons combine to form pairs called "Cooper pairs'" which then have bosonic properties. (22. MiillerKirsten [268]. W. A) or P{4>) for d = 3. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. of course. Landau [111]. (22.t) such that  ^  2 describes the density of Cooper pairs. R. D. a > 0. Tchrakian and H.89).89) the contribution (4 — d)T(4>). if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. (22. One can thus define a scalar field or wave function ty(x.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory.90) and we would have to add to Eq. ft V . H.g.e. L.92) The expression ^ ( x ) ! is known as the "order parameter. J. is temperature dependent and changes sign at the critical temperature Tc. Skyrme [253]. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a socalled pure gauge theory with socalled instanton solution (which we shall encounter in a simpler context later). 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. Thus such a contribution could counterbalance T{4>. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. T^{<t>x) = \4dT(</>). .89). (22. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. Ginzburg and L. exemplified by soliton and sine—Gordon theories.** e. 22.
Example 22.2) £(^. Sec. Clearly the length (see Eq. assuming there is no superconducting material in this domain).e. 22.e.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ .(tp) = 0. Variation as before. setting S£.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. (22. which we recognize as one half of the kink solution we obtained previously as the topological instanton solution of the Schrodinger equation with double well potential (in the present case ip(x) = 0 for x < 0.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf.^ 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. from x < 0) means — as is familiar from electrodynamics — replacing V by V .94) to check the dimension) 1 = I—a. is a measure of the distance from the surface of the superconductor to where . i. but for T < Tc it has a double well shape. (22. leads in a onedimensional case (considering this briefly) to the equation ^  = aV + /?^ 3 axz Imposing the boundary conditions (in the domain 0 < x < 00) ^(0) = 0. Then the static energy density becomes (cf. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. ^(00) = a (22.93) J' (22. The parameter 7 serves to put the minima of the double well at V = 0.22. i.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4.2): For T > Tc the overall potential has a single well (minimum).
9 Behaviour of ip and B in the superconductor. 450.M.98) where M is the magnetization resulting from atomic currents. The magnetic field on the other hand. MiillerKirsten [215]. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ). 22. the order parameter \ip\ attains (approximately) its asymptotic value. We can see this as follows. V x M = j s . J. as indicated in Fig. H. (22. 2mi Setting < / > \P\e vp B = ViA. (22.g. i. (22. yJ—a/[3. W. In the Maxwell equation See e. p. i.e. 22.e. Classical Field Configurations superconductor (x>0) Fig. The density of the Cooper or super current is given by the typical expression of a current.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B .e.572 no superconductor CHAPTER 22. i. exhibits a completely different behaviour — in fact.9.96) ipH. . h [^DV'V'DV].
V x B = [i0V x M = /lois. and V x (V x B) = / J 0 V x j s . (22. this becomes A B = no V x j 5 . This is zero in the present case. 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. K i ™*. Also dD/dt = 0 in a static situation like the one here..8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside.L B = O. In the onedimensional case we have B{x) = B(0)ex/XL. Approximating p by its equilibrium value.e. (22. i. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 .97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. those of the socalled type I and those of type II.5 to 10~6 cm).div grad" and V • B = 0. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter.e.M i. These vortices therefore carry magnetic flux which is necessarily quantized . Hence 0 = V x H = V x ( — B . 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.99) Inserting (22. Using the relation "curl curl = grad div .22. m xi A B . There are two types of superconductors.
In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).102) In the onedimensional soliton case of Eqs.101) (p(r.e.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. de Vega and F.. We can think of (e. In the two models of Example 22.g) twodimensional space as being bounded by a circle at r = oo. we must have that lim V ( H ) = 0.37). (22. Classical Field Configurations (see below). The existence of these vortices with quantized magnetic flux was predicted by Abrikosov* and are therefore frequently referred to as Abrikosov vortices (the idea was originally rejected by Landau).36).9) ^e^ '>\<poo\. Abrikosov [2]. .574 CHAPTER 22. i. A. i.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 < 1/4.9). Thus the asymptotic field </>(r. *A. The existence of these vortices has been confirmed experimentally. Finiteness of E requires \<t>\ = \<fioo\ on this circle. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. the space consists of two diametrically opposite points on the circle. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. A. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. in other words. Schaposnik [69] after their equation (3. In the NielsenOlesen model defined in Example 22. See also P. in fact a periodic function of the polar angle with period 27r. G. (22. 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. 'See the remark in H. 22. of two points on a line. where exp(ix(0)) is an arbitrary phase factor.e. J. De Gennes [67]. (22.
Suppose <j>. We may define this as « = ^[X(0 = 2 T T ) .e.118)). abelian (i. Then a continuous function of (j> on [0.* 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. Finkelstein and C. Sl := {zeC\\z\ = 1}.1. The map A : S1 > S1 defined on S1 C R2 by (22.10.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. 0 < 0 < 2TT.102) is called the exponential map of S1 onto S1. topological group with the usual multiplication as group operation. commutative). Fig. (22.1). States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. X is called the domain of / and y its range. W. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. (22. In mathematical language S1 is therefore a compact. 2n] remains continuous if the function assumes the same value at the two endpoints. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure.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". i. The 1sphere is the unit circle in the space of all complex numbers. 22.X ( 0 = O)].22.e. Naturally an integer cannot change continuously. defines the orientation of this line element.10 States of winding number 0. Misner [96]. Thus smooth deformations *See D. . 22. The two theories we considered above in Example 22.
r^oo 1 dx(9) n r o0 (22..g. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example).e. finite. the winding number must be a constant of the motion. The winding numbers n are such topological invariants. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. Thus two homeomorphic spaces have equivalent topological structures. The gauge field required in the asymptotic limit (22. We observe that since time evolution is continuous.) remain the same under homeomorphic transformations and are therefore called topological invariants.e. elements in the same class have the same topological characteristic. if it has one twist as above or a hole.e.576 CHAPTER 22.. connectivity. no hole.105) is described as "pure gauge" since it is determined entirely by the phase x(#). i.104) to be finite we must demand that g . we can write 1 d h J ee rde w At h r^oo . i.105) the corrections falling off faster than 1/r. we required that lim r—>oo w o. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. i. (22. The different kinds of connectivity of these topological spaces (e. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. For that reason n is called a "topological invariant?. one hole. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. In other .i'ldx(0) 9A Thus for (22..104) so that d2x hJ Since A has direction e#.g. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.
the higher order contributions in (22.106) J g j r d d g Thus the magnetic flux is quantized. if A = V x . (22.105) insure that B ^ O . 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.22. the group space of (the phase transformations of) U(l) continuously. the gauge field. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.9):=t(f>£\9) + (lt)(f)W(9) with te[0. we have B = V x A = curlgradx = 0.e. cannot be deformed continuously into one with two loops (n = 2). It follows that the field Fij. (22.110) . going once around S^.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. A field configuration with one loop (or twist). it is determined by a gradient (l/r)d/dd. with n = 1. we can go n times around S^. into S i . In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. cannot be pure gauge everywhere.<b rd9—4^ = n. (22. we can consider transformations which describe continuous transformations of a configuration with fixed n. 9){^ and hence the functions (22. the vector potential A. Of course.e. An example is the mapping H(t. for B / 0 . Therefore.9 Introduction to Homotopy Classes words. However. Ffiu.e.l]. i. is also zero at r = oo. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). i.107) The fields <f>(r. i.109) A W(0) = e^ W. However.108) map the spatial circle S^.
and the space of t. (b) symmetry and (c) transitivity.578 CHAPTER 22. Thomas [270].112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. Example 22. H(1. 8) = cfinl\9). we have to show that the map satisfies the three properties of (a) identity. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).9) = <fiW(9) and H(l. Reuter [77]. But this type of combination is excluded sind H(t.T. (b) symmetry and (c) transitivity. e. Two curves which can be related to each other in this way. 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.e. i. it is clear that this is also a homotopy. H.g. S%. The map H(t.lQSl^S^ with H(0. The relation (22. A superposition of classical configurations with different winding numbers does not minimize the energy.e.9) = ^°\0). Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G.8) itself must be a configuration with definite winding number. i. (fin (9) in the above example. In fact. We can consider (22.1]. Classical Field Configurations which is continuous in the parameter t and is such that H(0. n' ^ n. Steenrod [261] and S. the unit interval I = [0.0) = $\6). A brief introduction can be found in the very readable book of W. — S\ > (22. (22.113) . are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).111) In principle an expression like (22.110) could be written down for configurations with different winding numbers. Dittrich and M. i. i. to the internal group space of £7(1).e.e. f T h e standard texts on homotopy theory are N. time cannot make a field jump from one homotopy class (see below) to another. (22. these two maps (fin from S^. i. Since the passage of time is continuous.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). 9) is an equivalence relation.111) as a map from the space described by 9. We verify these in the following example. H.9) is called a homotopy (emphasis on the second syllable). the phase factors of (f)(9). like <pn (0).3: Homotopy an equivalence relation Verify that the homotopy map H(t. n. Hu [134]. Solution: As stated.e.
which therefore demonstrates that the homotopy relationship is transitive. (b) Next we check that if 0<°) ~ <p^. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. 9) := H(l . below). Then H(t. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). Moreover.e) tf2(2il. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). ' . is isomorphic to) the set of integers Z (positive.e. (c) We can demonstrate the property of transitivity as follows. i<t<l. 9) = H(l.22. On the other hand H(0. .9) H(0. .t. 0) = ^ (0) and 6(1. at t = 1/2 both i?i and H 2 give c/^1'.* 22. as they are called. Hence the set of all such maps breaks up into equivalence classes or homotopy classes.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. 9) = H(0. U(l)).0) = </>(0) is continuous there. 0) = tf2(l.10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the onedimensional sphere S1. 0) = <£{2). Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. <^(2). 9) connects <pn (0) with itself continuously. 0).9) = . 9) = <f>W (0).e. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. so that H(t. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. Hence <f>(0^ ~ <p 2 ). i? is a homotopy connecting </>(°) with <^>(2'. 0). Returning to the maps \(x) : Si ^ S1^ (i. We now define the function H ( f 0 ff . a class being a set of equivalent maps. 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. then </>W ~ <t>{0)• We define an H(t. 0) = 4>W (0). negative or zero).0) for for 0<t<. Homotopy classes are sometimes also called Chern—Pontryagin classes.0). <f>W . is such that H(0. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. we see that of the set of these maps some are equivalent and others are not. these groups are then called homotopy groups (cf. since and tf (1. i. being the continuous map connecting </>(°) with <j>(1'>.9) = <f>(0\6) and H(l.ffi(O.e. \ M)\ H!(2t.
Fig.e. T (22. and then introduce some of the related mathematics.) why 7i"i[S2] = 0. but TTI[S2] = 0: S 1 is a circle. 52. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9.115) Varying t € [0.g. . This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. S 2 the surface of a sphere in a threedimensional Euclidean space.11 The trivial n = 0 map allowing shrinkage to a point.580 CHAPTER 22. In order to really appreciate how topology comes into the picture here one should understand (e. In the diagrams^ of Fig. Sketch of idea demonstrating that nilS1] = Z.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 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. T < 9 < 2TT. 22. n=0: t=1: t=1/2: i. Appropriate maps. Rajaraman [235]. 22.11 the circle drawn with the continuous line is the circle S%. 22. to make the result plausible. which is to be mapped into another circle Si indicated by dashed lines. How can we see this? We first sketch the idea. p. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors.
I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. to the case of winding number zero. We make another observation at this point.22. (22. The maps for these cases are illustrated in Fig. It is seen t h a t in these cases the continuous curve can be distorted into t h a t of case (a) only by cutting the dashed circle — thus these cases do not allow a continuous shrinkage to a point. i.12. One can prove that the expression (22.117) 2^ d6X(x)\1(x). In the cases n = 1.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). 2 the maps are wound around the dashed circle.s. the transformations^ (the result may be looked at as 8n of expression (22. 0 < X(P) < 2vr. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n.h. and hence the winding number. . is invariant under infinitesimal (continuous) deformations. i.e.h. (22.117).116) In fact from this we obtain (see below) 2  n and this means r. For winding number n we have the mapping (i. 22. Coleman [55].117) in ±. d (22.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.s.e.e. of (22.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. 22.
there is a theorem which says thatH TTg(Sn) = 0 for q < n. In fact.e. 1 1 See N. only one homotopy class. 15. whereas 7ri(50(3)) = Z2 (i. Sec. Steenrod [261]. 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.582 CHAPTER 22. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. the identity class (the group space S2 is simply connected. .e. Thus there is only the trivial map into a point. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W).e.8. i. an integer modulo 2) i. which means that a curve connecting any two points of S2 can be continuously deformed into every other curve connecting the two points) and one writes 7T1(S2)=0. Classical Field Configurations Under this transformation n changes by 5n Now. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0. the cyclic group with two elements (since 50(3) is doubly connected).
We consider such a theory now with action S((p) depending on the real scalar field 583 . different from literature. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.Chapter 23 P a t h Integrals and Instantons 23. Since we have not yet introduced periodic (i. and in particular the exponentially small level splittings of the double well potential. In the following we employ the path integral in order to derive the same result for the ground state splitting (only that in this chapter!). An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential. in such a way that the factors appearing in the result are exhibited in a transparent way. finite Euclidean time) classical configurations (which are nontopological). our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. 23.e.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem.
* /•tf CHAPTER 23. Fig. in particular E.2.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O .2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. (23. The Hamiltonian of the simple harmonic oscillator (cf. Path Integrals and Instantons S(<f>) / dt \i>2 .m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. Gildener and A.e. Patrascioiu [110] and Chapter 18. 23.1. "For comparison with literature. ! " ^ j = ^ 2 . (23. note that here the mass of the equivalent classical particle is taken to be mo = 1. 2 2 4 ™ = ~. and ^2(0) : m* .1)) and hence E0 = mh/y/2.v{4>) <?VV m4 __ 1 .584 <)>{t). 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. Chapter 6) 1 2 . n = 0. i. 23.2) Fig.1 The double well'potential.
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. e.2 — and we can deduce from this the boundary conditions.23. 23. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . but there is tunneling between the two minima if the central hump is not infinitely high. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>. </. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig.x) = x2 V(x).2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. This tunneling affects the eigenvalues. We can develop a perturbation theory around either of the minima.e. The states of the system here must be even or odd under x — ±x. this alone does not yield the level splitting. must have definite parity as in a physical > situation._(0) = 0. Fig.In mechanics of a particle of mass 1 with ~ position coordinate x(t).2 The wave functions of the lowest states. as we did in Chapter 18 — but as we saw.g. 23. i. and in the above 4> corresponds to x. W(0) = 0.
Euclidean time T = it. For small h we can attempt a stationary phase method.2 V'(<l>) = . (23. In the semiclassical path integral procedure that we want to use here.0o.e.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo. classically forbidden domains then become "classically" accessible.e.586 CHAPTER 23. i. r .2. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23. But quantum mechanically it can tunnel through the hump from one well to the other. Equation (23. i.e.U) = J'D{<f>}exp[iS(<l>)/h] (23.9) . (23.4) J <bi =—(An rTf (ITLE = I JT.3). (23.8) V(4>) + const. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.3) for tf — oo.2 m 2 ( l m z (23.5) and iS = dLE d dr[d(d(f)/dT)} i.<h. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.£'(0) = 0.7) so that (23.e. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. dLE = 0. The transition amplitude for this.7. that given by Eq.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21). (23.6) d*l dr 2 .  . SE. dr We observe that Eq. is finite if we go to imaginary.tf\ . i. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. Then (0o.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time. dr m+™ (23.
3. 23. this means the total energy of the particle in Euclidean motion is zero. — (23. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest.23. Fig.3 The instanton path. The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description. Its trajectory. in the present case (T — To) = m tanh m(r — TO). from fa — —rn/g. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p).e. 23. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). the classical path. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".e.10) where TO is an integration constant.e. Thus we know the solution from there which is the kink solution (22.18). timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. . i. we see that the equation is the same as the classical equation for a static soliton (i. i. is depicted in Fig. and hence kinetic energy gained is equal to potential energy lost.2 Instantons and AntiInstantons 587 In fact.
' .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.]. We also observe that the expression is singular for g2 —> 0.13).Tj) oc exp 4 m3 m^2 (23. thus indicating the nonperturbative nature of the expression. The configuration (23.588 CHAPTER 23.5). Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = . 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.13). (23.4.7"/! ).r 0 ) ' (23.11) For this quantity to be finite for TJ = —oo. The classical equation (23. but with the same Euclidean action (23. This antiinstanton configuration rises monotonically from right to left as shown in Fig. (23.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .TJ T fro.13) into Eq.7) admits another configuration which is the negative of (23.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. (23. the constant must be zero. we obtain lim f+ OO. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26. Inserting the result (23. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.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 (23.14) But this is not yet enough.10) is given by (23. Tf = oo.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23. 23. Using Eq.4).8) we obtain T f SE dr[2V((j)c) + const.^ . Path Integrals and Instantons The Euclidean action of the instanton solution (23.
15b) since we require configurations from vacuum to vacuum. 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. / > Looking at the instanton (23. 23. Questions which arise at this point are: Is the nonmonotonic configuration (23. and an antiinstanton localized at (say) TQ. 0c(r) = &(T). . (23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g. from the sketch or from the variation of SE.4 The antiinstanton.15b) has the shape shown in Fig. 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. #(r) = </>C(T .15b) In the case of (23. 23.TO far apart and T'Q 3> TQ.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . 23. b u t if TQ.TO) + ^ c ( r .2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). Clearly we are here interested in configurations of type (23. e. as we can see e.15b) is not exactly a solution.5. The sum (23. in the case of (23. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.15a) we have a simple superposition.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.r0).23.15b) a solution of the classical equation? Is it a topological configuration — if so.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.g.g. (23.5.
e x p { . Finally.15a) is a configuration which starts from 4> = —i^/g and ends there. J —oo the classical equation is valid only up to a contribution dL E d(d<t>/dT) Ti oc <ty(Ti) coshr m(T~i .5 Superposition of an instanton and a widely separated antiinstanton. Ignoring exponentially small contributions the Euclidean action of the configuration (23. since (23. g cosh m(r — TQ) (23. it also does not exactly stationarize the Euclidean action. vanishes as T\ —> ±oo (of course.15a) is not exactly a classical solution.T 0 ) } ] . m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig.16) . T\] with LECIT = 0.15a) can be argued to be zero. at oo the variation is <<> = 0). Path Integrals and Instantons '. 5^ In order to answer the second question we observe that (23. #(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 . 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. 23. In fact.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. of course.2 m ( T o .CHAPTER 23. we see that in the domain [—oo.
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).
. the kernel K(r. the value of the determinant is half (or arithmetic mean value) along the diagonal. .T2) I exp I K(n. H. In the limit S —* 0 . so that (with T *T) 1/2 •Dz N(T) (23.68) D above.n) o K(TJ. ^This is the method explained by B. we obtain exactly the nonvanishing terms on the right of Eq... as in Eq.63a). whereas the determinant arises only once in the original determinant Dn. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. Felsager [91].a { l n i V ( r ) .64).64) Inserting the explicit expression for the kernel given in Eq. Thus D„ = l + 5 ^ K ( r i . (23.n) z=l x (23.T1)dT1 + J dn J dr2. T. 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. Thus in our case one of the kernels is zero and we have to decide what we do when.T')dT' (23.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.7 Y. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) .j.n) K(Tj.. 2! .TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. Cameron and W.64) D (23. (23. r i ) + . Martin [43]. With this specific discretization the Jacobi matrix becomes diagonal and hence we obtain with the determinant coming exclusively from the diagonal* T>z ~Vri and then Vz Dr) = lim exp In TT ( 1 + K(T.Ti) i=l : exp \ £ K(T. p.n) i=l N lim exp £*(T. X(TI.64) as above and consider the determinant of the expression.exp • /_:T JV(T')" dr' = e x p [ . 192. where a f 1 See R. are summed over all values from 1 to n. [44]. Our final step is to relate the VolterraFredholm determinant to the required functional determinant.e. (23.4 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.63b). 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.. Thus the factor a inserted above has to be taken as 1/2.lnJV(T)}] N(T) N(T) (23.N{T) Finally we add the following remarks concerning the discretization (23. We observe that if we discretize Eq.TI) 0 K(T2.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: . r ' ) is to be evaluated at r = T ' . i.TI) K(T2.67) i=l v ) = lim exp V In ( 1 + ' L K(T. n )d? (23.23. (23. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K).66) •Dr) . = 1 52 K(n.
59) the endpoint conditions »)(±T) = 0 and evaluate the integral. In = dzi dz2 • • • dzn e x p . represents a constant translation. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T).73) .— .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. (23. .Z i . (23.(T) + /_ dr — . (23. Briefly.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). Path Integrals and Instantons Example 23. and hence may be ignored. N(T') . Solution: From Eq.59)) + f(T)] = — f dz(T) f°° dae^W+f™. f(T) = N{T) J^ dr'^^z(r') (23. T + N(T) / T' = T JT dr' N(T') .70) and this now requires also integration over the endpoint coordinate z(T). n J=. rn+i = T. In every integration / dzi the contribution €7. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo.^ .l + £7t) «i (23.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this. Evaluating Ii.71) Hence (with h = 1) l = j .3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. With partial integration (in the second step) we obtain (since (cf. Eq. Eq.22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. (21.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. (23.Z ( T ).0.( .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 [ .608 CHAPTER 23. (23.2o) 2 rc^O.e—*0 In.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23. I2 and using induction one obtains (cf.
^E.^ 23. Patrascioiu [110]. (23.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE.e. 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.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.75) or more generally a threedimensional surface integration j dau by the volume integration f cftx through the relation dF(x) 6[F(x) — otTo]d x. This result verifies Eq. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const.59) and agrees with that in the literature.10) and (21. 11 L.12). i. (21. (3. a = const.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. / d£o <9r0 5[F(TQ. £n = ~2 dTT](T)7]n(l (23. J co § dr Recall that / f ^ doce^ = yfVfp. Gildener and A. Popov [89].£Q) — aTo]dTod£o." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function.4.. .11). Eq. N. j / dTT]n(T)r]m(l r .23. (23.74) where we reinserted h. / — / / We started off with V(T) = ^CnVniT). Faddeev and V. 1/2 T W 2 (r)J (23. D. Ignoring the remaining metric factor with the reasoning explained > between Eqs.
if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . Then cn = ~2 / dr(pc(T)rjn(T).80) i r or •i /*oo dr7?o(r)0(r) = 0.75) requires. (23. this constraint implies that we go to the subspace which is orthogonal to the zero mode.^O) with dependence on ro which the replacement (23. The delta function 8[F(TQ. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).81) the rdependence is integrated out and hence the integral does not provide the function F(TO.79) (23. i.83) **The collective coordinate in the present context is also called "instanton time". / dTm{r)n{T).81) 2 / dr»7o(r)0c(r) = . i /*oo (23. We consider the following integral / ( r 0 ) := .82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T).^n{r). In analogy to (23. We now impose (i. Path Integrals and Instantons _^ ^ w . (23.e.78) <t>ij) = Y. .77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. (23. demand validity of) the constraint £o = 0. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. The number £o is the component of the fluctuation rj(r). (23. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r).J drm(r)Mr + T0) (23.e. In the integral of Eq. Thus. (23.610 this means that CHAPTER 23.£Q) — aro] contained in Eq. We establish the function ^(To>£o) with the following reasoning.g.76) we set 4>c{r) = ^CnVnir). (23.£o)/a. in the direction of the zero mode. where L = in + cn.75) enforces the collective coordinate ro to be given by F(ro. i. e. at r = ro.e.
Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) . jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.84) (23.y/SET0 = 0. 158.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23. p. We have AFP = /F(T0.87) l/M) (23. Felsager [91].86) = Jdrrjoir^Mr F(TO. .£O) OTQ = = £ " ./ drri0(T)<pc(T) + PJ (23.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r).89) Here the factor App is called the FaddeevPopov determinant.y/S^ro] (23.. (23.^) .91) " F o r more discussion of this point and related aspects see also B. (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).Zo)y/SETQ We now define A FP '• =0 • (23.(23. .<t>c{r)} .23. We evaluate this quantity as follows in the leading approximation.75) 1 = AFP f dT06[{F{rQ.88) ro0 Then in Eq.85) PJ W i t h Eq. (23.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o. so t h a t with F(TQ.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 ) .
e.92) We find therefore that the value of the FaddeevPopov determinant is given by the square root of twice the classical kinetic energy or simply of SE^ The relation (23./ dT0AFP. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. (23. The factor p is a normalization factor which we > retain here only for ease of comparison with literature and may otherwise simply be put equal to one.91) into Eq.90). P the relative sign being of no significance. i. the "static energy" invariant. in d(j)c(To) + dr/fa). fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). i. =~ (23.0AFP8 4o H P . P J AT (23. ~ There is a quick way to obtain the result (23.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 .612 CHAPTER 23. if the coefficient of 770(70) vanishes. 770(7"). 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. /(TO) ^/SE~T0} V~SET0 c d£. tangential to (J)C(T)) leaves this and hence the square of this quantity. . 0 d r (23. i.93) where AT = TT — 00. if ^ + d£o = 0.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction.e.e. Consider.e.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. Inserting (23.93). 7 Thus there is no fluctuation along the direction of the zero mode in the sum. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). using the relation (23. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe.
Since w\ is for finite T given by WQ of Eq. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.A T A F P i " o e x p T—>oo p where I'o n^O. This very important intermediate result is the contribution of a single instanton or kink configuration to the p a t h integral (ii has been defined by (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.60a) for I0 and Eq.96) lim . (23.96) does not yet allow a comparison with (23.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.60a) in not involving the zero mode contributions.55b).4.97) is a constant still to be evaluated.52a) and replace in this J d£o by the expression (23.23. we can set I 2TT ^o = l'o\ wlp2'' which with Eq. (23. We observe t h a t the result (23. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23. We determine I'Q by observing t h a t this differs from IQ of Eq.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.43)).93).4 Field 23. (23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo. 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. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. (23.98) . (23.
In view of their normalization of the normal modes. Although we choose p2 = 1.* Finally we rewrite the expression (23.a n t i . 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 .4>C exp h (23. in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells. i. 0c M. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .55b)) w0 = v0e VQ 23. This means we are concerned with configurations like* 2n+l d(r . Gildener and A. . (23. (23.20). ^See E. the result becomes h lim 2 m r ^ ^ 1 e .614 CHAPTER 23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.r).2 m T e x p 2 irgh ' T»oo (23.100) in such a way that it exhibits the crucial factors of which it is composed. we drag the factor p2 along for ease of comparison with literature.100) We observe that the dependence on the normalization constant p cancels out.4.102) *For comparison with the parameters of E. Patrascioiu [110]. they have to introduce the inverse of that factor on the far right of their expression (3.32). *For the validity of the calculation the instantons or kinks have to be far apart. their Eq.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. Gildener and A. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23.TiMiy+'Mr «>. ~ rKo')e{Tx . the approximation is then known as the "dilute gas approximation".5 I n s t a n t o n . (3.101) 2T > oo (cf.i n s t a n t o n contributions As already mentioned.e.
6.T2}((t)2.<t>2)eSE{<t>2.(l>i) (23.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0.r / 3 .fo) = Id2 St SE(<Pf.4 Field Fluctuations 615 for n = 1. (23.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations. .103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig. T3 = T) d(f>l((pf. (23. 7 i = .102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).Tl}{4>l.<t>ff + h . (n = 0 implies the single instanton solution).104) Considering the case n — 1 of a kink with one additional kinkantikink pair. Taking these configurations into account.4>I) ld2SE SE(c/>i. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. are varied. r 0 ( 2 ) = 0.Tf\(p2. ( 2 3 .106b) SE{<t>l.H)2 + . the total Feynman amplitude is given by (^/.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed. . 1 0 6 a ) and SE{(J>I.<Pi)2 + • • • . .T2\(f)l.l ) i + 1 in the sum of Eq. Proceeding as before we now consider first in analogy with Eq.r 0 ( 1 ) = .<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f. r0(3) = 2T/3.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .<j>i) . here (f>2.23. 23.Tl\(j)i. (23. Since these endpoint configurations. (3) (23. 2 .4>i.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.4>l)eSE{4>l. T2 = T/3.r/l&.36) (/.2 T / 3 . this means we have to consider the amplitude (23. The factor ( . Consider (with T 0 = T.
rn ^2 J2H 9 1m' ±2gcj). (23. dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .e.2d2SE ~c 4>i d(f)i e x p SE^f.107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.106a).4>i) { 2 " 2d SE .4>f) d4>\ 4>s72 (23. the .e.. d2SE{4>) d(j)2 2m.105) can in other respects be treated in analogy with the single instanton amplitude. On the assumption that the kinks of the instantons are widely separated.109)