•
Schrddinger Equation and Path Integral
INTRODUCTION TO QUANTUM MECHANICS
Schrodinger Equation and Path Integral
This page is intentionally left blank
*»«*?:#•
m
m
't
Schrodinger Equation and Path Integn
University of Kaiserslautern, Germany
\(P World Scientific
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • HONG KONG • T A I P E I • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
INTRODUCTION TO QUANTUM MECHANICS: Schrodinger Equation and Path Integral Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9812566910 ISBN 9812566929 (pbk)
Printed in Singapore by World Scientific Printers (S) Pte Ltd
Contents
Preface 1 Introduction 1.1 Origin and Discovery of Quantum Mechanics 1.2 Contradicting Discretization: Uncertainties 1.3 ParticleWave Dualism 1.4 ParticleWave Dualism and Uncertainties 1.4.1 Further thought experiments 1.5 Bohr's Complementarity Principle 1.6 Further Examples xv 1 1 7 12 14 17 19 20 23 23 23 29 29 31 34 38 41 41 41 49 53 54 55 59 59 60
2 Hamiltonian Mechanics 2.1 Introductory Remarks 2.2 The Hamilton Formalism 2.3 Liouville Equation, Probabilities 2.3.1 Single particle consideration 2.3.2 Ensemble consideration 2.4 Expectation Values of Observables 2.5 Extension beyond Classical Mechanics 3 Mathematical Foundations of Quantum Mechanics 3.1 Introductory Remarks 3.2 Hilbert Spaces 3.3 Operators in Hilbert Space 3.4 Linear Functionals and Distributions 3.4.1 Interpretation of distributions in physics 3.4.2 Properties of functionals and the delta distribution . . Dirac's Ket and BraFormalism 4.1 Introductory Remarks 4.2 Ket and Bra States v
4
VI
4.3 4.4 4.5 5
Linear Operators, Hermitian Operators Observables Representation Spaces and Basis Vectors
62 68 71 73 73 73 77 78 80 83 83 84 90 91 98 98 105 105 105 Ill 113 118 118 123 129 129 130 133 139 143 146 147 152 155
Schrodinger Equation and Liouville Equation 5.1 Introductory Remarks 5.2 The Density Matrix 5.3 The Probability Density p(x, t) 5.4 Schrodinger Equation and Liouville Equation 5.4.1 Evaluation of the density matrix Quantum Mechanics of the Harmonic Oscillator 6.1 Introductory Remarks 6.2 The OneDimensional Linear Oscillator 6.3 The Energy Representation of the Oscillator 6.4 The Configuration Space Representation 6.5 The Harmonic Oscillator Equation 6.5.1 Derivation of the generating function Green's Functions 7.1 Introductory Remarks 7.2 Timedependent and Timeindependent Cases 7.3 The Green's Function of a Free Particle 7.4 Green's Function of the Harmonic Oscillator 7.5 The Inverted Harmonic Oscillator 7.5.1 Wave packets 7.5.2 A particle's sojourn time T at the maximum TimeIndependent Perturbation Theory 8.1 Introductory Remarks 8.2 Asymptotic Series versus Convergent Series 8.2.1 The error function and Stokes discontinuities 8.2.2 Stokes discontinuities of oscillator functions 8.3 Asymptotic Series from Differential Equations 8.4 Formal Definition of Asymptotic Expansions 8.5 RayleighSchrodinger Perturbation Theory 8.6 Degenerate Perturbation Theory 8.7 DingleMiiller Perturbation Method
6
7
8
Vll
9
The 9.1 9.2 9.3 9.4
Density Matrix and Polarization Phenomena Introductory Remarks Reconsideration of Electrodynamics Schrodinger and Heisenberg Pictures The Liouville Equation
161 161 161 166 167 169 169 169 170 173 176 178 184 185 189 191 195 199 199 199 205 206 210 213 215 215 . . 219 223 227 234 237 239 243 249 249 250 251 254
10 Quantum Theory: The General Formalism 10.1 Introductory Remarks 10.2 States and Observables 10.2.1 Uncertainty relation for observables A, B 10.3 OneDimensional Systems 10.3.1 The translation operator U(a) 10.4 Equations of Motion 10.5 States of Finite Lifetime 10.6 The Interaction Picture 10.7 TimeDependent Perturbation Theory 10.8 Transitions into the Continuum 10.9 General TimeDependent Method 11 The Coulomb Interaction 11.1 Introductory Remarks 11.2 Separation of Variables, Angular Momentum 11.2.1 Separation of variables 11.3 Representation of Rotation Group 11.4 Angular Momentum:Angular Representation 11.5 Radial Equation for Hydrogenlike Atoms 11.6 Discrete Spectrum of the Coulomb Potential 11.6.1 The eigenvalues 11.6.2 Laguerre polynomials: Various definitions in use! 11.6.3 The eigenfunctions 11.6.4 Hydrogenlike atoms in parabolic coordinates 11.7 Continuous Spectrum of Coulomb Potential 11.7.1 The Rutherford formula 11.8 Scattering of a Wave Packet 11.9 Scattering Phase and Partial Waves 12 Quantum Mechanical Tunneling 12.1 Introductory Remarks 12.2 Continuity Equation and Conditions 12.3 The ShortRange Delta Potential 12.4 Scattering from a Potential Well
vm 12.5 Degenerate Potentials and Tunneling 13 Linear Potentials 13.1 Introductory Remarks 13.2 The Freely Falling Particle: Quantization 13.2.1 Superposition of de Broglie waves 13.2.2 Probability distribution at large times 13.3 Stationary States 13.4 The Saddle Point or Stationary Phase Method 14 Classical Limit and W K B Method 14.1 Introductory Remarks 14.2 Classical Limit and Hydrodynamics Analogy 14.3 The WKB Method 14.3.1 The approximate WKB solutions 14.3.2 Turning points and matching of WKB solutions . . . . 14.3.3 Linear approximation and matching 14.4 BohrSommerfeldWilson Quantization 14.5 Further Examples 15 Power Potentials 15.1 Introductory Remarks 15.2 The Power Potential 15.3 The ThreeDimensional Wave Function 16 Screened Coulomb Potentials 16.1 Introductory Remarks 16.2 Regge Trajectories 16.3 The SMatrix 16.4 The Energy Expansion 16.5 The SommerfeldWatson Transform 16.6 Concluding Remarks 17 Periodic Potentials 17.1 Introductory Remarks 17.2 Cosine Potential: Weak Coupling Solutions 17.2.1 The Floquet exponent 17.2.2 Four types of periodic solutions 17.3 Cosine Potential: Strong Coupling Solutions 17.3.1 Preliminary remarks 17.3.2 The solutions 259 265 265 265 266 270 272 276 281 281 282 286 286 290 293 297 301 307 307 308 315 319 319 322 328 329 330 336 339 339 341 341 350 353 353 354
ix
17.3.3 The eigenvalues 17.3.4 The level splitting 17.4 Elliptic and Ellipsoidal Potentials 17.4.1 Introduction 17.4.2 Solutions and eigenvalues 17.4.3 The level splitting 17.4.4 Reduction to Mathieu functions 17.5 Concluding Remarks 18 Anharmonic Oscillator Potentials 18.1 Introductory Remarks 18.2 The Inverted Double Well Potential 18.2.1 Defining the problem 18.2.2 Three pairs of solutions 18.2.3 Matching of solutions 18.2.4 Boundary conditions at the origin 18.2.5 Boundary conditions at infinity 18.2.6 The complex eigenvalues 18.3 The Double Well Potential 18.3.1 Defining the problem 18.3.2 Three pairs of solutions 18.3.3 Matching of solutions 18.3.4 Boundary conditions at the minima 18.3.5 Boundary conditions at the origin 18.3.6 Eigenvalues and level splitting 18.3.7 General Remarks 19 Singular Potentials 19.1 Introductory Remarks 19.2 The Potential 1/r 4 — Case of Small h2 19.2.1 Preliminary considerations 19.2.2 Small h solutions in terms of Bessel functions . . . . 19.2.3 Small h solutions in terms of hyperbolic functions . . 19.2.4 Notation and properties of solutions 19.2.5 Derivation of the Smatrix 19.2.6 Evaluation of the Smatrix 19.2.7 Calculation of the absorptivity 19.3 The Potential 1/r 4 — Case of Large h2 19.3.1 Preliminary remarks 19.3.2 The Floquet exponent for large h2 19.3.3 Construction of largeh 2 solutions
361 363 371 371 373 375 377 378 379 379 382 382 384 391 393 396 402 405 405 407 412 414 417 424 427 435 435 436 436 438 441 442 446 455 458 460 460 461 464
X
19.3.4 The connection formulas 19.3.5 Derivation of the Smatrix 19.4 Concluding Remarks 20 Large Order Behaviour of Perturbation Expansions 20.1 Introductory Remarks 20.2 Cosine Potential: Large Order Behaviour 20.3 Cosine Potential: Complex Eigenvalues 20.3.1 The decaying ground state 20.3.2 Decaying excited states 20.3.3 Relating the level splitting to imaginary E 20.3.4 Recalculation of large order behaviour 20.4 Cosine Potential: A Different Calculation 20.5 Anharmonic Oscillators 20.5.1 The inverted double well 20.5.2 The double well 20.6 General Remarks 21 The 21.1 21.2 21.3 Path Integral Formalism Introductory Remarks Path Integrals and Green's Functions The Green's Function for Potential V=0 21.3.1 Configuration space representation 21.3.2 Momentum space represenation Including V in First Order Perturbation Rederivation of the Rutherford Formula Path Integrals in Dirac's Notation Canonical Quantization from Path Integrals
466 468 470 471 471 476 479 479 486 493 494 495 500 500 501 502 503 503 504 510 510 513 514 518 524 533 537 537 539 544 549 554 557 564 570 574 579
21.4 21.5 21.6 21.7
22 Classical Field Configurations 22.1 Introductory Remarks 22.2 The Constant Classical Field 22.3 Soliton Theories in One Spatial Dimension 22.4 Stability of Classical Configurations 22.5 Bogomol'nyi Equations and Bounds 22.6 The Small Fluctuation Equation 22.7 Existence of FiniteEnergy Solutions 22.8 GinzburgLandau Vortices 22.9 Introduction to Homotopy Classes 22.10The Fundamental Homotopy Group
XI
23 Path Integrals and Instantons 23.1 Introductory Remarks 23.2 Instantons and AntiInstantons 23.3 The Level Difference 23.4 Field Fluctuations 23.4.1 The fluctuation equation 23.4.2 Evaluation of the functional integral 23.4.3 The FaddeevPopov constraint insertion 23.4.4 The single instanton contribution 23.4.5 Instantonantiinstanton contributions 23.5 Concluding Remarks
583 583 583 592 596 596 603 609 613 614 618
24 Path Integrals and Bounces on a Line 619 24.1 Introductory Remarks 619 24.2 The Bounce in a Simple Example 625 24.3 The Inverted Double Well: The Bounce and Complex Energy 631 24.3.1 The bounce solution 631 24.3.2 The single bounce contribution 635 24.3.3 Evaluation of the single bounce kernel 637 24.3.4 Sum over an infinite number of bounces 641 24.3.5 Comments 644 24.4 Inverted Double Well: Constant Solutions 644 24.5 The Cubic Potential and its Complex Energy 645 25 Periodic Classical Configurations 25.1 Introductory Remarks 25.2 The Double Well Theory on a Circle 25.2.1 Periodic configurations 25.2.2 The fluctuation equation 25.2.3 The limit of infinite period 25.3 The Inverted Double Well on a Circle 25.3.1 Periodic configurations 25.3.2 The fluctuation equation 25.3.3 The limit of infinite period 25.4 The SineGordon Theory on a Circle 25.4.1 Periodic configurations 25.4.2 The fluctuation equation 25.5 Conclusions 649 649 650 650 659 663 664 664 667 669 670 670 671 673
Xll
26 Path Integrals and Periodic Classical Configurations 675 26.1 Introductory Remarks 675 26.2 The Double Well and Periodic Instantons 676 26.2.1 Periodic configurations and the double well 676 26.2.2 Transition amplitude and Feynman kernel 678 26.2.3 Fluctuations about the periodic instanton 679 26.2.4 The single periodic instanton contribution 684 26.2.5 Sum over instantonantiinstanton pairs 688 26.3 The Cosine Potential and Periodic Instantons 690 26.3.1 Periodic configurations and the cosine potential . . . . 690 26.3.2 Transition amplitude and Feynman kernel 693 26.3.3 The fluctuation equation and its eigenmodes 694 26.3.4 The single periodic instanton contribution 696 26.3.5 Sum over instantonantiinstanton pairs 700 26.4 The Inverted Double Well and Periodic Instantons 702 26.4.1 Periodic configurations and the inverted double well . 702 26.4.2 Transition amplitude and Feynman kernel 705 26.4.3 The fluctuation equation and its eigenmodes 706 26.4.4 The single periodic bounce contribution 708 26.4.5 Summing over the infinite number of bounces 710 26.5 Concluding Remarks 714 27 Quantization of Systems with Constraints 27.1 Introductory Remarks 27.2 Constraints: How they arise 27.2.1 Singular Lagrangians 27.3 The Hamiltonian of Singular Systems 27.4 Persistence of Constraints in Course of Time 27.5 Constraints as Generators of a Gauge Group 27.6 Gauge Fixing and Dirac Quantization 27.7 The Formalism of Dirac Quantization 27.7.1 Poisson and Dirac brackets in field theory 27.8 Dirac Quantization of Free Electrodynamics 27.9 FaddeevJackiw Canonical Quantization 27.9.1 The method of Faddeev and Jackiw 28 The 28.1 28.2 28.3 QuantumClassical Crossover as Phase Transition Introductory Remarks Relating Period to Temperature Crossover in Previous Cases 28.3.1 The double well and phase transitions 715 715 717 720 723 726 727 734 736 740 740 745 745 753 753 755 756 757
Xlll
28.3.2 The cosine potential and phase transitions 28.4 Crossover in a Simple Spin Model 28.5 Concluding Remarks 29 Summarizing Remarks A Properties of Jacobian Elliptic Functions Bibliography Index
759 760 771 773 775 779 797
This page is intentionally left blank
Preface
With the discovery of quantization by Planck in 1900, quantum mechanics is now more than a hundred years old. However, a proper understanding of the phenomenon was gained only later in 1925 with the fundamental Heisenberg commutation relation or phase space algebra and the associated uncertainty principle. The resulting Schrodinger equation has ever since been the theoretical basis of atomic physics. The alternative formulation by Feynman in terms of path integrals appeared two to three decades later. Although the two approaches are basically equivalent, the Schrodinger equation has found much wider usefulness, particularly in applications, presumably, in view of its simpler mathematics. However, the realization that solutions of classical equations, notably in field theory, play an important role in our understanding of a large number of physical phenomena, intensified the interest in Feynman's formulation of quantum mechanics, so that today this method must be considered of equal basic significance. Thus there are two basic approaches to the solution of a quantum mechanical problem, and an understanding of both and their usefulness in respective domains calls for their application to exemplary problems and their comparison. This is our aim here on an introductory level. Throughout the development of theoretical physics two types of forces played an exceptional role: That of the restoring force of simple harmonic motion proportional to the displacement, and that in the Kepler problem proportional to the inverse square of the distance, i.e. Newton's gravitational force like that of the Coulomb potential. In the early development of quantum mechanics again oscillators appeared (though not really those of harmonic type) in Planck's quantization and the Coulomb potential in the Bohr model of the hydrogen atom. Again after the full and proper formulation of quantum mechanics with Heisenberg's phase space algebra and Born's wave function interpretation the oscillator and the Coulomb potentials provided the dominant and fully solvable models with a large number of at least approximate applications. To this day these two cases of interaction with nonresonant spectra feature as the standard and most important xv
In the first part. but arise also in recently studied models of large spins.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. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. in fact. we recapitulate the origin of quantum mechanics. its mathematical foundations. Excluding spin. even though the expansions are mostly asymptotic. Thus this first part . be treated to a considerable degree of satisfaction perturbatively. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. in Feynman diagrams in quantum electrodynamics). These basic cases will be dealt with in detail by both methods in this text. 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. again point the way: For scattering problems the path integral seems particularly convenient. basic postulates and standard applications. with degenerate vacua) were first and more easily derived from the Schrodinger equation. To what extent the two methods are actually equivalent. like periodic potentials. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. 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. screened Coulomb potentials and maybe singular potentials. With various techniques and deeper studies. the path integral method of Feynman was soon recognized to offer frequently a more general procedure of enforcing first quantization instead of the Schrodinger equation. but also about complex energies that he encounters in a parallel course on nuclear physics.g. numerous problems could. However. Chapters 1 to 14. whereas for the calculation of discrete eigenvalues the Schrodinger equation.e. With the growing importance of models in statistical mechanics and in field theory. has not always been understood well. one problem being that there are few nontrivial models which permit a deeper insight into their connection. the aforementioned exactly solvable cases. that is the Coulomb potential and the harmonic oscillator.
and their eigenvalues. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). 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.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. 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). i. 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. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. the chapter thereafter deals with Yukawa potentials. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. Using perturbation theory. which is presumably the only such singular case permitting complete solution and was achieved only recently. In the second part. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. Chapters 15 to 20. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. we deal mostly with applications depending on perturbation theory. The most prominent examples here are the double well potential and its inverted form. After a treatment of power potentials. The following chapters deal with the derivation of level splitting formulas (including excited states) for periodic potentials and anharmonic oscillators and — in the oneloop approximation considered — are shown to agree with those obtained by perturbation theory with associated boundary conditions. The potentials with degenerate minima will be seen to reappear throughout the text. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). Thereafter the concepts of instantons. 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.e. and the elliptic or Lame potential — here introduced earlier as a generaliza .
The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. except that these are no longer of hypergeometric type. for instance. such as limiting procedures. 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. anharmonic oscillator potentials on a comparable level with. 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.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. this comparison on a transparent level being one of the main aims of this text. This method is therefore more complicated. it is. and whenever available also with the results of WKB calculations. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics.g. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. 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 can make the following observations. This puts Schrodinger equations with e. 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. All results are compared with those obtained by perturbation theory. 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 Mathieu equation. from our precise reference to unavoidable elliptic integrals taken from Tables). With a fully systematic perturbation method and with applied boundary conditions. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. In fact. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. for instance. cubic). We then illustrate the relevance of this in the method of collective coordinates. however. quartic. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. an approximation whose higher .
J. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). B. B. Throughout the text some calculations which require special attention. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. in particular Professors J. as well as applications and illustrations. Whittaker and Watson [283]. shows him that each such formula here has been properly looked up). 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. at the end of a chapter (after troublesome turning of pages). the references referred to are never cited by mere numbers which have to be identified e. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). As a rule. The line of thinking underlying this text grew out of the author's association with Professor R. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. like E. W. with the source given in the bibliography at the end. Park (Masan). The author is deeply indebted to his onetime supervisor Professor R. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. N.g. MiillerKirsten *In the running text references are cited like e. D. 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. Instead a glance at a nearby footnote provides the reader immediately the names of authors. Andrews). so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. 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. Nonetheless.Q. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. Watson [283]. Since this comparison was the guideline in writing the text. e. T. 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. also for the verification of results. Whittaker and G. are relegated to separate subsections which — lacking a better name — we refer to as Examples. H. For ease of reading. D. we also consider at various points of the text comparisons with WKB approximations. thereafter University of St. .* H.g. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. Liang (Taiyuan). K.g. Dingle (then University of Western Australia.
Kirchhoff's law in thermodynamics says that in the case of equilibrium. very few explain in this context what he really did in view of involvement with statistical mechanics. 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. A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it.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).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. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 .Chapter 1 Introduction 1. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. We do not enter here into detailed considerations of Planck. Instead. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). the amount of radiation absorbed by a body is equal to the amount the body 1 . we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery.
T) were known and tested experimentally.1 Absorption in a cavity. Introduction emits.3 4 J s.1) ' y Here u(v. energy per unit volume) of the radiation (i.1) c is the velocity of light with c = u\.38 x 1(T 23 J K'1.1) 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.l ) where x = ^ = ^ . These expressions are: (1) Wien's law. v + dv in equilibrium with the black body at temperature T. and the other in the region of large A (or AT). i. y J c 3 \ex . (1. Black bodies as good absorbers are therefore also good emitters. 1.e. (1.T) = 2*?£(?)kT. In Eq. the formula (to be explained) u(u.. Fig. kT kXT (1.2) and the . and then measuring the increase in temperature of the heat bath.2 CHAPTER 1.e. Let us look at the final result of Planck. It was found that one expression agreed well with observations in the region of small A (or AT). T)du is the mean energy density (i. How did Planck arrive at the expression (1. The parameters k and h are the constants of Boltzmann and Planck: k = 1.T) = dv3eC2U/T.e. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. radiators. h = 6.e.626 x 10 . A being the wavelength of the radiation. 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. u(u. i.
3) are contained in Eq. he searched for a derivation. (x large). Here W is a number which determines the distribution of the energy among a discrete number of objects. in the first place Planck had tried to find an expression linking both.e.3) Ci. that the formulas (1. and thus over a discrete number of admissible states.T) = av e6"/Ti' where a and b are constants. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons".1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. C3 being constants. and he had succeeded in finding such an expression of the form u(v. which are indistinguishable) among the N indistinguishable oscillators at . Moreover Planck assumed that these oscillators do not absorb or emit energy continuously.1) in regions of a. T) u(i/. "small" (i. (1. . We see. To this end he considered Boltzmann's formula S — klnW for the entropy S.2) and (1. (1. 3 (1.47TZ/ 2 {x small).1. exp(x) ~ 1+x) and "large" (exp(—x) < 1). Indeed. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. C2. 1. discretization begins to enter.2 Distributing quanta (dots) among oscillators (boxes).T) = 2^C3T.1) as approximations. we obtain: u(i/. e xhv. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. but — here the discreteness appears properly — only in elements (quanta) e. This is the point. Considering Eq.T) 2^^kT. where the Fig. When Planck had found this expression.
there not called Stirling's formula. h = const. as in most other Tables.2) requires that e ex is. 1. i.2. = 1/T). N * oo.4) IniV! ~ JVlniViV + O(0). Oberhettinger [181]. (1. The Stirling formula or approximation will appear frequently in later chapters.T)dv. . Introduction temperature T.e. Agreement with Eq.343(2). Then W is given by w = (N With the help of Stirling's formula* {N + piy. U{T) being the average energy emitted by one oscillator. Gradshteyn and I. 1. with the quanta represented schematically by dots as indicated in Fig. formula 8. as for small values of e). .1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.e.6) Fig. Ryzhik [122]. M.4 CHAPTER 1. Magnus and F. 940.3 Comparing the polarization modes with those of a 2dimensional oscillator. i. (1. with riydu — 2 x —w—dv.g. v + dv. S. one obtains (cf. e = his. W.g. We visualize the iV oscillators as boxes separated by N — 1 walls. e. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. p. u(i>. p.3. (1. and the second law of thermodynamics ((dS/dU)v Example 1. We now obtain the energy density of the radiation. I.1)!P! (1.7) *See e.
The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.3 (as for ideal conductors).3.1)) The solutions of this equation are ^max = 4.1).8) This is Planck's formula (1. i.T) has a maximum which follows from du/dX = 0 (with c = vX). L for i = 1..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. is given by .2.2.UJ = 2KV.1. so that . We obtain the expression (1. rii = 1.e. 2 2 2 r2 L K — 7T n .. so that the derivative of u implies (x as in Eq.7). nvdv per unit volume.3.V 2 ) E = 0. (1. so that (lvL\A 0 I I = rr. . '''From the equation I \ JW . We observe that u(v. 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 in electrodynamics..7) for instance.1.. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. Then L^j = nrii. dj\l dM . where^ 2 [2KUY .UJ2/C2 + K? = 0.T)dX = ^ehc/*kT_idX.2. We obtain therefore u^T) = Unv = 2^fJ^—i.i = 1. (1.965 and xmin = 0. In terms of A we have u(X. (1.. The number with frequency v in the interval v. as indicated in Fig.. KT = I J . v + dv. .3.
We therefore examine such contradictions next. Lorentz and Planck that Eq. n = 0. However. One might suppose now. Introduction he 4. . 2 . i.10).e. but one can expect an analogy. we then have at T = 0 independent oscillators. and from which the constant h can be determined from the known value of k. which had also been known before Planck's discovery. Since temperature originates through contact with other oscillators.1. .e. 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=—. of course. then (with x = hv/kT) its mean energy is En=0e = nX dx ^0 — /ii/— In = huf r%e dx 1 — e_x (1 — e~x)z hv . If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. One expects. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1.8) could be derived much more easily in the context of statistical mechanics.965K = Const. Later it was realized by H. x — oo) the mean energy vanishes (0 < U < * > oo). We are not dealing with the linear harmonic oscillator familiar from mechanics here. that it is easier to consider first the case of T = 0. which can assume the discrete energies en — nhv.6 The first value yields AmaxT = CHAPTER 1. This is Wien's displacement law. which did not arise in Planck's consideration of 1900. that of an oscillation system at absolute temperature T ^ 0.l. ra = 0. . .10) Thus here the socalled zero point energy appears.—v <L9) We observe that for T — 0 (i. Thus we have a rather complicated system here. the behaviour of the system at zero absolute temperature. (1. A. .2 (1. such a procedure leads to contradictions. which can not be eliminated without a different approach.
kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.1. In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente"). . 1. {NW •• 1)! (TV1)!P! and P = UN Show with the help of Stirling's formula that the mean energy U of an oscillator is given by U •• exp(e/fcT) .1)! .2 Contradicting Discretization: Uncertainties 7 Example 1. for 2 degrees of freedom. we obtain T V S =fc[ln(TV+ P .In .k dU 1+U\ f ln(l T + U\  U U k ( e .1 ' u~ e .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.ln(TV .e.1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV.InP!] ~ kN The second law of thermodynamics says 1+ 7 ln 1 + ( 7)~7ln7 \au)v For a single oscillator the entropy is s = S/N. i. kT/2. with Heisenberg's discovery of the uncertainty relation.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .i n .. We "This is what was effectively done before 1925 in Bohr's and Sommerfeld's atomic models and is today referred to as "old quantum theory". so that 1 f ds\ T — \dUjy .= . d .1)! . where k is Boltzmann's constant and W is the number of times P indistinguishable elements of energy e can be distributed among T V indistinguishable oscillators. around 1926.2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.
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
.t). H. Rather.n n. It suggests itself therefore to consider more closely the properties of the symbols (2.) = £ [wm + WiK) = £ [WiWi . chapter VII.pi) of qi. which can be observed directly at time t. ^As H. (2. since .24 CHAPTER 2. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics.2) If we have only one degree of freedom (i = 1).P.1) as d .2) as This equation is. as mass times velocity. Hamiltonian Mechanics is wellknown.^ ^ j . Pi has to be considered as an independent quantity. A system consisting of several mass points is therefore described by a number of such variables. qi(t + 6t) . Goldstein [114] remarks at the end of his chapter VIII. one obtains the Hamilton equations* • OH .g. (2.Pi) is contained implicitly in the canonical variables q^ and pi..e. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.3). (2. (2.p% are therefore again observables. The following properties can be verified: *See e. The total time derivative of u can therefore be rewritten with the help of Eqs. Goldstein [114]. i. M.1). whereas the velocity requires observations of space coordinates at different times.4) as the generalization of Eqs. (2. which all together describe the state of the system. Compared with an arbitrary function f(qi. chapter VIII. in analogy with Eqs.4) contains as special cases the Hamilton Eqs. du \ x^/dudH du dH\ . Dirac [75]. All functions u(qi. We can therefore consider Eq. A. .qi(t) qi = hm f . the standard reference for the application of Poisson brackets is the book of P.Pi. the equation of motion of the observable u.1). the entire timedependence of observables u(qi. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. 6t>0 5t Real quantities which are directly observable are called observables. x^fdu. (2. One can verify readily that Eq. S »(«. (2.1). (2. this expression is simply a functional determinant.
like here.B}C or C{A.Pk} = 5ik. If.5b) (3) complex conjugation (note: observables are real.BC} (5) Jacobi identity: {A. that the very general Eq. The original definition of the Poisson bracket will not . (2.Qk} = 0.B} = {B.5a) (2. {C. a i S i + a2B2} = ax{A. {pi.6) We can now show. we obtain the fundamental Poisson brackets.B}* (4) product formation: {A. for example. that of the linear harmonic oscillator. where A and B are arbitrary observables. If we evaluate the Poisson brackets for qi. B}} = 0.5d) is useful in calculations. the Poisson bracket {A.5d) = {A*. C}} + {B. 25 (2.5e) = {A.3) of the Poisson bracket.B}. we wish to evaluate {A. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2.2 The Hamilton Formalism (1) Antisymmetry: {A. {qi. Bx} + a2{A. (2) linearity: {A.A}. Later we shall consider noncommuting quantities.C}. A}} + {C.5c) The first three properties are readily seen to hold. (2. As long as we are concerned with commuting quantities. As an example we consider a case we shall encounter again and again.2. B} is completely evaluated. These are {Qi. it is irrelevant whether we write {A.6). (2. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain.B}C + B{A. {B. which combines the Hamilton equations. Since Eqs. i. B2}. {A. Property (2. (2. in other words without any reference to the original definition (2. but could be multiplied by a complex number): {A.Pk} = 0. then the ordering is taken as in (2. (2.Pi.6) give the values of these.5d) above. B}. (2.e.4).B*}.
Since constants are also irrelevant in this context.p}p + p{q. q = {q. These are transformations qi—>Qi = Qi(q. Hamiltonian Mechanics be used at all. or q(t) = qo+ Pot . (2.26 CHAPTER 2.11a) 'q' = q.12) .8a) and use the properties of the Poisson bracket and Eqs. In the evaluation one should also note that the fact that g.p) is an ordinary function is also irrelevant. Then we obtain: (2. we consider as Hamiltonian the function H(q.H}.\(p2 + q2)} = l({q.7) into (2.p. (2. from which we infer that q(t) — qocost + posint. q + q = o.yPot3 + •••• (2.11b) (2. and so q = q. (2.H}. and Pi are ordinary real number variables and that H(q. and P={p.p.t).6). (2. From Eqs.. According to Eq.7) q = [q. "4' = q. (2.8a) (2. (2..9) Similarly we obtain from Eq..8b) p=q.. (2.9) and (2.P2} + {q.8b) We insert (2.10) In classical mechanics one studies also canonical transformations.4) we have for u = q.p) = ±(p2 + q2).p\) P.t).qot2 . Pi—> Pi = Pi(q. (2.p.q2}) = = 2i{q.10) we deduce q = p = q.
12) Qi—>qi = qi{Q.2 below contains a further illustration of the use of Poisson brackets. i. (2.P One can then show that {A. that (dropping the subscripts therefore) {Pi. i.B}q.Qk} = 0.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P).P is canonical in the sense defined above.B}Q>P.13) We write the reversal of the transformation (2.15a). Of course. which transforms qi. Example 2.pk} = 5ik. B} of two observables A and B in terms of either set of canonical variables.1 we verify only Eq. i.t). which means that a Hamilton function K(Q. . as {A. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.B}q. and Example 2. i. .Pk} = 0.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. Pi^Pi=Pi(Q. those at a time t — 0. for which Hamilton's equations hold.2. (2. {Qi.e. verify that the Poisson bracket of two observables A and B is invariant.P provided the transformation q.t). *H.e.3) we can now express the Poisson bracket {^4.P. {qi.e. (2. _ax p__dK_ (2.p = {A.* Hence in Example 2.3 deals with the relativistic extension. Eq. (2.qk} = 0. {PhPk} = 0.15b) The proof of the latter invariance is too long to be reproduced in detail here but can be found in the book of Goldstein. (2.14) With the help of the definition (2.15a). The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.15a) °raS {AiB}Q.e.p <> Q.e. {qi. Goldstein [114]. {QhPk} = Sik. chapter VIII. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here. P) exists. Example 2. i.Pi back to their constant initial values.P.
A}q.B}q.Qj}q.p) . § E x a m p l e 2. Qk}q. their product Et — qp being relativistically invariant. {A.p(0) = poSolution: The solution can be looked up in the literature. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.E(t).3: Relativistic Poisson brackets By extending qi. V V °Qk dPk dPk dQk J E x a m p l e 2. 236. Solution: Relativistically we have to treat space and time on an equal footing.28 CHAPTER 2.P^pr.p^ "Pk.q/c) and (E. Thus we extend q and p to spacetime vectors (t.p = J2 d~Fk Replacing in the above A by P and B by A. .p v ^ (9A dB ^  _ _ *? V dqj dpj dA 8B —— dpj dqj 8B_dI\\ dPk dpj J _ 8A_f^B_dQk dpj \ 8Qk dqj dB 8Pk dPk dqj = E E k dAL/'^B_dQk dqj \ dQk dpj {A.15a) r„ ™ v .+ dQ.s F 5 7 r ={A. we obtain as claimed by Eq. we obtain dA {Qk. dA {Qk.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential).B}QIP.F}r ~di d~E ~ d~E~8t Consider F = H(q.pc).Pj}q. we obtain analogously dA dQk' Inserting both of these results into the first equation. we have —Et. p. OQk Pk}q.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. Thus whenever q and p are multiplied. Mittelstaedt [197].( 9A = {Pk.p = dB dA dB \ rA 1 {A'B^ E b r s r .A}q.p^— + {A.15b) dA {Qk. (2. § See P. (2.p^v dP. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. Replacing here A by Q and B by A. Hamiltonian Mechanics S o l u t i o n : U s i n g t h e definition of t h e P o i s s o n b r a c k e t a p p l i e d t o ^4 a n d B w e h a v e r „ „•.
Pi are the coordinates of some mass point m.po is a point on Go. one obtains Gi with q = q((lo.. g oc AqAp. and E as a function of t). Probabilities 29 (This is. du . I . and the space spanned by the entire set of canonical coordinates is described as its phase space.t0. if qo. dudH s l u . also boundary points of one domain are mapped into boundary domains of the other.dt y/l  qZ/c2' 2.3. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . (*.p) . But now we consider a system whose phase space coordinates are not known precisely.po) = Gi(q. p= p(qo. Since Hamilton's equations give a continuous map of one domain onto another.._. Of course.17a).p) to G\{q. (1. dE .2. c dt 2 2 w^ du du dr . of course.y. since H is expressed as a function of q and p. so that Go(qo.)•'=(*)**£. and at time t > to in a domain G\(q.. Then . but partial derivatives of F do not vanish.14) and area A of the phase space ellipse given by Eq.po.16) For example.E(t)}r = Hence at Relativistically we really should have clu/dr.p). i. g. Let us assume that at some initial time to the system may be found in a domain Go(q. We distinguish in the following between two kinds of probabilities.1 Liouville Equation.. Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.t). rrr . numerically zero. in the case of the linear oscillator with energy E given by Eq..e.p) around some point qo. it is the equations of motion which lead from Go(q. We consider first the a priori weighting or a priori probability.to. 2TTE UJ and hence g oc — . This probability is evidently proportional to AqAp. .3 Liouville Equation. we have A = (p apaq — J . (2. du q\ P = —. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.p). (1. .e.t).Po.' ±=j1V c \dtj' u^. to . H(q.Po. 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.3 2. i.p).
Thus g must be independent of t. — dp A A — In(AqAp)s = — + — = Example 2. Pi 2 / \.)curve for constant E and $ is — as may be seen by comparison with Eqs./.1): d . Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle.4: Liouville's theorem Show that A q A p is independent of time i.14) t o (1. dpdq and similarly dt = — Ap. = 2TrIEsm6. dt dq and with with Hamilton's equations (2. Example 2. Show that the total volume of phase space covered for constant E is 8n2IE. the (pg. (1. in view of this independence it can be expressed in terms of the conserved energy E. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates.eded<t> = 8TT2IE. Example 2. dt dq dq dp dp d2H dqdp d2H = 0. this has the same value at a time to. 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.— = — Aq.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 . which means.30 CHAPTER 2. J0=0 Hence g oc 8n2IdE.p^. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. Je=0 . as is demonstrated by Liouville's theorem in Example 2.15) — an ellipse of area § dpgdp. Hence from the difference: — . and hence g oc %ir2IdE.. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' .5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E.4.
(2. it is suggestive to introduce a factor 2KK with every pair dpdq without.p + dp.3.e.19) Thus W is the probability to find the system at time t at q. leaving the basis of classical mechanics! Hence we set / «w)^ =. (2. Probabilities 2. however.p.1 The system moving from domain Go to domain G\. Thus dn is that number of systems which at time t are contained in the domain q.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.q + dq.p.t).t)dqdp= 1. (2.Po. W=p(q. i.p.17) F ^GT^^ 0 Fig. (2. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space.20) . by jr = P(9»P. i. The total number of systems N is obtained by integrating over the whole of phase space.18) With a suitable normalization we can write this / W(q.Thus we assume a large number of identical sytems. 2.*). Since W has the dimension of a reciprocal action. whose positions in phase space are characterized by points.t)dqdp = N. d d 1 P = JJdqidpi.p.p.e.2. dn — p(q.3 Liouville Equation.
p. In order to derive this equation.p.p) denote the velocities in directions q and p — p{q. t + dt)dqdp — p(q.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. 2. t + dt)dqdp. i.p) and vp(q. p(q.t)dp{ jt Q.p. 2. t)dqdp p(q. i.p / .3 The region G. * q Fig.e.e. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. We are now interested in how n ov W changes in time.t)dp( — I . that in our consideration no additional points are created or destroyed. — if vq(q.a t dt + q+dq.p. In doing this.p. The equation of motion for n or W is the socalled Liouville equation. 2.2 The ensemble in phase space.P dt p(q + dq. p + dp G P O q qndq Fig.e. how the system moves about in phase space. we consider the domain G in Fig. i. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.32 CHAPTER 2. we take into account. The number of points at time t + dt in domain G.
p + dp) dp = or  However.P *' % = iH>P} ( 2 .p)dtdp +p(q. (2.p.p)dtdq .21) with Eq. t)dqdp = p(q.p.p) dq p(q. (2.t)^ = 1.p) . .t + dt) . .2.t)vp(q.p. (2. .p.20) and (2. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { .21) we can also write dW(q.p(q.p. (2.P)].p)dtdp . p.p + dp.3 Liouville Equation.p).t)vp(q. = K(9. put differently.t)vp(q.p + dp)dtdq.t)vq(q + dq.t) dt {H(q. t)vq(q + dq. p. .t)} with JW(q.p.4) shows that p and u satisfy very similar equations.t)vp(q.t) dt p(q.p) = q = g^. (2.W(q.p(q. the probable motion of the system under consideration.p) .p. Probabilities and thus p(q.p(q + dq.t)vq{q.t + dt)dqdp — p(q. Comparison of Eq.t)vq(q.p(q + dq.p.P)]K(Q.p + dp.p(q. dH Mq. vp(q.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . With Eqs.21 ) This is the Liouville equation which describes the motion of the ensemble or. 33 Dividing both sides by dqdpdt this becomes p(q.p.P.19).p.p)=p = 3H —.
e.e. We define as expectation value of u(q. q+dq. (.that the density or probability W(q. Thus in Fig. t) depends explicitly on time t (if determined at a fixed .26) we described the time variation of the observable u(q. p. i. i.p (cf. We now inquire about the time variation of the expectation value (it) of u.4).24).1 the area Go is equal to the area G\. (2. that dtj\y v if qo. p+dp.p) the following expression: (u)=Ju(q. Equation (2. 2. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.po are the initial values of q. We deduce from the Liouville equation the important consequence that ^ M = 0. (2. and hence that equal phase space volumes contain the same number of systems. We shall see that we have two possibilities for this.4 Expectation Values of Observables Let u = u(q.p) be an observable.24) since the total derivative is made up of precisely the partial derivatives contained in Eq.t)(^J.p.p). (2. The first and most immediate possibility is — as indicated . 2. With Eq.p)W(q. Example 2. for i<«>!/«<*p>"w>(^)".34 CHAPTER 2.24) implies that p is a constant in time. since no systems are created or destroyed.4).p. (2.
Goldstein [114]. since W oc p is constant in time according to Eq.31) i.t).Po.f(q0. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo. Reversing Eq.22). (2. we can express these in terms of their initial values qo.p.t).0) = W0(q0.2. and hence u(q. Thus we can write.p). (2.Po. Observables 35 point in phase space).p.t).t). at t = 0.p. ie. However.e. (2. (2.34) S e e also H.p. of course.p) = u(q. We verify this claim as follows.t) = = W(g(q0. 1 (2. (2.t).p) = uo{qo.p. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.29) The distribution of the canonical variables is given by W(q.t).t)}(^y.^ Solving the equations of motion for q. we have Qo = g(q.f(qo. where we used Eq.32) In this expression the time t is contained explicitly in the observable u(q. we can also employ a more complicated consideration.24): W(q.p0.t)W0(q0. that (u) = (u)0. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.0) = u0{qo.Po. .po.4 Expectation Values. 8.i). (2.33) po = f(q. (2.27) becomes !<»>  /«(*P>!"W>(^)" (2.p0)(^^J.30) p = f(qo.t).po.p.po. We expect. Then Eq. so that Q = g(qo.t). Sec. (2.p) assumes certain values) that depends explicitly on time.t).8.29).p.Po) at time t = 0.po.Po.t) W(q0.28) = Ju(q.t).W(q.Po.p){H(q.po. (2.0) =u(g(qo.
[ du0{q0. we obtain an expression which is different from that in Eq. d dt {U)0 = dt.30) Moreover. (2. Eq.t). (2. (2.P0.Po.t).) f / u (q0.t).p.37) and (2. p = f(<j(q. i.t) M f 9u\ V °P / P=f(q0. we obtain («)o = / u{q. / ^fdq0dpc .f(q.f(g(q.f(q.p.po.t).t).po)(~ .30) u0(g(q.p.t) d t _ (du\ \ d<i dq / P=f(qo.t)W "(q0.32).POit)i p=f(qo.35) Inserting these expressions into no we obtain uo(qo. (2.t) (2.30): duo(qQ.p)W{q.M 27Th i . Taking now the total time derivative of (it)o.t) 0H_ 0 q =  {{H(q. (240) .p.p.39) Substituting this into Eq.t).po.P.p).36 CHAPTER 2.f(q.p)})q=g(QO.32).Po. (2. .) Wo(qo.PO.=9(. Eq.o.e.u{q.t).f(q. (2.Po.t).36) = (2.t) = (2 30) (2.36) and (2. (2.t).28).t) v{g(<j(q.p).t) dp M du\ ° q 'P=f(qo.P.37) as had to be shown. (2.t).Po)(^^)n." V " " ™0' > • • » « » 0 .\„.P0.u(q.f({H(q.25) into Eq.p0) = W(q.P)}).t).t) dB__ (du\ °P \°PS P=f(qo. (cf.p.t).(dq0dp0\n We deal with the partial derivative with the help of Eq.38) we obtain £<«)„ = .t).t) . Hamiltonian Mechanics so that on the other hand with Eq.p.(2.p).t).p.t)\—^\ = (u). (2.p. (2.Po.35).0)=u(q. Inserting Eqs. (2.po.31)) W0(q0.0) u(q.t)J(q.p.29) q = g(g(q.p.
In the first case.28) and (2.W} in Eq.p). lim ——uW — 0.p).t) of u assuming certain values q and p.p)}W(q.t)} . and the time variation of (u(q.31).. the observable u is treated as a function u(q. ^ „. so that ~d~t <«>0 = J{H(q.(2.p).41) This expression contains {H. 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.t)}.34) and use (2.p.p. (2.p){H(q. However.u(q.42) the relation d . (2.25) and (2.p)) is attributed to the probability W(q. x r „. Observables 37 Here we perform the transformation (2.uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2.28).p).p.p. (2. . we obtain zero after partial integration of I and hence from Eqs.45) the Liouville equation. from the properties of the Poisson bracket.41) and (2.43) ——uW = 0. u}W instead of u{H.28). (2.. / d q d p \ n 2irh J (2. The phasespace integral of a Poisson bracket like I (2.2.W(q. we obtain {H.4 Expectation Values.W(q.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. .45) in agreement with Eq. Consider {H. Alternatively we could deduce from Eqs. dt <«)o v f u(q. W}. described as "Schrodinger picture".44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). This timedependence is described by the Liouville equation dW(q.. u}W + u{H.t) dqdp\n 2irhJ (2. uW} = {H.t) dt {H(q.
p)}.p0. The equation of motion is then that of an observable. Hamiltonian Mechanics In the other case.4) of an observable on the other. does not have to be commutative. the probability of the initial values Wo(qo.rt.4)) ^ = {«(. 2. " This means: Only in the Heisenberg picture dW/dt = 0.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. but does satisfy the usual associative and distributive laws. (b) As usual a muliplication by a complex number is defined.t) dt du(q.t). as the equation of motion of an ensemble or of a probability distribution on the one hand. This formulation deals with observables u representing physical quantities.po. the reason being that — since qo. as we observed.p)}. (2.** These considerations point the way to a generalization which results if we permit u and W to belong to a more general class of mathematical quantities." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo.H(q. (cf.po are constant initial values — we have du0(q0.po. (2. Thus we arrive at a more general theory if we define u and W with the following properties: (a) An addition is defined between the quantities. " T h e author learned this approach from lectures of H.t) dt __ ~ (2J30) ( 4) du0(qo. and the equation of motion (2. Koppe [152] at the university of Munich around 1964. which.e. called u Heisenberg picture". Eq. i.F(«.. (c) A multiplication of the quantities among themselves is defined. .p) It = {u(q.p). for which the axioms of a commutative group apply. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. which describe the state of a system. The time dependence of the expectation values can be dealt with in two different ways.po) is assumed.46) We thus also recognize the connection between the Liouville equation. and probabilities W.38 CHAPTER 2.
if u ^ 0. also called "density matrix".48) In matrix theory the symbol "trace" has a welldefined meaning.2. Eq.p)(^y. for a phasespace integral we define the word or symbol "trace". The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''.p} = l {q. (2. Moreover. (2. and to be able to correlate these with the above classical considerations. it is helpful to introduce already at this stage some additional terminology. y:\p. Introducing it here assumes.q}=0. which all apply to (2.q} = 0 {p. In Chapter 9 we attempt a corresponding approach for classical systems with .26)) (u) = Tr(wW). (d) Tr(ut.48): (a) TV u* = (Tr «)* = TV^. by Traceu:=Ju(q.6) a socalled "commutator''' [A. That this is the case. so that we consider this next.p] = l. can be deduced from the following characteristic properties of a trace. that its use here implies the essential properties it has in matrix theory. ~[q. px > in—. Thus we can write the expectation value of an observable u (cf.) = TV (vu). (2.p} = 0 • — ^ ~[q. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. (c) TV (u*u) > 0. (b) TV(cm + j3u) = oTV u + /3Tr v. therefore. also written "TV".49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. ox In view of our later considerations.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. B] := AB — BA in the following way: 'i {q.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.
e. excepting the Poisson brackets. it will be shown that in electrodynamics.e. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. a constraint) has to be taken into account. . and. electrodynamics. the Poisson brackets require modification to Dirac brackets. Hamiltonian Mechanics a wavelike nature. i. since gauge fixing (i.40 CHAPTER 2. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. However. These aspects will be considered in Chapter 27.
which turn out to be those of the theory today known as quantum mechanics. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters. we can define the Hilbert space as the space of states of a physical system. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. Building upon this. of measurable quantities. A set M.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space.e. if the elements ipi of M satisfy the usual axioms of addition and K 41 . We found that the Poisson algebra permits extensions to noncnumber formulations. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. i. 3. 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. i. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. and selfadjoint operators in this space as representatives of observables.e.
3) If all on = 0. n is called the dimension of . such that n ^ = YJCi^i. (a.02)* (hermiticity). In each case n linearly independent vectors are said to form a basis..4) The vector space . ( n (3. f > 0 if V T ^ O .  .aiV>i + a 2 ^ 2 ) = ai(ip. . . (3. .ipn are said to be linearly dependent. ^ i = 0 if ^ = 0. x M. .ip2. 2 . . and n is the smallest such number.2) Vectors ip\.~ip2) '• M. . Mathematical Foundations of Quantum Mechanics multiplication by complex numbers./?eK). fa. If n + 1 elements ^ € M. (aptyi. ip2 of this space can be associated with a complex number (V'l. n exist (not all zero). are linearly dependent.ip2). if every vector ip E M can be associated with numbers Q . n.M. i=l (3. .1) aipi + aipj. i = 1. . • • • . (V>.e.ipn are said to be linearly independent.^) called inner product.tp2. E (3.5b) .i = 1.42 CHAPTER 3. A + tpj = tpj + i'i.5a) (3. 2 . —»• IK.^i) .M is said to be a metric vector space or a preHilbert space. the vectors il)i. (ipi. . if numbers eti. so that n 5 > ^ = 0. (ipi + ipj) + ijjk = ipi + (ipj + ipk). if any two elements ip\. (3. . i. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M).^1) = (01. (3 5c) + a2{ip. with the properties (a* G IK): (^2.
In order to verify Eq.^1) 2 h 11 0 <%«e + M .V.)> (36) i.7) In addition the following relations hold in a metric space M.2 Hilbert Spaces 43 where the asterix * means complex conjugation. (3. ^ 2 )  .9d)   V i   = sup  ( ^ i . The distance between two vectors if>i. for arbitrary A and ip2 ^ 0: (V>i + AV>2. (310) (^2.9c) HV'i + tp2\\2 + \\1p1 . for ip\. meaning oneandahalffold linearity). The first two properties imply ( a i ^ i + a2ip2. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2. (3.*h)=\\AH\(38) (3. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).3.V') +"2*(V.V>i + A</>2) > 0 .rh) (3. ^ i ) + A ( ^ i . if Wi.2.tp2 £ M.e. »2 2 .is defined by d(ipi..9b) = 0 (Pythagoras theorem). which we can write 0 < ( ^ i .^) = ai*(V'i. (3. tp2 £ M: 1(^1. (3.9a) we start from if) = ipi + \if>2 6 M. WA+Ml2 = ll^il 2 + ll^2 2 .9a) (3.^ 2 ) =   ^   2 + 2 ^ i . A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. ^ ) + A 2 (V2.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation). ^ i ) + A * ( ^ . antilinearity in the first component (also described as sesquilinearity..^2)1 < H^ill • 11^211 (Schwarz inequality). linearity in the second component.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.
10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. hence verification in each case is necessary.* Two vectors i f i . ll^ill + llVdl. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. Examples of metric vector spaces: (1) Let M. (3. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . ^ e M are said to be orthogonal if (</>!.ii) thus verifying Eq. so that   ^ l + ^ 2   <   ^ l   + ^2. .^1)1 Vi 2 + ^2 2 + 2   ^   .V2) IIV'l2 + V'22 + 21(^2. (3.9c). o ^ V i = o. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0.9a).9e). i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. (3.  ^ l l = (HlMI +   « 2 .13) the vector is said to be normalized. (3. HIHI (3.\\ih\\. (3.^)1 <IIV>il.9d).12) We omit here the verification of the remaining relations (3. (3.44 or CHAPTER 3. for which CO 2 := V^l^l 2 < 00. In verifying the triangle inequality we use this result (3. beginning with A = 1 in the second line of Eq. V 2 ) = 0 .11). Mathematical Foundations of Quantum Mechanics 1(^2.
3. i.[$]):= / JM /(x)* 5 (x)d 3 *. i. which satisfies relations (3. Eq. = 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.5b). (3.5c). all squareintegrable functions / which are "almost everywhere equal".5c)) / ( x ) = 0 = > ( / . (3.w) :=J2v*Wi. In order to avoid this difficulty. Then L2 is the space of all these equivalence classes. •'^ \ 0 otherwise. are combined to an equivalence class [/] (with space L2). But this applies also in the case of any function which is nonzero only on a set of measure zero. (3. With the scalar product (f. Elements of the classes are then called representatives of these classes. which differ solely on a set of measure zero. . */ x } = / /o for x = 0. etc. and one defines addition and multiplication by complex numbers with respect to these classes.g)= [ /(x)^(x)d3x. with inner product oo 45 (v. / ) = 0. 2 and so on. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. is defined by ([/].2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). for which the scalar product. (2) Let M.e.e. Js the space C? is not yet a metric vector space although for (cf.5a).
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.
that the vector \u) E V is associated with a vector in V. (3) (u\u) = 0. i. This expression is linear with respect to \u) and antilinear with respect to \v). which we write (u\.9a): \{u\v)\2 < (u\u)(v\v). (2) (u\u) > 0.e. 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.\u) = (u\0*.4b) (4. if \u) = 0 (see (3. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. The dimension of V is the same as the dimension of V. Dirac's Ket. or (4. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. V \u) (in the case of £) with infinite norm!). if (u\v) — 0. \v) E V are said to be orthogonal.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). Two ketvectors \u). the norm squared.3 Linear Operators. (4.5c)). if x\u) = 0 V kets \u) E V.and BraFormalism Furthermore we have: ( x  = 0. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV .5) 4.4a) \v) = J\(m<% and (4.62 CHAPTER 4.(t. It follows that we can now write the Schwarzian inequality (3.
or Kronecker. Furthermore. i. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). also acts on the dual space V. Let the following bravector be given: (x\ £ V.. on bravectors are similar to those on ketvectors..e. Finally we define the tensor. i. The inverse of a product is (AB)'1 =B1A~1. denned as a linear operator on V.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). In this way the operations of operators A. The inverse of A is written A . This product is commutative. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). the space Vi <S> V2. the space Vi <S> V2 has the dimension . Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on.4.. This does not always exist. ltt) = \u). Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. Then the following rules apply: A + B = B + A.3 Linear Operators. let (771 6 V be the bravector defined by this functional.or direct product of vector spaces. defines the unit or identity operator. if AB = 1. Two operators A and B are said to be inverse to each other. The product of such vectors is written: \u^u^) = \u)^\u)^. The product vectors l?^ 1 )^ 2 )) span a new vector space. if \v) — A\u) i. Multiplication by unity.e. and one writes: (V\ = (X\A.e. AB^BA. B. Furthermore linearity applies. Then one says: (771 results from the action of A on (x. 1 and \u) = B\v). as we can see as follows. If Vi has the dimension A/i. The operator A. BA = 1. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . It follows that (r.
i. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. also Sec. i.e. then \v) is a linear function of \u). Every such operator is then also associated with an operator. we obtain the conjugate relation <tAt«> = (u\A\t)* (4. here designated with the same symbol.3). AW\U)W = \V)W.64 CHAPTER 4. \t) G V. (v\t) = (u\A\t).7C) («)(u)t = \v){u\.e. i. the operator A^ is called hermitian conjugate or adjoint operator of A (cf.A^] = 0. Next we construct the following scalar products: (t\v) = (t\Ai\u).g. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. Let A^> be an operator in the space Vi and A^ space V2. Dirac's Ket.6) for all \u). we write \v) = A*\u). Since we demanded the scalar products to have the property (t\v) = (v\ty.7a) (4. [A^. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW).7d) .e. and (v\ = (u\A.e. 3. Every operator A^> commutes with every operator A(2\ i.7b) (4. in the product space. AW\uW)\uW) = \vWUW). If (v\ = (u\A. As a consequence we arrive at the following properties (e.and BraFormalism an operator in the Mi A/2. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.
Analogous considerations apply in the case of bravectors (u. it follows that (u\A\u) = a(u\u). An important theorem is the following. Verification: Since A — A^ and A\u) = a\u). The commutator of two hermitian operators is antihermitian: [H. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). but only if these commutators commute. [H.e. or (u\A\u)* — {u\A^\u) = (u\A\u). if H — H' and antihermitian. as is also (u\u). so that (n^4n) is real.K\. . if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. i. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. so that (a)(a)t = a)(a. hermitian The product of two hermitian operators is not necessarily hermitian.3 Linear Operators.K]^ = = [HKKH]* KHHK = = K^H^ [H.e. It follows that a is real. Obviously the quantity \a)(a\ is an hermitian operator. 65 The linear operator H is called hermitian. The operator H is said to be positive definite.4.K}. An hermitian operator A has the properties (a) that all its eigenvalues are real. K] = 0. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. for we had (\u){v\)i = \v){u\. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . i.
which includes vectors of infinite norm. Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. and let S* be its complement.e. If these form a complete system.9a) . Let S be a subspace of the Hilbert space !K. a ^ b. i. i. since a is real. but the continuum vectors are normalized to a delta function. Ps is idempotent. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). {us\us*) = 0.Au) = a(v\u) — b(v\u) — (a — b)(v\u). i. The entire set of ketvectors spans the extended Hilbert space.u'). i.9b) (4. Ps\ua. ( n  z / ) = 0 . so that (u\Ps = (us\.8) The projection operator has the following important properties: Ps is hermitian.e. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). Next we introduce the very useful concept of projection operators. Dirac's Ket. Assuming A\u)=a\u). (u\Ps = {us\. as we can see as follows. (u\u') = 5{v .66 CHAPTER 4. Ps2 = Ps(4. The above theorem remains unchanged. (4.e. A continuous spectrum can immediately be included by passing to the extended Hilbert space. \u8*) E S*. (n\n')=5nnl. since a ^ b. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.e. (v\A = b(v\. we have 0 = (uArt) — (u.) = 0. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. then the hermitian operator defined on this space is an observable.and BraFormalism Moreover. so that (v\u) = 0.
(1 — Ps) is projector onto S*.. 2). Then 0 = (P2P)\p) and hence p = 0. Very similarly we can construct operators which project onto the subspace spanned by the continuum states. (4. i. The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw)..9c) This property can be seen as follows.p = 0. c = (a\u). Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. The quantity \a)(a\ is called elementary projector. i. . Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. = (p2p)\p). Let a continuum state be written £).e. Set \u) = \ua) + \ua*) with (a\ua*) = 0.e. where £ is a continuous index. iV) is a set of orthonormalized states. i. if the set of vectors 1). Let p be an eigenvalue of the projection operator P.e.4. P\p) =p\p).e. Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. so that (Ps2 ~ Ps)\u) = 0. p 2 .3 Linear Operators. Ps is projector onto S. so that \ua) = (a\u)\a) = \a)(a\u). 67 (4.1. i. as can be seen as follows. \ua) = c\a). (1 — P)\u) are orthogonal.. . Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace.9d) Thus the vectors P\u). \u) arbitrary.
.e. u2). (11. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. (11.r)(ui.6.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator).. However. in this case the spectrum also has a continuous part. 8. 4.68 CHAPTER 4. one says the degree of degeneracy is r . that one and the same eigenvalue Aj is associated with several eigenfunctions.1 and 11.114c). i. Coulomb potential. see Eq.g. Examples 8. If there are r such vectors. Observables are representatives of measurable quantities. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors.6. See e. which are orthogonal to each other (i. n=l In the case of continuum states (which are no longer countable) one has correspondingly as projector of all states in a domain £ e [£2>£i] or in the case of the differential domain d£ of £: dp = j '£ de'ion This latter operator is called differential projection operator. be a system of basis vectors in this space.r\ "An example is provided by the case of the hydrogen atom. Degeneracy will be discussed at various points in this text. Dime's Ket.. the operator P is correspondingly P2\u)= f2\0m\u). Sec. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.4 Observables Operators which play a particular role in quantum mechanics are those called observables. Let \u\).. are linearly independent).e. which we introduced earlier. and Eq. P = £n)(n. In general it is possible.114b).
r'} = Sii>6rr>. is always implied).e. PA = y£pi i = Yt\ui.r =^ i. then PiPi' = 0. i i (4.10) expresses the completeness of the orthonormal system.10) gives the linear combination it) = ^ i.e.r\ = l. (4.r)(ui.r \(ui.10).4.e.r (4. 69 i If A is an observable and if the spectrum is purely discrete. i. and the eigenvectors \ui.r) (the convergence. and APi = XiPi. which we do not enter into here. we obtain i i i i i. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.e. The uniqueness of the expression (4.r\u) i. or also as subdivision of unity or of the unit operator. i. the projection of this operator is the entire space. (4. the relation (4.r. the operator A is completely determined by specification of the eigenvalues A.r)(ui.12) . Together with the orthogonality condition.\i)Pi = 0.r \ui. in the case of degeneracy with (ui. The operators Pi are linearly independent. ^ A i P i = A = ^Aiui. Let us set {A .10) follows from the fact that i Applying the operator A to the projector (4. Applied to an arbitrary vector \u) G "K Eq.11) i.4 Observables The dimension of this subspace is that of the degree of degeneracy.r\ui/.r)(ui.r\u)\2. i.r>(ui.r\u).10) This expression is known as completeness relation or closure relation. If Aj ^ Aj'.
17) previously. now however. as explained earlier. Let v be the continuous parameter which characterizes the continuum. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. written in Dirac's notation. In a similar way we can handle functions f(A) of an observable A. for instance. (4.e. (3. the eigenvector \ui): f(A)\ui) = f{\i)\ui). (4.70 CHAPTER 4. for instance exponential functions. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). If the spectrum of an observable A consists of discrete as well as continuous parts. Dime's Ket. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv).14) . (u„zv) = 8{v .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). we have the corresponding generalizations. i. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. i.and BraFormalism This is the expression called Parseval equation which we encountered with Eq. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. One defines as action of f(A) on.e.v'). The ketvectors \uv) are orthonormalized to a delta function. 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.
58) in terms of ket. 4..15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx. \k)€Fk.15) and (4. with energy E.59). For many practical purposes the use of the Fourier transform is unavoidable.B.C.5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. if and only if they possess at least one common system of basis vectors. for which the completeness relation is f dk\k)(k\=l..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. and (b) if they possess a uniquely determined system of basis vectors. Eq. as a formal orthonormality condition.17) The Fourier transform provides the transition from one representation and basis to the other.and bravectors. B]. First of all we can rewrite the integral representation of the delta function.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. / dx\x){x\ = 1. if (a) they all commute pairwise. again later. their commutator vanishes.e.16) represent subdi . In the following we consider more generally ketvectors \ip) as representatives of the physical states. i. Since both expressions (4.. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. form a complete set of commuting observables.4. (4. Extending this we can say: A sequence of observables A. where the symbol ip is already indicative of the Schrodinger wave function ij. (4.. In the onedimensional case we write {x\x') = S(xx'). which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). (4. We shall return to this theorem. (3. x) € Fx. here presented without proof.e. Therefore we want to reexpress the Fourier transform (3. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. [A. i. which all commute with one another.
Fk. (x\k) = ^=eikx. i.15): 5(x .e. Dime's Ket. . Rather \x) and \k) serve as basis vectors in the representation spaces Fx.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).59) or Eq. called wave function. (4. V 27T (k\x') = . ${k) := (k\if>).e.21) provides the ketvector \ij)) in the A. (4. Obviously we obtain this by inserting a complete system of basis vectors of Fk. (4. (3. 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).18) According to Eq.19) Comparison with the orthonormalized system of trigonometric functions (4.e. il>{x) := (x\i/}) : "K * C. <a#) = [ dk{x\k)(k\ip). The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. i.space representation.L e " * * ' = {x'\k)\ V27T (4.and BraFormalism visions of the unit operator. (4. we can rewrite Eq. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. this expression has to be identified with — / dkeikxeikx\ i.22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.2a) etc. (4. which are representatives of the states of our physical system.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').1a).72 CHAPTER 4.
We mentioned earlier (in Sec.p.e.2 The Density Matrix We shall establish a matrix p. which is the quantum mechanical analogue of the classical probability density p(q. the calculation of which for specific cases is the subject of Chapter 7. 5.Chapter 5 Schrodinger Equation and Liouville Equation 5. with the complementary system or *E.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. Schrodinger [244]. i. if we do not take into account the complementary part of the universe. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time).t). 73 . This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. Thus we distinguish between socalled pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator. 1.1 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.
b. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a.i..e.. which in Dirac's notation we can write as = J]C a »i) a.74 CHAPTER 5. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. T . we have to deal with a socalled "mixed state". are called "pure states". a (5. i.. of an operator that acts only on the states of the system we are interested in.i. (52) (5. p" ( C C * ) t = CTC* = p. where i represents collectively all quantum numbers of the state of the system under consideration. where T means transpose: p = CTC*.b..3a) and N J  a ) ( o  = l in the complementary subspace. i. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. (a\i)=0 = {j\b). \a).j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. \a). as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij.. The states \i) £ "K.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. The actual and real state of a system is then a superposition of two types of states..^ Since p is hermitian. that is. \i).e. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows.. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system.3b) The quantum mechanical expectation value of an observable A. but also of the states of the complementary set of the universe. These are the quantum mechanical states which are also referred to as microstates. we have (a\b) = 5ab. Schrodinger Equation and Liouville Equation the measuring instruments. For an exact treatment we would have to express the actual state \ip) of the system as a supersposition not only of the states \i). called the density matrix.
as we discussed previously. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).e. in the position state representation or xrepresentation.3 with Pi^YtCaiCaj*.) = YJc t . can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i.x) := (x'\p\x) = yju.6) (i'\p\i") = Wi'Si'vi = real. (5.. a complete orthonormal system with properties (5. in a specific representation.4) we had the expression (A) = Y/Pij(j\A\i) i. Thus the real system is in the pure state \i) with probability uii/ J2i Ui. which today is a fundamental term in statistical mechanics.\p\i")(i"\ = Y. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.2 The Density Matrix 75 \i') 6 "K. Gibbs introduced the concept of ensemble.5. i.* We can specify the state of our system. i i (511) In Eq..ji(x / )i*(x).3a).9) is diagonal.e. i. This transition to a new set of basis vectors in which the matrix becomes diagonal.7) p = ^2\ii)(i. seen macroscopically of Z copies of the system). This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities.e. . (5.10) and is therefore. quantum states. (see below) eM. In the following we write simply \i) instead of \i'). (5.2) and (5. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann. e.e. but occupy in general different microscopic. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i).^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j.. or (5. where x represents collectively the entire set of position coordinates of the (particle) system. This ensemble is interpreted as a large set of identical systems. The ensemble can be represented as a set of points in phase space. P ^ ) = i)(#) = (#)i) (5.g. which are all subjected to the same macroscopic boundary and subsidiary conditions.j(x'i)(ia. a projection operator. If the ensemble corresponding to the system under consideration consists of Z elements (i. as we saw earlier. In this sense we have p(x'.
4). We can also show that Ui > 0. (A. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) . Then the expectation value of the observable A is (cf.18). In particular for A = 1. (5. as claimed by Eq. Eq.17) Hence the sum of the classical probabilities is 1. i t (5. 3 (5. (5. (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.16) (5. according to Eq.5) above) (i\p\j)=Pij.18) Substituting this into Eq. (5. On the other hand. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). (5.76 CHAPTER 5.4)) (5. (5. Thus Tr(j>ii)<i)=l.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j).14) which we can write (A) = Tr(pA). Hence Wj/ > 0. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf.) := (mm = (#') W> = l(#')2 > 0. we have Tr(p) = l.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. (5. Eq.3 = pvv = uv.
that — different from u>i — P. i. and in the other states with probability zero.t). The effect of the interaction of the system under consideration with the surroundings.21) p=\i){i\=Pi. (5.t) = \^{x. the only remaining one is 1.5. that these are actually not sufficiently general. not in a mixture of states.22) In the following (x\i) is always to be understood as (x\i(t)}. but we see here. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2.t) and instead of (x\Pi\x) we now write p(x. In the following we are mostly concerned with considerations of systems in pure states. the system is in a pure state. i. Equation (5.20) expresses that the system can be found with probability 1 in the state \i). (Question: Is the state of the universe a pure state?) 5. thus enters these states. (5. defines the statistical operator. represents a quantum mechanical probability.e.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. We see this more clearly by going to the position space representation.t) = {x\i). by representing the vector \i) by the wave function i(x.3 The Probability Density p(x. so that p{x.t) we now write tp(x.x').t)\2. which is contained in the coefficients Cai.23) Instead of i(x. ^2ui\i){i\.24) s (x\x') = 5{x . (5.8). If all ui but one are nought.e. since they do not take into account the (interaction with the) rest of the universe. t) p= i The expression (5.3 The Probability Density 77 We see therefore that p defines the mixed states. Thus we have a system in a pure state. We now consider the latter expression. one says. We observed above.e. in all other cases the state of the system is a mixed state. i. For the case n we have TJ=J (520) (5. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. .
t)\2. the expression (5.t) = / dx{i\x)(x\i) = (i\i) = 1.25) J Thus the particle whose state is described by \i).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. This differentiation requires the time derivative of a state vector \i) E "K.26) 5. the Schrodinger equation is always postulated in some way. but the linear operator states are moving — even if the system is observably stationary).22) can also be written (xi) 2 = (i\x)(x\i).e. Schrodinger Equation and Liouville Equation The relation (5.^ The equation which is the conjugate of Eq. 2. To this end we have to differentiate the density matrix p. (5. i.28) Differentiating Eq.t) = \rp(x. (5.8) and multiplying by ih. and hence the integral over all space dxp(x. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j). 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.27) where H is the Hamilton operator of the system. or the wave function tp(x. (5.3.27) is known as the Schrodinger equation.e. (5. The generalization to a threedimensional position space is selfevident: JK 1 / p(x. We therefore postulate first an equation for the time dependence of a state vector. it is not "derived"). (5. can be found with probability 1 in the space R1. t). . which is chosen in such a way that the desired analogy is obtained. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. (5. We shall convince ourselves later that the postulated equation is sensible (in fact. We now want to obtain the quantum mechanical analogue.8). Equation (5. with respect to time. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).27) is iHJ\ jt= m .78 CHAPTER 5.
31) The eigenvalues u>i of p determine the fraction of the total number of systems of the ensemble describing the actual system which occupy the state \i).32) Comparing Eq. (5.e. i.27) and (5. (5. Eq. (3.27). (5.5.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. Eq. With Eqs. We can now consider the Schrodinger equation (5.e. we observe that here on the left hand side we have the total time derivative. We thus obtain the operator form of the Liouville equation.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.28) the states are considered as timedependent. (5. p also contains "quantum mechanical probabilities" (i. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. dp/dt ^ 0.p}.28) we obtain (see also Eq. The reason for this is that in addition to the "classical probabilities".e. the relation ih^ = [H. i. (5.37)) (5. such a formulation is described as Schrodinger picture.21). not the partial derivative.Xi)\j). The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf.34) .p} + ihJ2^\J)(J\. (2.32) with its classical counterpart (2. (cf. (5. both uji and i)(i). M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.24)).
x 1 ^ ( x .0): where H(ih—.. T meaning temperature. with respect to time t. appearing in the Boltzmann distribution.t) = Hld . i.e. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H.and temperaturedependent Green's functions. operator" the exponentiated oper .Xi){x\En) = En(x\En). it is plausible to refer to this equation at this stage.36) in in The initial wave function or timeindependent wave function ^(x. a (5. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not.** In view of the close connection between time. With the help of Eq. n (537) This equation is described as the timeindependent Schrodinger equation. .. (5. P. In expression (5. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. in p = ^2i^i\i}(i\. since the solution of Eq.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.t) = exp ip(x. to) = exp[—iH(t — to)/H\. in Sec. t ) .4.36) however. Feynman [94].8) for p. but with respect to the parameter /3 = 1/kT.. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed. 0) or \il))t = \il>)t=o exp (5. however.t): > ih—</>(x. **R.35) can be written'! ip(x. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later.4. we shall describe as "time development ator given by U(t. 10. (5.38) (x\j)t=o = Yt(x\E^(En\J)t=o. the state \i) is still timedependent.e. „ / ith—. i. or H\En) = En\En). we can replace these states by corresponding timeindependent states.35) Here we can look at the Hamilton operator as a "time generator3''. (5. we write V>(x.80 CHAPTER 5. which we shall need later. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1.
(543) Since H(f>i(x) = Ei<f>i(x).^ n tne position space representation Eq.x'(3) = Si e ~^W*V).9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT.40). (5.4 Schrodinger Equation and Liouville Equation 81 Example 5.44) r (TrePH)' (5. (5.e. (1. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. we can rewrite p.5. Eq.1).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.x')^p(x. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. we obtain p(x.42) Inserting here Eq.40) 1.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). i ( 5 .45) since on the one hand •PEi i i i and on the other hand . (5. i (5. or ^ = ^ e 0Ei _m.41 ) or with 4>i(x) := (x\Ei) this is p(x. (5. x') = ^ i LUiMxWix').
49) In the position or configuration space representation Eq. (5. Example 5. (5.(3). We now rewrite the factor exp(—(3H) in Eq.45) without normalization as PN(J3) := e~pH.48) = 1. (5.X'. [A.(3) := (x\pN(P)\x') = (x\e~0H\x'). B}} . This solution is the Green's function or kernel K(x. whose calculation for specific cases is a topic of Chapter 7 (see Eq. B] + L [A.45). [B. Wilcox [284].47) (5.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5.L [B.52) Hence this expectation value can now be evaluated with a knowledge of pN. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5. Differentiating Eq. (3). (5. (7.51) is seen to be very similar to the Schrodinger equation (5.x'.51) where the subscript x indicates that H acts on x of PN(X. Equation (5. With Eqs.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.x'. Tr(pN) = / dxpN(x.4> = ^ g ^ . (5.tt tt For a more extensive discussion see R.46) is pN(x.x'.47) with respect to (3. M.45) and (5. (5./3). (5. Schrodinger Equation and Liouville Equation so that p is given by Eq.51). A}} + • Solution: The relation can be verified by expansion of the left hand side. we obtain = HxPN{x.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij./?). the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.35).50) Differentiating this equation with respect to j3. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).46) we can write the expectation value of an observable <^Tr<„. (5. .55)). (5.x. (5.82 CHAPTER 5.
Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. * Comment of Y.. that the oscillator. can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory.the present form of quantum mechanics is largely a physics of harmonic oscillators.51) to (8.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. Noz [149]. This is a fundamental topic since the harmonic oscillator. 83 . 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. like the Coulomb potential. can — in fact — be quantized in quite a few different ways. 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:* ".53))." Quite apart from this basic significance of the harmonic oscillator. 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). However. Since the spectrum of the harmonic oscillator is entirely discrete. we shall see on the way. E. S. its quantization here in terms of quasiparticle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles. (8. Kim and M.
h = l. which is called Weyl ordering. This can be done in a number of ways. in the present case + [EmooJ2x2)ip = 0.84 CHAPTER 6.2) which for mo = 1/2. we can also proceed in a different way. When such a term arises one has to resort to some definition. so that there is no ambiguity due to commutation.2 The OneDimensional Linear Oscillator In the case of the onedimensional. Quantum Mechanics of the Harmonic Oscillator 6. of which the only nontrivial one here is [q. 2mo dx \ 2 2 (6. p^ih—. However.^ The first problem is to determine the eigenvalues and eigenfunctions of H. To this end we introduce first of all the dimensionless quantities rriQUi fl q. hence we do not use a distinguishing notation.to = 1 reduces to the parabolic cylinder equation.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers.p]=ih (6. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. This equation would then have to be solved as a second order differential equation. Also observe that H does not contain terms like pq or qp.p. For instance we can go to the special configuration or position space representation given by q^x. like taking half and half.' ^mgLOh by setting and *~. which is — in the first place — representationindependent. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions.1) for the hermitian operators q. . linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 .
e.B}C + B[A. (6. We now use the relation (3.6.2 The OneDimensional Linear Oscillator 85 (p.14b) [A* A. for ^ A  a ) = a\a). (6.13) ( ^ A ) i t = A\A^A From Eq.14a) (6. A^] = A\ (6. (6. .8) The eigenstates of H are therefore essentially those of N := ^ f A (6. (6.6) and p = y/motoh AAi =^.e.1 ) . C] = A[B. With the help of Eq. (6. A*} = A^A. (6. C] + [A.14a) we deduce for an eigenvector \a) of A^A. A] = [A\ A]A = A. i. (6.1) we obtain immediately the commutation relation [A.BC] = [A. if A*A\a) = a\a).9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. A\ we obtain h A + A^ rriQLU y^ (6. q are hermitian). i.34d): [A. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .A^] = 1. Prom these relations we obtain ( ^ = ^ .11) Thus the eigenvalues are real and nonnegative.10) then a = (a\A^A\a) = \\A\a)\\2>0. and [AB.12) in order to obtain the following expressions: [tfA. (6.5) Reexpressing q and p in terms of A.C].7) (6. C}B. + 1).
86 the relation CHAPTER 6. unless A^\a) = 0.18) = '= a ) A(A^A .e.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a).11) this is not possible.17) (6. unless A\a) = 0.19) i.A)\a) A{AAU .21) .l)\a) = A(a . In this case we have (A*A)A2\a) (6 (6. If we continue like this and consider the vectors An\a) ^ 0 for all n.2)A a).l)A\a).16) This means. unless A 2 a) = 0. However. (6.  a) is eigenvector of A^A with eigenvalue (a + 1). Similarly we obtain from Eq.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1). The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).l)\a) = (a .2A)\a) 2 (a . in view of Eq.. Thus for a certain value of n > 0. (6. A 2 a) is eigenvector of A^A with eigenvalue (a — 2). This would mean that for sufficiently large values of n the eigenvalue would be negative. or Ata) = v ^ T T . (6. we must have (a) Let \an):= Ana)^0.l)A\a) = A(A^AA . (6. (6. Next we consider the vector A 2 a).2A)\a) = A(Aa . since this equation implies that the eigenvalues cannot be negative. (6. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1.20) . but (b) An\n>\ llA An+1a)=0. or \\A\a)\\ = Va. (6.
or a = n > 0. The state 0) is called ground state or vacuum state.e. (6.17) a by (a — n). I „ . i. (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a ..22) With relation (b) of Eq (6. (6. v. \ 9 \ I / II 4 T ) . Moreover.27) According to (b) of (6. (6. \ I 2 2 x 87 Replacing in Eq.n ) = " A»a>=» ' = A«a)F = 1 . For a — n = 0 we deduce from (b) of (6.20) we have for a = n = 1: A2\1) = 0. In the following manipulations we always use the commutator (6.25) (6.20) that A0) = 0.I n U . (6.5) to shift operators A to the right which then acting on 0) give zero.' " /.28) .20) we obtain from this that the right hand side vanishes.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.6.24) This is a very important relation which we can use as definition of the state vector 0).18) we obtain for a = n: p t  n )   2 = n + l..23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.n  a .2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n). (6. 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. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. that an = 0.
i.29). According to Eq. In general we have \n) = ^(A^n\0) (6. as we can see as follows. and in general [A. But using Eqs.32) (6.88 and CHAPTER 6. we have 2> = ^=dAi\0). (A*)2} = [A. A^A* + A*[A.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). (6.5) and again (6. and in view of Eq. (0^ 2 (At) 2 0) = 2.11).11) we have [A. l>ocAt0).(A*)n] = n(Ai)n\ (6. A^} = 2A\ and [A. (6.^ ( O l A ^ n o ) . (6.31) (6. (A^)3] = [A.(6. (6. 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) ^ 0.26) the equality l) = A t 0). (At)2]A+ + (A*)2[A.32) we have (nm) = . Similarly we find 2)oc^Ut0).34) (6.29) Hence we obtain and in view of Eq.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.e.33) . The states \n) thus defined are orthonormal. (6.
•(» (6.40) (6. The terminology is.34) A\n) = ±=A{Al)n\0) . however.6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.33).37) n^)711}^) = v n  n .J \n) = hw (n + i J \n).34) and (6. l . in quantum mechanics.39) A^A\n) = y/nA^\n . (6. (6.1) = n\n). ldnm = n\5nm. i. . .35) Inserting this result into Eq. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv. In view of the properties (6. 2 .1) (6. . . (6.41) The contribution hu/2 is called zero point energy. .'n\ and = v n ![ ( A t ) M i= + (6. Here. whose number is given by the integer eigenvalue of the number operator N = A^A. chosen in close analogy to the postulated "second quantization relations" of field theory. we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle").1 (A + ) m . we obtain (n\m) = 5nm We also deduce from Eq.8) we obtain therefore H\n) = hu (A^A + ]. 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 .1 ) .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 .35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. . With Eq. (6.e.38) (6. .32): rf\n) = . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . n = 0 .1 0). (6.36) + 1) (6.
43) = = y/n(n'\n — 1) = y/n5n^ni. n = 0. other elements zero.90 CHAPTER 6.3 The Energy Representation of the Oscillator The energy representation. (6..2. Eq. and — in fact — we have used this already previously (see e. ( 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 / (6.7) the energy representation of the operators q and p. other elements zero. (5. Vn + 1.1..39): {n\A.39)).. i. is defined by projection of operators on eigenfunctions of energy.e.45) Correspondingly we obtain with Eqs. (6.'\n) i'\A*\n) (n + l\A'\n) and similarly (n'An) (n — l\A\n) In matrix form this means 0 0 0 0 0 0 0 0 0 A/4 = = Vn + l(n'\n + 1) = \/n + 15 n / iTl+1 . also called Heisenberg representation.46) v^ .. (6. (6.42) We can deduce the energy representation of the operators A. Quantum Mechanics of the Harmonic Oscillator 6.6) and (6.38) and (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. En = hu(n + ).44) yfn. A^ from the relations (6.g. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn..
(6. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.32). (6. A\Q) = 0. 2 . mow / (6.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs.e.4 The Configuration Space Representation We saw that the eigenstates are given by Eq. 6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .47) LP 0) = 0 .1) and (6. Eq. i.5) are also satisfied as matrix equations.3)) ^ (q 2n \ + (6.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). 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. by (cf. (6.6. (6. The ground state wave function (/>o(x) is defined by Eq.24).
l ) n f f „ ( 0 . ) e"^2. (6. Vn! Now {XlA (6. c/>n(x) := (x\n) = . (6.e.54) (6 55) i/4 rzri7r 1 / 4 vo. i. In order to obtain the wave functions of higher states.0 = ( . so that <t>n(x) = ( .l ) > n ( x ) .2 n n! The functions Hn(£) are obviously polynomials of the nth degree. (6.92 CHAPTER 6.dx'{X] so that <*'>^(irrv^i« mQU dx Setting a we have 7T 1 /4 v ^!2«/2 \a Setting a and ff„(0 we have = ^ / 2 u^n^m^''^ h mouj dx) K ' ( f .L ( x  ( A t ) »  0 ) . it suffices according to Eq.51) \l 2h {X q \ m0uP)\l 2h \X m0u. they are called Hermite polynomials.32) to apply the appropriate number of creation operators A^ to the vacuum state 0).56) < ^ = We have as an operator identity the relation . Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. These polynomials are real for real arguments and have the symmetry property # n ( .
93 . (6. p. ( 2 0 4 . d£" (6. (2036(20.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .en_ dt. .g. Oberhettinger [181].58a) ' T h i s definition — apart from the usual one — is also cited e.2)m = (±t)m for some function VKO. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.58b) by W.4 The Position Space Representation i.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.12(20 . (202"2. (6.6.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1. e.57b) • Generalizing this result and inserting it into Eq. 81. 2£.12.e. Magnus and F.
6 £ 2 + 3. Magnus and F.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . (6. = nHe^iO. .59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.^ + n) He n(0 = 0 (6. (6. (6. £33£.He'n{£). = ffen+itO+ntfeniCO.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . (6. £. £2l.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. 80.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.(e^/2) or Hn(0 = 2n/2Hen(V2ti). £ 4 .59b) = (l)^2/2^.60b) J—< "See for instance W. pp. Oberhettinger [181]. 81. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1.94 CHAPTER 6.58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists. (6.
g. .s)=F(£.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.s) = J2 n=0 (6.4 The Position Space Representation 95 The relation between Hermite polynomials and parabolic cylinder functions Dn(0 — which we use frequently here — is given by" Dn(0 = = Dn(V2£) = ( . p. and so on.63) The meaning is that if we expand F(£. W.s) = eM?(Zs)2}(6.61) e?l*2nl2Hn{Z). F(S. s) as a Taylor series.6. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.l ) ^ e^HeniO 2 / 4  .64) n\ See e.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.. 93. Thus the generating function can be written F(Z. Oberhettinger [181].. Their normalization is therefore given by OO /"OO / OO J —OO z (6. (6. Magnus and F.^ = e^/42~n/2Hn(aV2).
for x(t) = 0 and p(t) = mox(t) — 0.96 CHAPTER 6. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). 2 . i. and we see t h a t even and odd states alternate.e. We also see from the eigenvalue t h a t the energy of the lowest (i.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. In fact we can use the uncertainty relation AxApx ~ H/2.e. ground) state is fko/2. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig.1. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . mow' Apx = ^m0Luh/2. In classical mechanics the energy of the oscillator can be zero. 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. 6. We see t h a t the lowest state is symmetric. 6. i. Quantum Mechanics of the Harmonic Oscillator Fig. has parity + 1 . But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero.e.
: 27 T mov2a2.1. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). 14. to 0. setting q = 1/z. 27rmoi/ <f> V'a 2 — q2dq = n*h. /classical orbit Here q goes from 0 to a. like twoatomic molecules. E . mass mo. n*h.plane pole at z=0 ^ . Example 6. back to 0. pdq. 6. which is the complete orbit. we obtain the integral .4) to obtain by contour integration the eigenvalues of the harmonic oscillator.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity".1/a O l/a Rez all at infinity Fig. in corrected version instead with n* = n + 1/2. Thus. But it has a pole at infinity. and is directly observable.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. Sec. The analogous Example 6. The kinetic energy is always positive or zero. According to "old quantum theory" with integer n..4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. to —a.6. 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)). In Example 6. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.2 The original path and the transformed path at infinity. original path q . This observability can be taken as direct proof of the uncertainty relation. classical orbit i.plane Imz z .e. In the plane of complex q the integral has no poles for finite values of q.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf.
(11. z = 0 Vi 2Kmov(Ka2) = n*h = E/v..1)! (6. in q at infinity). . . 6. are the most basic and exactly solvable cases in quantum mechanics. harmonic oscillator and Coulomb interaction. since these two cases. i. 2 .It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 .98 CHAPTER 6.e.e.1.1 D e r i v a t i o n of t h e g e n e r a t i n g function From the above we know t h a t the energy is quantized. but here we leave this .5 . 1 . for instance — without deriving all properties of the solutions.a) m = 2 .e. Example 6. hence we consider the following differential equation in which n = 0 . £ = n*/w = ( n + . « qdq + I2 = f dq . (m . After removal of a Gaussian or W K B exponential. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . . mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .2TTI a ^ = .2: Contour integration along a classical orbit Setting q = 1/z. Eq. / " ^ . With a2 = E.2b = Ze2.) hv.114a)).( 6 .c2 = (I + 1/2) 2 (natural units.V62 a2).5. i. 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). we start with a Laplace transform ansatz for the remaining part of the solution. E = Z2e4/4(l + n+ l ) 2 (cf. and obtain with this all properties of the solution. In the following we investigate in more detail the important differential equation of the harmonic oscillator.65) f ^Va2z2 . a J \d Solution: The solution proceeds as in Example 6. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11.3. 6. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.1.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials.
Eq. Oberhettinger [181]. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. (6.69) **See e.66) First we have to remove the £2term.e. Substituting these expressions into Eq.g. (11.68) § = J p2s(p)e*dp. fp{ps{p)) = {v2 + n)s{p).6. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). so that the integrand of the remaining integral vanishes. / dpf(p)e~px the Laplace transform of f(p). We demand that C(p) := — \ps(p)e~pt] = 0. This has to be true for all values of t. p.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. Magnus and F. . E. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .p t dp = \ps(p)ePt} . Then (with partial integration) *dt = ~*/Ps(p)e. 80. (cf.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. (6.111)) the Mellin transform of e~p is n!.g. i.t*ij& + ny{t) = O. (this determines the limits). (6.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . . (6. and / dpf(p)pn the Mellin transform of / ( p ) . W.i ^  (ps(p)) P n P Integrating this expression.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.J ±(ps(p))e*dp.
then for large t we have y(t) ~ exp(i 2 /2). consider the exponential in the integrand. Therefore we take ptp2/2 y(t) .70) = 0 P. This can be attained if p is allowed to be large (t can be of either sign). To see this. (6. so that p cannot go to infinity. = / • Im p Rep n not an integer Fig. For singlevaluedness. Hence a possible path of integration > is from — oo to oo.n+l between the two limits of integration. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. so that y(t) ~ exp(i 2 /2). This exponential assumes a maximum value when pt \—p2 = minimal. (2) If n < 0.3. 6. Eq. n must then be an integer.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible.e. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. we can choose a path from zero to infinity (the condition will vanish at p = 0). We can show that if \p\ is allowed to be large. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. 6. We now investigate possibilities to satisfy the boundary condition C(p) = 0.71) .3 The path when n is not an integer.. as indicated in Fig. of course). when p = —t (which is possible. The value of the exponential at this point is ~ exp(£ 2 /2).100 and hence CHAPTER 6. The quantum mechanical wave function is then (cf. (1) This will be satisfied if \p\ — ±oo. (3) We can choose a contour once around p = 0 in the complex pplane. i.
this becomes jfj.. .5.6. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).„y2 v l { ' ^ r (^) \ 2 ' where v = 0. 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) = ( . n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . pt p2/2 (6. In y(t) above.. .l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.P 2 / 2 00 C—tY ( . . the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt).if n is odd..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.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e .3. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. We have 27T! Setting q — p + t. 2 .5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. 4 . . (676) . We can also obtain the derivative form of this Hermite polynomial. if n is even and v = 1.
8. (n.n + l ) f f e n ^ i ( £ ) + (n.7 that we use throughout this text) and their orthogonality and normalization to Examples 6. ( n . as in the normalization of asymptotic expansions of Mathieu functions).102 CHAPTER 6. Example 6. and elsewhere. (6. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.n— 1) = n. (6.59a).1 '2dt using Hen(t) = (l)n27 r . Hen(t). Example 6.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO .3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. n + l) = l. Pptp 2 / 00 /2 dp. one can construct the tower of polynomials Example 6. we obtain tHe„{t) (l)"n! ~ i. for the evaluation of expectation values. 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). n — l)Heni(t).75) as Hermite polynomials.70). (6. 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+<.3 and 6. in radiation problems. 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.e.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. Reinterpreting the remaining integrals with Eq.4. Solution: We have to evaluate °° 2 .it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t).77) = t.g. Starting from Heo(t) = 1. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. Hei(t) Henj.
6.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).82) ^ . g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. and 0 otherwise \/2n(—l)n+mn! if n = TO.5 nm .5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy. (6. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.66) is \Zn\V2w Example 6.jQ) n n!I ^—' ^Q fi\(n .^ + 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 .q) = = e . so that /2 Q2(n+rrf in / ( p . It follows therefore that the following integral (with an exponential exp(—at) from whose expansion we pick later the power of i) is given by °° 2 / where OO eatHen(t)Hen+2r(t)et l2dt = n\(n + 2r)\ X coefficient of pnqn+2r in f(p./(n __ 2r — /i)!.q).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.78) \/27rn!. zero otherwise (6.80) f(p. g) is n! Next. Hence the normalized wave function of Eq.n)\ i s an+2rij. (6.
the coefficient of T 2s »{sr u)\ As we observed. 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) + . i.(t)e«2/2dt Now. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ .u)\ 2s<—" v\(n . the relation (n.104 CHAPTER 6. n + 2) + (n.v)\ ^ Multiplying both sides by the value of the normalization integral. n + l ) ( n + 1. n ) ( n .n + 2) + (n. n.n+ l)(n + 1. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). n + 2)(n + 2. n + 2)(n + 2.( n + Jr). so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. i / ! ( n . n + 2)]ife„ + 2 (t) + ••• .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .l . n + l)(n + l .e.i / ) ! ( 2 r + i/)! > .A i ) ! (6. n + 2) + ( n ..— — ^ .83) = (n + 2r)!e a / 2 V „ „. .e '' ^ o^M ! ( n .e. n + l ) ( n + 1. .v)\(s . n)(n. As an example we obtain for the integral from 0 to oo (also reexpressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e . n + l ) ( n + 1.e.r .^ ) ! ( n + 2 r .r . It follows that the coefficient of a 2 s / ( 2 s ) ! in this expression J is the quantity K given by f^ot2sge„(t)Jfen+2t. n + 4)_ffe n +4(t) + [(ra. The result then follows with half the normalization and orthogonality integrals (6. n + l ) ( n + 1. n .Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . Prom Example 6.77).^ / 2 * . n + 3)(n + 3.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.l ) ( n .v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .79). n — 1) are given by the first index." ) a2^) r coefficient of (2s)! in f^ (s .2s _oo ( a 2 / 2 ) i .A o?(. we obtain the result (6.3 we know that upgoing coefficients are 1 and downgoing coefficients (n.' .78). i. n + 2){n + 2. by n!(7r/2) 1 / 2 . i.r+v) a2s coefficient of in (n + 2r)\ V (2s)! h .ra + 3)(n + 3. K is the coefficient of a / ( 2 s ) ! in the expression J . n + l)(n + 1. 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. /i = (n — v) _ / r o o e .° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e .
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. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. We 105 . We consider first timeindependent Green's functions. With reference to this case. in Feynman's path integral as the kernel or free particle propagator.Chapter 7 Green's Functions 7. which we write G{x. and those which are timeindependent. and is thus an important point to be noted here.x'. of a differential equation of second order. which is a point of unstable equilibrium for a classical particle. 7. in Chapter 21 (with a different calculation). Let H^ = E^> (7. These are the Green's functions of the timeindependent Schrodinger equation or. 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.1) be the timeindependent Schrodinger equation which has to be solved. more generally.t).2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms.x'). which is normally not discussed. which we write here K(x. The first of these cases will reappear later.
Green's Functions H = H0 + Hi. i.EW)xG{x. We can obtain an expression for G(x. for instance.g. a perturbation part like (3q4. (7.x'). 'The case of £(°) equal to some Ei is considered later.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. The timeindependent Green's function G(x. Eq.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). although the explicit calculations can be much more involved.x') by recalling the completeness relation. where £ ( 0 ) = lim E. e. (7.x') can be written t G (7.g.g. (7.x') is defined by the equation* (Ho . since the generalization to higher dimensions is selfevident.13).106 assume a Hamiltonian of the form CHAPTER 7. x') has to satisfy correspond to those of * (to insure e. of the kinetic part p2/2mo and some part of the potential. e. square integrability. cf.e.x') or ^2^(x)^*(x') i = S(x . the relation X') = 5{x .3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . . and Hi is some other contribution. i. aq2.x'). appropriately decreasing behaviour at infinity). We can readily see that G(x. The boundary conditions which G(x.e.2) where Ho consists. (7.
i (7.10) The contribution on the right hand side is the socalled perturbation contribution. We see that the initial condition (7.#.11) = = 3) = fdx'(H0 .£(0) . . (7.x'. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.x')(E .EM)X*(X) ( .8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x). (7.x'.E^ We can check this by considering (H0 .(*'))*fr') Hi{x'))^(x') f dx'5{x .t).. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.t) since ih^K(x.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').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.E^ of the eigenvalue.2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei.x'){E .) = f dx'G(x. The timedependent Green's function K(x. x'){E . (7.x'. (JL x'\ t) = H0K(x.Hi{x'))^(x').9) = i H0YieE^i%(x)^(x') = H0K(x.E® (EE{0) Hi{x))m{x).E<®)xG(x. x'\ t) (7. .t) (K for "kernel") is defined as solution of the equation ih—K(x.x'.7) with the initial condition K(x. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .0) = S(xx'). We can see the significance of the timeindependent Green's function for the complete problem as follows.x.
x'. In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj.2 (not contained in the first edition!). (7..6) we would have (7.15) ''This problem and its circumvention can be formulated as a theorem.t) for the complete problem as follows. Green's Functions Can we add on the right hand side of Eq. H. But if we assume that E^ ^ Ei for all i. 17. MiillerKirsten and A. The theorem is also known as Fredholm alternative. Booss and D.6) is not defined. Maharana. In general. But then the Green's function (7.11) 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).108 CHAPTER 7.12) Equations of the form of Eq.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. Wiedemann [183]. integral equations are more difficult to solve than the corresponding differential equations. Examples in various contexts have been investigated by L.11) with a system of linear equations of the form Vi = Mijyj. This will be different. J. (7. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). see E. 2nd ed. In cases where an inhomogeneous contribution is given. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. Sec. that (E . it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). (7. Bleecker [33]. D. All such methods of solution are based on the analogy of Eq. (7. however. .11) are called (homogeneous or inhomogeneous) Fredholm integral equations.14) We can now see the significance of the timedependent Green's function K(x. Merzbacher [194]. (7. i. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution.e. W. if we assume that E^ = Ej. Instead of Eq. see B. then at best such a contribution would be something like cipi(x).
21) dx'K(x. in different formulations.' which obviously satisfies the initial condition K(x.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) describes the evolution of the wave function from its initial value ^(a^O). x'.g. Comparison of Eq. H\En) = En\En). (7.0). E. (7. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7.19) i/>(x. \En)(En\ip)t=oeE^\ n (7.42)) Wi=0 = E l ^ ) ^ l ^ = o n (717) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o.x'.x'. 2nd ed.Q) = 6(xx'). 1 See e. 109 (7. *>0.15) which contains (7. . (7.22) According to Eq.20) is the timedependent Green's function.5.t) = J2 fdx/^n(x)^*n(x')^(x'. i. (7. t)\2.t)il>(x'. (5. Eq. t > 0. for the oscillator potential.158.16) As usual.7.e. t) with the Note that when K(x.The solution of Eq.t) = ^ e £ " f / « V n W < ( ^ ) . We can write this expression also as§ ip(x. We use this in Sec. x'.17) as initial condition is obviously W)t = Y.18) or.7) with Eq.x'. Merzbacher [194]. e. (7. 7. p.20) the Green's function K(x. (6.2 for the computation of the sojourn time.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. 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.17) is the initial condition of \4>)t.0)eEnt/in. (7. this relation provides the probability density \ip(x.t) is known..t) = where if(x^'.g. (7.
i /f e «/.x') and K(x. (7. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.x'. x'. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x.«^«2 W .x'. 7.x') = GE(x.1 The contour of integration.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. we obtain / W : = .t) K(x. 24 )  ReE Fig. t) = J2 e £ " f M i ( x ) < ( x ' ) .t). x')6(t) (7.ie (7 . In the following we shall derive the Green's function for the case of the harmonic oscillator.1. t > 0.25) in agreement with the timedependent Green's function . Green's Functions difference that (5 = 1/kT plays the role of it. 7. between (with E^0' = E) G(x. i. Inserting for GE+ie the expression above. Jc 27T „ hn . we obtain then also the corresponding density matrix. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.E .110 CHAPTER 7.e.
29) and (7.30) into Eq.x'. i.0) = 5(xx'). which is moving in one space dimension.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. and identifying coefficients of the same powers of t on both sides. (7. (7. B m0 n _ 2ih 2iK (7.31) The constant A has to be chosen such that K(x.t). A and B being constants.29) tJJ 2 2B(x .27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.e.7).— d2 —2K(x.t) t h2 = . (7. we obtain *I£I = h2 —(2AB).26) This is the case of a free particle with mass mo. It is clear that it is nontrivial to solve an equation like Eq.30) Inserting Eqs. (7.7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.27).28) B(xx')2 t2 DB(xx') 2 /t (7. Thus we try the ansatz.x'. K(x. / dxK(x. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.X>.3 The Green's Function of a Free Particle 111 7.e. In this case the Green's function is the solution of the equation d hK(x.x>.0) = 1.x'. v ' 2m 0 v h i. 2mQK h h2 ih(AB) = —{4B2A). .t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.
(7. p.x'.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—.zEnt.35a) K(xy.112 CHAPTER 7. 32) It follows that x'.20) and can then obtain ip(x.t) into Eq.34) K(x. from K{x. i. 158. *>0.21). we > have to make the replacements V ^ fdk. Green's Functions For parameter values such that the following integral exists.0)oce"Qx2+ifcox. (7.38) . (7. (7. we have This is 1 provided A=JB= ^ (7 .x't) = Y. Merzbacher [194].36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.ihMxWn{x'). (7. 1st ed. v27r En^^.oo 27T (7.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq.e.t) — for instance for a wave packet given at time t = 0 of the form ^(x. (7. j3 = i(xx'). "See the excercise in E."737) \27rii/i in agreement with Eq.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS .(7. n For a free particle moving in the onedimensional domain \x\ < L — oo.x'. 2m0 (7.33). Mx)^M^) „ n so that J =^ .H We can insert the expression for K(x..t) = — / dkeak2+Pk 2vr J. t) = /jJo_ e mo(««') 2 /««.
(7.e.——K(x. x'. .x'. i. . K(x. x'.t).x'.x usn ot i.e. x'. I (7. 5{ax) = ± J eikaxdk = ArS(x). We rewrite this initial condition in terms of £ and use for a = const. z .t) 1 + m0u2x2K(x. (7. 7./ ) + fK(x.t) = h2 2 d2 TW .t).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.7. .37) will later be obtained by a different method — see Eq..e.t) + ~x K(x.x'. —KK(X.25) — in the context of Feynman's path integral method. .0) = 6(xx') at / = 0.43) 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.3 The Green's Function of the Harmonic Oscillator 113 The result (7..42) . (7. (7.e. (7.33).x'. . §jK(x. in this domain *Recall that S(x) = ± f eikxdk.d r.40) becomes 2 . (21.40) Then Eq.x'.t) We now set h2 d2 = . HQ = ^—p2 ZrriQ + \m0uj2q2.X .8). —ih—K(x.7) is the solution of d ih—K(x. / ) (7. the relation* 6(x) = o<y(os). mQLu . . / ) = ~^K(x. 1 2mo u>n ox n with the initial condition (7. i.. i. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.39) In this case the timedependent Green's function K of Eq.
(7. a = . and we expect K(x.t) become similar to expression (7.48).46) If we interpret Eq.47) becomes (7.51) See R. 50.50a) we obtain (7.x'.f) with a(0)»y. we obttain a' = l .50c) (7.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7..114 CHAPTER 7. we must have /o = 0. Green's Functions to the particle behaves almost like a free particle. (7.4a 2 )£ 2 .33). Identifying coefficients on both sides.b2.4a&£ + 2a . as the limiting case / — 0. .) a'£2 + b'i + c' = (1 .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7. it • is suggestive to attempt for K the following ansatz^ m.42) and obtain the equation (with a1 = da/df etc. W ) .Z'. Integrating Eq. (7.47) We insert this ansatz for K into Eq.e.e. (7. (/ = 2a.45) in this sense.50a) (7.45) in accordance with (7. (7. b' = Aab. 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. Feynman [94]. c(0)^. (7. f i.50b) (7. so that for / — 0: > a^—.^ . P. p.c o t h 2 / . i.4 a 2 . that is.b2.
53) or.i) (7.33) for the Green's > function of a free particle.t) = . p. / —————rexp y 27rmsin(u.45). as one can verify. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x. (7. . (5. Finally Eq. 7.e. Felsager [91].40) of the timedependent Green's function with Eq. t. 115 To ensure that in accordance with Eq. (7. In order to satisfy Eq. (7.50c) yields for c(/). x .* With this result we have another important quantity at our disposal.48). 174.e. B independent of / .In B.x'. K(x.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. x'.b(f). we must have (besides A = —£') = /m 0 u.t) to obtain this element (x. with A.50b) Kf) = —r^77) smh 2 / A independent of / . (7. (7.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq.x'. in particular for the derivation of the sojourn time in Sec. i.7.5. i.X'. and to ensure that we obtain the prefactor of Eq.55) 'For an alternative derivation and further discussion see also B./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. c(0) = £' / 4 / . Comparing the Eqs.c(f) K = 5 =exp Vsinh 2 / into Eq.2. if we return to x. (7.27).54) For t —• 0 this expression goes over into the expression (7. we can use K(x. (7.48) 6(0) = — £'/2f. 2TT^ Inserting a(f).0). as we shall see in the following. we must have A = ('. (7.51) for the density matrix PN(X. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / .
e. Eq.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7.116 CHAPTER 7.40)) 1 epEi 1 Ui = y.52): Tr (PNq2) Jx2pN(x. Eq.56a) the expression (7. Thus the system is in a mixed state and the expectation value "Cf. Green's Functions With this expression we can evaluate (cf. (5. 11 For T ~+ 0: u>0 .x. 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"}.» l. R. (7.x. we obtain § < Z> = ^ c o t h ^ ^ ° — . Feynman [94]. <2 (7. (5.e. . 52.. [ [ fdxdx'dx"(i\x}(x\p\x')(x'\q2\x")(x"\i) Idxdx'dx"{x"\i){i\x){x\p\x'){x'\q2\x") J2 [ [ i = fjf^»(x>)(xW)(xW) dxdx'dx"5{x" i.x.52)) the expectation value of an observable A in the canonical distribution (i. (5.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf. P. = jdxp(x. ^—• Irpiv J pN(x.uij>0 >• 0. p.56a) {q2) = = Y. we skip the algebra here.55).P) 2. eP E i = i + y. at temperature T): {A) = Tl{pA) = ^ > ..p)dx W ) = —^r = —p . For instance we have with Eq.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. )x Inserting into Eq.
56b) is that with respect to this mixed state (whose cause is the finite temperature T). 2 + x'2)le^2xx' to—>w—it t—>co 2 . We assume t > 0 and t —> oo.J (w — ie)< e (n+l/2)Et e i(n+l/2)u. 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 (7.59a) E 0 = ^fiw. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. (6. If we consider the system in the pure state \i).7. x'. 1 mo^ exp lmow / 2 2 ft X (7. In this case the Green's function K(x. of K(x.3 The Green's Function of the Harmonic Oscillator 117 (7.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . and we replace to by LO — ie.x'. We return to the Green's function (7.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. which means in the oscillator state \i) with eigenenergy fku(i + 1/2).54) is K(x.21). i.58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states. t) of Eq. e > 0. the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m.e. setting a = mooj/h.50) we obtain.54). x'.59b) n+ l \ ftwt here 2hf exp _ i ( n + .QUJ (7. This is the first and hence dominant term of the expression (7. (7. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( .
5. In a very analogous manner we obtain the solution of the equation for the density matrix.e.. Considered classically. lie on surfaces. of dpN as pN(x. In Example 7.59a).e. The wavelength A is given by The phase velocity v<p. Green's Functions For t large (i. k = k. i.118 CHAPTER 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)]. However quantum mechanically in view of the uncertainties in position and momentum. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. and with this we estimate the sojourn time T. 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 = 0.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const.e. i.x'.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. is defined as vv = . which are planes.1 we estimate T semiclassically. to infinity) the contribution with n — 0 dominates. the velocity of planes of equal phase. (7. In the following we want to calculate (more precisely estimate) with the help of the Green's function the time interval T which a pointlike particle can stay at the maximum of the potential before it rolls down as a result of the quantum mechanical uncertainties.62) . 7. (7. as in Eq. (7. the particle will stay there only for a finite length of time T.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples.
t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.k ') 2 . or dco' da The centre of mass determines the particular phase. oo (7. xt— + — = 0. If we assume for / ( k ' ) a Gauss distribution. to the function ip given by oo / fik'y^'x^dk'. the fundamental postulate on matter waves in Chapter 2): E = hu. duo' da . ' t) dk'.)ll/(fc")l[cos(¥'(fe. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k.e.63) The relation u = w(k) is known as dispersion or dispersion law.<p(k"))}dk'dk".e. essentially again a Gauss curve.Jt + a.7.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. u = w(k).64) where / ( k ' ) differs substantially from zero only near k' = k. i ) = /*/(k')e i(k '" r '' . then ^(r.)¥'(*:")) +ism(<p(k') .Q ( k . (7. i. e. e . ^ ( r . i.e. t) is the spatial Fourier transform of this Gauss distribution — as we know. One defines as centre of mass of the wave packet that value of x for which d(p — = 0.65) Let f(k') = \f(k')\eia and <p := k'x . / oo J—oo . The wave packet describes a wave of limited extent. for which \ip\ assumes its largest value: OO . (7. We now ask: How and under what conditions does the time variation of the function ^(r. a > 0 . i.
Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. we obtain the deBroglie relation p — hk.e. we can use a simpler method. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. Example 7. = — (768) It is instructive to consider at this point the following examples. as claimed for =0. For E = hio. however. 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 = .— g(k).120 CHAPTER 7.69) In general one uses the theory of functions for the evaluation of this integral.. In the present case. (7.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . i. The threedimensional generalization is evidently du .^ 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.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. . p = hk the group velocity is equal to the particle velocity v.fc w> vv = ~a^ = S r a d P E «i7 Sd u . a > 0. dE v 9 = . (7. . Solution: The function to be calculated is the integral dxeax OO ezkx.
7.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 . .71) (7.2: Representations of the delta distribution Use Eqs.75) ' E . (7.1 verify the uncertainty relation Aa..Afc = 81n2. MiillerKirsten [215]. XZ (e > 0).43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x . J.a) + 6{x . (7. (7. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. (7.73) 27T 7 .7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.r „• With the help of Eq.70) we obtain 6(h) = lim — f°° dxeax2eikx = .69) and (7.&)] for a + b. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . (7. (7..69) and (7. \a — b\ (7..4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .72) Solution: From Eqs.74) Example 7. Example 7.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7. (7...70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. W. with partial fraction decomposition. or see H. 2 f + . g . where c = 9(0) is a constant. Appendix A.0 0 7T 7 0 7T e From Eq.
e. A simple calculation yields A i = 2v/2~In2a.* o ) 2 / 2 . i. Solution: In Fig.r)1/2a Re f(x) Ax Fig.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.t) is a measure of the probability to find the particle at time t at the position x. 7. According to Eq.2 The Gaussian curve. requires according to Eq. i. when Ax is very small. The wave function corresponds to the function f(x) in the above considerations (e. Green's Functions 1/(27i)1/2a — 1/2(2.e. (7. and hence implies large values of Ak.78) It follows that the product of the uncertainties is AxAk = 8 In 2.« 2 ( * .77a) small values of a. Physically . the width Ax is a measure of the uncertainty of the probability. The breadth A/c of the curve g(k) around k = kg. at time t = 0). In quantum mechanics the square of the modulus of the wave function ip(x. Thus a slim maximum of the curve of /(a.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. (7. Thus a sharp maximum of the function f{x). 7. where  / ( x )  = m a x  / ( x )  / 2 . where g(k) = max<?(fc)/2.77b) (7.77a) g(k) = e .CHAPTER 7.) leads to a very broad maximum of the Fourier transform \g(k)\.g.70) the Fourier transform of f(x) is (7. is Afe = 2 V 2 1 n 2  (7.
Thus G. which means all values of the momentum are equally probable. E.54) with the substitution ui — iu. ^In Eq.u> = l. the less precise is the determination of its associated momentum.2 ) and U)t.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate.«> = / ^ C ' W .e. In the limit a — 0 we obtain from Eqs. As G. O ) .81) *G. For reasons > of simplicity in the following we set in addition t mo = l.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. . (7. 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. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.h = 1. Barton [15]. (7.7. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. 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.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} . i. Barton mentions at the beginning of his paper.x'. 7. <7*» We assume that initially. Barton recalls that W..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. at time t = 0. This means the particle is localized at x = 0.5.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. (7. The result (7.71) > and (7.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. so that Ki0(x.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O .". The function g(k) is therefore described as momentum space representation of the wave function f(x). (7. nr = S(x).79) According to Eq. In the same limit the momentum uncertainty Afc grows beyond all bounds. (7..75) implies therefore: The more precise the coordinate of a microscopic particle is determined.
Green's Functions Inserting (7.t) = J oo P .cosh t sinh t' Wisll2b ^ (7.83) ip(x. .2 (7. = .81) and (7.79) into (7. 9 (7.t) = A\/2msmh 2TT 2 /2C2+i<px2 tVir^b In t h e further calculations it is convenient to set t a n 29 = b2cotht Then ~A 4 and VA + iB := ReiB (7.88) . ' 2\Aswht % cosh t (7.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.(i/2) sinh 2t 2 2 2C 2 (i) + *c^..86) Evaluating the Gaussian integral of Eq.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7. + cosh 2 1.82) i/f(x. A .84) we obtain* ip(x.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.85) = —r= o2 7 2 sinh 2i. (7.124 CHAPTER 7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = . (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 .(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .
In taking the modulus of ip(x.7. t) the phase ip drops out and one has (x. _2 1 exp C 2 (i) = l + 2sinh 2 i.t)\2 Ji == 71 / V " Jo dO ? 6' (7. (7.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .95) . We have therefore .t)= where Co(t) / dx\^(x..t)\' 2 125 (7.94) / dxf(x)=g'(y)f(g(y)). •&(t)=//c(t) = / or.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.£ 2 (7. C2(t) (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).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.4 The Inverted Harmonic Oscillator The real part of Eq.91) C(t)' (7.89) = exp[x2/C2(t)} 7rV2C(t) (7.90) with the expected limiting behaviour \im \ilj(x. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.
+ 0(e'2) (7. we obtain dT _ 1 ~dJ~l 'See for instance H. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr..= \ / l + 2sinh 2 t. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7. p.126 We set I C(t)' so that dT ~d~l Now.2)(p_7?2)(7.1 ) 2 \n J ^"•""ff^^V^K?. 136. B.(4 2 . Dwight [81].=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.From this we obtain C(t) C'(t) Z2 ^(r?2_.97) sinh 2 1 = .. C(t) = .97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.e ~' V^~i Jo Vn~i Jo ^ rx=VM _Jl 2 Using the following expansion of the integral* _ _ r e~*' dt L 2 « 1 2e~*2/2 7T X /2^7.98) We insert this into Eq.=0 V^ C'(t) I Jt=0 V^ (7.99) . (7.96) °° l Jr.
7.7. also motivated by this paper. B.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. 238. Zwiebach [294]. See e. Explain the parameters entering the calculation.100) Reintroducing the dimensional parameters we had set equal to 1.g.102) 'As a matter of interest we add that in the theory of a free scalar field x. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7.4. this equation is the equation of motion of this scalar field with negative masssquared.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg. T h e following Example 7. p. (7. Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. It follows t h a t T~lnZ. 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. Fig. The equation there describes the classical "rolling down" of this tachyon. . is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. U) (7. this is T ~ ]n(ly/mou>/h). Such a field arises in the spectrum of states in string field theories and is there called a "tachyon".
2V ' Let A and Bio be the values of x and x at time t = 0. Green's Functions x = Aco sinh cot + Bco cosh cot. 2 10 x(0) = B u . and for simplicity we take (cf. (x 2 ) h 2mou 2mou> and so l^J^ULl). Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T. (7.p(0) cannot be determined with arbitrary precision. i. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. . but instead are subject to an uncertainty relation of the form AxAp > h. 2 x(0) ^ where A i . x(0) = A. Quantum mechanically x(0). v A x A p = \ (x2)(p2) v = m0u(x2). pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. For h —> 0 the time T —• oo in agreement with our classical expectation. (7. the smallest macroscopic length. Thus quantum mechanically x(0).5)) Ax = <J(x2){x)2. Ap= v /(p 2 )<p)2. A p are defined by the mean square deviations given by (see Eq. (p2):=mW(x2). and solution x = A cosh Lot + B sinh cot. ^(x2)(p2).102)) p(0) = ^/(p 2 >. (10. so that e. Eq.128 with x2 = LO2X2 + const. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*).e. i. CHAPTER 7.g.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. so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . / 7 ^ 2 =2J(x }.e.
2h2 cos 2 a # = 0. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. In general perturbation series do not converge. Frequently even the calculation of the next to leading contributions ip^. i. For mathematical purists the question of convergence of the series is an even bigger challenge. 129 .(i)+/^(2) + ". in closed form) only for very few potentials. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. one would set tt E = = ^(°)+/ty. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.e.E^ is already a bigg problem.e.Chapter 8 TimeIndependent Perturbation Theory 8. 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. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion).
• + (ir~ X X* 1' 2' r?l Xn + Rn(x). TimeIndependent Perturbation Theory In the case of this equation. N. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ .. T.* i. Whittaker and G. (8. Watson [283].= 0 < 1. . the expansions in descending powers of h2. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence. E = E^ + h2E^+hAE^ 71 72 7 + . p.2 However.130 CHAPTER 8. 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.. Meixner and F . E.g. (8.+ .1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. at the end of Chapter 26) — and by comparing the results of WKB. This convergence can be shown with D'Alembert's ratio test since''' lim .g.g. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •.. h hl are in the strict mathematical sense divergent. . 0.2) n—»oo n It is interesting to compare the behaviour of the series (8. J.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. 581.. 8. Schafke [193]. ^For further details. concerning also the uniform convergence of the exponential series. which is known as the Mathieu equation. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. W.g. It will be seen later (e. e. see e.3) *See e.e. In the following we explain the difference between convergent and asymptotic series.
From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .4938. e.9876. / ( I ) = 1 — 1! + 2! — 3! H • However.0.g. 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.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. 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. after reaching 6/27 = 0.22222 .l n ( l + ^ ) n=l L ^ ' (8. p..2. for x = 3. the series (8.. +0.. . . and we see that the moduli of successive terms first decrease and then. It is evident that the larger the value of \x\. /(1000) = 1 .. The theory of asymptotic expansions claims that if the expansion is truncated at the least term.376 .3) diverges for every value of x.000 000 06 + • • • .e. begin to increase again. Whittaker and G. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.g.. . However.8. for larger values of x. i..4) for every value of x.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.. E. Watson [283].001 + 0.0. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 .33333 .22222 . .222. T.3) with the convergent series (8.29629. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. . Thus e.. Comparing the asymptotic series (8. We observe in the first place that since lim n—>oo co (8. . in spite of this the series (8.1).000 002 00 . We can see this as follows. + •••. N.5) Cf. and in fact increase indefinitely. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!.. the better the approximation obtained.. + 0.. 235..0.0. + 0.0.
Sec.e. Watson [283]. § However.132 CHAPTER 8. p. Schiff [243]. TimeIndependent Perturbation Theory where 7 = 0.1." These "rapidly converging power series" in physical contexts are most likely very rare cases. p. that E and \ijj) (i. p. . It seems that Schiff tries to cling to the idea that a proper series has to be analytic. 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. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. It is interesting to observe the comments of various authors on this point. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. Messiah [195]. meaning convergent. E.41) and (16. In many respects Messiah aims at more rigour in his arguments. In fact. 11 L. the leading term of an asymptotic expansion. Merzbacher [194]. This latter observation hints already at the importance of asymptotic series in applications." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. I.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. T. ^E. ip) can be represented by rapidly converging power series in e.J In2. (16.5). it is sensible to make the assumption.5) are practically unknown.1 between Eqs. in fact. whereas applications of the convergent series (8. N. 371. the dominant approximation is. Vol. 16. **A. Here the series on the right is an asymptotic series which is particularly useful for large values of z but is readily checked to yield very good approximations for values as small as 2 or so. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. This is not appreciated by mathematical purists. 152. 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. although this has not been investigated except for a few simple problem^. 236. Whittaker and G. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems).
II. 2 (8. B.10) Vn Changing t h e variable of integration to u = t2x2. which have been studied in great detail. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion .1 T h e error function a n d Stokes discontinuities Another extremely instructive example which illustrates t h e n a t u r e — a n d in addition t h e origin — of asymptotic expansions is t h e error function 4>{x) denned by t h e integral <f>(x) = 7= / V71" Jo 2 fx e _ t dt.7) as / e'* dte* dt = 1 .2. Vol. He says. (8.7) This function has been considered in detail in t h e book of Dingle. (8.. (8. 198 (German edition). 8. we can rewrite Eq. (8.8. It is therefore suggestive to write iy poo <t>(x) = 1 . .. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.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. solutions of second order differential equations. ft Jx du = 2tdt.= / e~t2dt. Messiah [195]. Dingle [70]." Of these three authors t h e first.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.^=e~x2 / e^^dt. We know from books on Special Functions t h a t all of these functions. Merzbacher.11) A . seems to be closest t o t h e t r u t h a n d does not attempt t o give t h e impression t h a t t h e series has t o converge. p..2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. there is no need for such bias.* and we follow some of t h e considerations given there.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. In fact. (8. *R.
(8. since (—4)! = y/n. Then e (n —\)\( n\ 2du u \n Xy/n o du VK n = 0 e 2 + XypK ~"(1+ ^ °° /•oo —u (ni)!/ n! (nn\ 2)! u du+ Jx : e u du 1 ^ ( 1 + n=0 oo /_u_ V x2 ^ ) K ^ n=0 (8.^ X e U ~ ( 1 + * \ 2 .16) .12). 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .e.^e~x \Ar i.i)[(_!)«' Then. du.134 we obtain CHAPTER 8. (8. Using the reflection formula (*!)!(*)! = for z = n + i . TimeIndependent Perturbation Theory <P(x) = 1 . we obtain 1+ i.13) presupposes t h a t < 1.15) < 1 we can insert this expansion into Eq.' we have 2 nH 7T 7T (8. « \~2 =E n=0 l 1 V5F(ni)!(l)n nlTT (M x2) / « r oo 1+ For the domain \u/x2\ <f>(x) U \2 1 \/K n=0 2< n! V (8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8. 7T (ni)!sin{7r(n+^)} ( „ .e.14) sin7T2.
i. whereas in a convergent expansion the terms first increase in magnitude. and in fact so slowly. (8.2 Comparison of Asymptotic Series with Convergent Series 135 We can evaluate the first integral with the help of the integral representation of the gamma function. .e. convergent) only in the restricted domain u < x2. i.r<7r.12) ignoring the fact that the latter is valid (i. In an asymptotic expansion the first few terms successively decrease in magnitude. (8. It is fairly clear that the above considerations remain unaffected for argx < 7T. We now see how the asymptotic expansion (8. (8. We can actually understand the deeper reason for this in practice.19) originates. much more useful than the convergent expansion. In the foregoing discussion of the error function we assumed that x was real and positive. Frequently "=" is written instead of "~". ' arg. As in the case of the gamma function.—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^.18) or that we insert the binomial expansion (8.e. 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.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. We now want to relax this condition and allow for phases argx ^ 0. Whichever way we look at the result.8.19) implies either that we ignore the remainder or the correction term given by the integral in Eq.l)! = / Jo Then 2 cHnxdt. The expansion (8. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. /•oo T(n) = (n .17) 0(X) = i. (8. argx = 0.15) into (8. in general. the asymptotic expansion of the error function is.19) approximates <f){x) the better the larger x is.e.
(8. however.20) where we set t = is.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.22) we obtain 2 4>{iy) = .26) or else return . We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . i. (8. thus with \v/y2\ < 1.^ = . =dv. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. We therefore consider this case in detail. (8.24) We can read off the binomial expansion of the factor in the integrand from Eq.136 CHAPTER 8. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.^ f V e ^ . we can try to proceed with Eq.vds. 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. V7T Jo Changing the variable of integration to v (8.15). The situation becomes critical. (8.23) Throughout the range of integration 0 < v < y2 the integrand is real. VK Jo V^ Jo (8. (8.2 y V .e. dv = 2sds = . we set x = iy in Eq.21) = y2s2. when argx = 7r/2.
27). We know that for v < y2 the integrand is real.. Eq. Returning to Eqs.28) we observe that the expansion (8. ey is much larger than something of order 1/y. — (8.24). i (8. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 .2 yir (8. e ^ ( i y ) ^ ^ . Suppose now we consider Eq.28) n=0 We observe that this expansion differs significantly from (8. (8. We have therefore <j>(iy) = ^e^TL fV e~v ( l .17): J£° e~vvndv = n\.e. (8. (8. and the larger y. (8. i. (8.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . . which will be studied in more detail below.27) Then . If we proceed with Eq.e. for large values of y.30) yvn f Jo V y J Cf.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions.19). For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.^] * dv. (8. (8. i. However. y y 2 ( n _i)! y arg(* = iy) = i * . (8.8. the better the approximation expressed by Eq. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .^ ^ p . 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.24).29) where 5t means "real parf. and so the integral is real.27) and (8.2 Comparison of Asymptotic Series with Convergent Series to Eq.28) ignores the integral contribution of Eq.28) because then the correction term oc f°£ becomes smaller and smaller.
28). (8. y. Dingle [70]. i. and • higher order terms of the associated series all have the same sign.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. (8. We have therefore found that XTT n=0 ^ ' and .e.e. The corresponding phase lines are called Stokes rays. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. since (8. Since (cf.. (8. . e xZ ^ (n — £)! „ ^ ) .31) and (8. Chapter 1.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. (8.33)) would have • a maximally increasing exponential e+x .3!) which is (8.31). 7r = .25) and integrate from 0 to oo. With a similar type of reasoning one can show that the same expansion.138 CHAPTER 8.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. . *Cf.2 OO rz=0 V . (8.33) The sudden disappearance of "1" at argx = 7r/2 hints at something like a discontinuity of (j>{x) at arg x = n/2 which was discovered by Stokes and is therefore known as a Stokes discontinuity. (8. Looking at Eqs. R.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. This result is identical with that obtained previously. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. Eq. is obtained for <f>(x) with arg x = —ir/2." — E ^ ^ forarg. B.26). TimeIndependent Perturbation Theory If we insert here the expansion (8.
.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. (6. 1/1 ipv{x) v1 vl (±ix).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. 5. '"Cf. Chapter 1. in the sum of the right hand sides of Eqs.37) Cf. apply predominantly to asymptotic series of the solutions of second order differential equations. Eq. B. See also Tables of Special Functions. though. e.2. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . Equation (8. 9 . i.g.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x.36) which is of the type of a Schrodinger equation for an harmonic oscillator potential* In view of the considerable importance of the harmonic oscillator and the associated parabolic cylinder functions in later chapters we consider this case now in more detail.34) the contributions 1.13.8. and so may not be universally applicable. 91.32) and (8. Thus if we set x = XR + ixj.e. W.2). and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. Oberhettinger [181]. p. These rules.35) 8. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0.e. pp. R. i. We also observe that the expansion for  arg x\ = ir/2 is half the sum of the expansions on either side of the Stokes ray. Dingle [70]. (8. Magnus and F. —1 cancel out.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). for arg x = 7r/2.
Hence Dil){x) = xv(j>v(x). We consider Dv (x). (8.40) § Cf. (14). Dingle [70]. 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. (8. B. Eq. Chapter 1.e. The continuation of (8. B. i. Chapter 17.x 2 oo «*)  — E^fi=0 v . Dingle [70]. TimeIndependent Perturbation Theory i. v — — f — 1. R.37). 1 1 Cf. x~l/~1ipu(x). Chapters I (in particular p.38) are of the type (2i)! i\ i\(2x Y 2 (2x2Y and so have the form of a factorial divided by a power which (as discussed previously) is the behaviour typical of asymptotic series.e. 9) and XXI. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. We also observe that the late (large i) terms of the series in (8.39) onto the Stokes ray and from there into the neighbouring domain is determined by the following rules which will not be established here:" • Dingle's rule (1): On reaching arg a. > > The equation (8.36) is invariant under these replacements.7.140 where^ CHAPTER 8. as can most easily be seen in the case of the Mathieu potential. 8. = 7r/2. 0 < argz <  (8.39) (as in the example of the error function). .38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. ti asymptotic series (8.ei<v+VxvHv{x\ argx = £ . R.e. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. i. with a real proportionality factor. Thus D{J\x) (i.
i. We therefore proceed to continue (8.41) to arg a. (8.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).e. with \x\ = (—x) for arg a. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i.40). (8. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. we must have at arg x = 7 in the dominant factor on the right of Eq. where in the first line on the right hand side the second term contains the real proportionality constant a/2. = ir.e. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray.e. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. = ir. 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.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. another half of the discontinuity appears on leaving the Stokes ray on the other side. (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv).e.44) + ^iap) = 0 .41) the Stokes multiplier a is.8. for xj = 0) and therefore possesses a Stokes discontinuity there. beyond arg x — 7r/2. as yet an unknown real constant.41) xuct)v{x) + a{x)~u~lipu{x). (8. On reaching ir the part containing ^ ( x ) . i. < 7T. < 37r/2. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. In (8. the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first). Since Dv (x) is real when x is real.42) T ^(eiwu i. The value of Du (x) on the other side of the Stokes ray.
42) that a multiplies ipv(x).42)) for argx = ir (in the second line of Eq. (3{v . TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. i.1)! TT. (8.1) = ±a(y).e. We can see that the equation is satisfied by (3{u) = ±a(u . (z)\(ziy.1) (8. (8. (8. (8. i.v .45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1. (8. (8.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .1) = 2sin7ri/.42) we insert from Eq.1) (8. a{y)P{y) = a(u .1).50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. argx = vr.14).—^ ON • — ^ > + ^ ^ L .e.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .45) that af3 = 2sin(7r^)) D^\x) = (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. Hence the left hand side must have the same property.48) We compare this with the reflection formula (8.e.142 CHAPTER 8.e.45) is therefore given by a{v)a(v . (8.51) .l)0(v .46) For a given function a(y) this equation determines /3(u). Eq. i. i. V i=0 \U0 v 2w . . 8.47) Equation (8. Dl \X) is (cf.
the second contribution vanishes and we obtain £>«(*) .e.1 ) ! 4~L i\{2x2Y i=0 V2^(x).7r). (8.35). (8.e^ .53) We have therefore obtained in a natural way the quantization of the harmonic oscillator. which vanish at x = ±oo. 8.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. —f.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator.A / ± sin^Cx)"M* 2 V ^ . (8. For details we refer again to the book of Dingle. Chapter I.. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .„ _ix2^(2ivl)l e 2 > —.50a) because this enables us to define normalizable parabolic cylinder functions. i. (2x2)1 = COSTTU(X) v .I / .36). (8. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.52) (argx = 7r). In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0. i...7). i. ' ^ (i/l)!z! ^ . (8. Eq.. ' (X) (cf.e.1 ) ( .. We observe that for v — n = 0.8. However. It should be noted that in Eq.e. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.)(. u 1 y> {2i + v)\ 1 n 0.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function.1.2. the error .*"ei°> f ( ^ _  _ L _ = (!)««(.36)) in the domain 0 < arg x < IT.. solutions of Eq.) (8.
59) and equating to zero the coefficients of powers of x2. % = {. how series of the type (8. x°. Instead of dealing with Eq (8...36).60) 2 x x l Inserting (8. 0. 5„=4 Sl=o. Eq. S2.37).38) can be obtained. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. (8. Eq. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. cf.55) The large x asymptotic expansion can also be obtained from the differential equation. and demonstrate.e. • • • can be determined. (8. x.144 CHAPTER 8. \ + U (8. i.^ + 1 ) .36)) 0 where + x2Q(x)y = 0. (8.54) we consider again the important equation of the harmonic oscillator. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 .e. We write Eq. (8.54) (8. S\.. i. ^ + 2 . + — + ^ + • • • .60) into (8.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. previously we had ^.36) in the form (coefficient of — x2 here chosen to be ~. Solving these equations we obtain = = 0. x—>±oo (8. (8. we obtain equations from which the coefficients So.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.
are asymptotic series in a variable x. T h e expansions (8. In the following we are mostly concerned with asymptotic series in some parameter. do not enter here into their derivation and simply quote the result: W i t h \x\ . 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+ . 4~l~2 ' i i _i. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. function. We. i. The derivation of the asymptotic series from integral representations (cf. and it is clear t h a t higher order terms follow accordingly.z) is a confluent hypergeometric T.61) (the dominant terms of (8. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. etc.e.35) and (8. 2 X 2Wi.36) have correspondingly e±x / 2 ) .37) with (8. » 1.63) **W.37). \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . (8. (8.b.62) 4 \ 2 XiFi 2'2 .8.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before).38). therefore. n 22e"T JziL (8. Oberhettinger [181]. .19). p.+ X M W = Q. Nonetheless we shall need asymptotic expansions like (8.91. / i N . Magnus and F.38) in a variable for the solutions of approximated differential equations.
Whittaker and G. E. T.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 " ' (8.0.65) 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. 8. 4 (8.146 CHAPTER 8. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . N. .1 a2 an a0 + — + ^ + . Watson [283]. Thus there is no contradiction^ between (a) and (b).66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially. whereas e~x '4 increases exponentially. top of p. 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 3.67) Cf.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.4x2 (8.+ — + ••• of a function f(x) for a given range of a r g x .4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. 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. 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.
e.69) and the consequent nonunique definition of the expansion. we set E^En * > ^ n = En°1+eEnV = + e2EnV + . 8. (8. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". (8. in particular with regard to possible exponentially small contributions to the expansion for which e. lim xneW x—>oo = 0 (8.73) Vn + q/#> + e2T/i2) + • • • . A critical discussion of the definition can be found in the book by Dingle [70]. The definition is due to Poincare (1866).68) is (effectively) simply a statement of the observations made at the beginning.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .Sn{x))  = oo n—>oo (8.5 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. Chapter I. 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. which is assumed to be small. . i. first published in 1902.8.70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (871) / dx%l)*m(x)lpn(x).72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. possesses a whole chapter on asymptotic expansions. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.g.68) The definition(8. In fact. (8.
75) HoW + vW = E^+E^+E^.e. The second equation can be rewritten as {Ho .72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n. TimeIndependent Perturbation Theory We insert these expansions into Eq.i4 0) )V4 1} = {EM . We obtain then the following set of equations: Hrj.70) and equate on both sides the coefficients of the same powers of e. (8. We insert the ansatz (8.n = 40)V>n.n.4 0 ) M = / ^C(4 1} . so that we choose / • dxrpM^O.78) into Eq.148 CHAPTER 8. J f°° dxrn E(^ 0) . Ho^+V^n = EW^+E^n. i.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.74) (8.As a consequence the state vl/) is a vector in this space. ^ ) = 0.e. 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.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . i. (8. 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. (8.e. . °° i J or 0 = EWJdx{rnVil>n). ( ^ .76) The first equation is that of the unperturbed problem. With the help of the orthonormality condition (8. (879) We multiply this equation from the left by ip^ and integrate over x. (8. (8.w » .V)i.
4 0 ) M 2 ) = ( 4 1 } .79). (8.e.^(EW ~ V ) ^ + E& / J—oo dx^miln . (8.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n.V)^ Setting + 42tyn.Viprt e (8. (883) Thus the procedure forbids the equality of any E\ '.8. (il)i. 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).Vil>n) (880) In order to obtain ipn .76) which we rewrite as (H0 . (8. (8. We therefore consider Eq. ^ n ) + 0(e 2 ).82) *> = ^ + E . we return to Eq. S ' ti^+Q^)• We observe that Eq.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. i. with En . In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. (8.i ^ 0. We multiply the equation by ipm. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n. multplying Eq. the coefficients a\ .83) makes sense only if Ef] ^ EnV for all i + n.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. i.m ^ n and integrate: oo /•oo / oo dxrmY. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates.e.
£i°») 2 Hence +^£o (£«? .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .85) together with (8.B i 0 ) ) ( £ f .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.89) • • • where we guessed the term of 0(e 3 ). TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.V^n) .e. En0) (8.81) we obtain (2) _ _ {lt>m.Vlpn] M _ P (0K 2 ^r .150 CHAPTER 8.Vlpn)E.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n . j. Inserting the expression (8.(1) (i&0) . V ^ i ) X (Q) (Vi.Vlpn)(lpn.4 0) ^+ e 2 E (lpk.Vlpn) E^ .86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.^ ) + $& w £ (^fc.V^j)(^j. FVn) + e 2 ^ ( V > n . (8. n) £f) (8. ~Ej E E. (8.V4>n){lpn. (8.2 (V> m .^w .87) (i>k.^(8.VV .88) (Q) Inserting Eq.V'>pTl (E^Eny V'fc (8.79) •/—oo oo / Using (8. ^ n ) j&iift + ^n .^j)(V . ipn ) = 0 i.
8. One would prefer to lump it together with the other contribution. 8. We then obtain (lpk.n(x) = J2crtn+ii (8.90) is the term with only one summation J2 m the contribution of 0(e 2 ). in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w.90) can be found in just about any book on quantum mechanics. An ugly feature of the expansion (8. The functions ipn(x) are eigenfunctions of some secondorder differential equation.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations.Vlpn) = y^Ci(lPk.7 is based on this procedure. We can therefore rewrite tpn ': The perturbation theory described in Sec.e..89). .g. Merzbacher[194]).91) The coefficients Cj follow from the recurrence relations. (8.lpn+i) = ^ Cj5k.. i.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0).
93a) (8. (8. (8.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8.96) (Ho .e. m = I.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . all n.We now insert Eqs. In order to deal with the problem of degeneracy we consider the simplest case. (8. . Equations (8. is not uncommon. degeneracy. For example the magnetic states enumerated by the magnetic quantum number m.94b) with (ipn.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . Equality of eigenvalues.152 CHAPTER 8. TimeIndependent Perturbation Theory 8.ipn).93b).93b) The lowest order eigenfunctions belonging to En. i. are allowed to be equal since otherwise the expansion becomes undefined.e. .95a) (8. (8. (8.96) are effectively the known equations of ipnitpm and so yield no new information. (H0 + eV)^n and (HQ + eV)Vm = Emym.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. Equations (8. Em = 4 ° ) +eE$ + e2E$ + . .V)(c2l^n + ci 2 ^ TO ).97b) with ipn y£ ipm. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.97a) (8. i. i.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. (8.e.97a) and (8. so that En = 4 ° ) + eE™ + e2EP + ••• ..94a). in a central force problem all have the same energy unless the system is placed in a magnetic field.70). (8.93a). I — 1 . . —I. ..4 0 ) ) ^ } = (Em] .97b) are . Em can now be linear combinations of ipni'ipm. (8. (8. + c22^m). however.94a) into Eq. that of double degeneracy En = Em .ipm) = 0 = (^m.
98a) Setting I = n. ^ ) = ( V . .H0il>W) = cn[EW . and ( ^ m .EJ®(lPn. V^n).m).99) and ( ^ ^ O ^ ) " ^ ^ .^ ( ^ . Similarly Wn. Vi/>n)] .ipm ) the equations (iPn. H0lP$) = C22[E$ ~ (Vm.H0^) . £ aW^°Vi) .ci 2 (V„.J#»a&> = 0j since ii4i = En (degeneracy). H0lpV) . (^m.V^n).98b) + Culpm) + cuV^m). Wm.(Vn.y>m. (8.C 21 (^ m .m. ^ ) ) = CU[EW . we obtain with (ipn.6 Degenerate Perturbation Theory 153 inhomogeneous equations. ^ o ^ ) ) = 0. Equations (8. But (8.H0xl>$) .^ ) ( ^ m .Vil>m)] V^m).(lPn.cnVipn and tyi. Then (^.8. cn(ipm. (8. Vl/. We now multiply these equations from the left by ip* and integrate. Vlf>m)] . c2iV^n + c22V7pm).ipn n) = 0 = (ipm.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .£ < « = ag)^) .(^04°))41)) = EM^CuTpn (ipl./>£>) = C2l[E£> . V^n)\ .C22(Vn. t f ) = a k 1 } 4 0 ) " Eg>aU = 0. (^n.
This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate.100a) determine the coefficients Cn.100b) determine C2i. in particular in parabolic coordinates (see Sec.101) This secular equation determines the first order correction En • The equations (8.E)tp = 0. i. \nip = ln(j> + c o n s t . ip = c<p. It is then possible that a special perturbation. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. 11. i> 4> Hence there is no degeneracy. TimeIndependent Perturbation Theory (l/>m.. Example 11. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. At x = ±00 the constant is zero. However.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. like that providing the Stark effect (proportional to rcosO). cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate.e. A onedimensional system. 9} ip. . 4. Then / > i>" 2— = V ip Hence 0 = V ' V .E™ (Vn.^Vm) 0. as we saw.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). d>" E=^.1. (V>V</>V>) = const.6. the degenerate perturbation theory must be used. If one wants to perform a perturbation calculation in such a case. It follows that ip and <j> are linearly dependent.Vrl>n) . Example 8.154 and V (^m. In this particular case therefore nondegenerate perturbation theory can be used (cf.e. dx i.5).^ ' V = — (•>!>'<t><t>'i>).e.Similarly Eqs. (8. if)' (/>' — = —.3).Vlpn) CHAPTER 8.C22. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . i. the secular determinant. Let ip and < be two eigenfunctions belonging to the same eigenvalue E.ci2. splits the problem into two independent onedimensional systems. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. . (8.
Dingle and H. and the solutions then have to be matched in their regions of overlap. W.t Since in general a solution is valid only in a limited region of the variable. We assume that V can be expressed as the sum of two terms. V = V0 + v. We consider an equation of the form Dcj) + (E . anharmonic oscillator potentials and screened Coulomb potentials. the considerations given below would be valid even if D contained only the derivative of the first order.104) are chosen such that the equation Dcf) + (E0 .106) (8. T For instance spheroidal functions can be studied with this method. Eq.102) can be written D<f> + (EQ . J. i.V)(p = 0. and such that one can even obtain a recurrence relation for the coefficients of the expansion. the method has to be applied in each domain separately. cf. of course. J. (8. Applications of the method are. (8. The subdivisions (8.V0)<t> = (v. B. J. with some restrictions on the function V. a potential. H. It is immaterial here whether D does or does not contain a first order derivative — in fact.103) such that if E is written correspondingly E = E0 + e.* In various versions this method is used throughout this book. Muller [216]. W. Dingle [72] and H.102) where D is a second order differential operator and E a parameter. (8.103). It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. See also R. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation.e. also possible in areas not of immediate relevance here. .7 DingleMuller Perturbation Method 155 8. (8.105) (8. Muller [73]. B.V0)<f> = 0 (8.e)<t>.104) "This method was developed in R. in particular in applications to periodic potentials. Thus the method is a systematic method of matched asymptotic expansions. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term.8. Muller [210].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.
V0)<f>m+a = 0. m + s)0m+s. s an integer: (ve)(pm = 'Y^{fn. (8.113) .+s ' (8. v = {D+EmVo)<t>m.e\ < \V0 .VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .m + s)4>m+s s (8. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . and so v (8.e ) ^ = 2 ( m .V0)<f>^ . where m is an additional integral or nearintegral parameter (depending on boundary conditions). the equation Dcf>m+S + (Em+S . the function 4>(0) = <t>m (8. (8. m + s) which are determined by the recursion formulae.Em+s)(/)m+s.m + s)" in order to relate them to "steps" from m to m + s.) This result is obtained as follows.102) with E ~ Em. since D<f>m + (Em .108) with coefficients (m.Vo)(j)m+s = (Em .m + . TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.109) D(j)m+s + (Em .110) If we write the next contribution <f>W to <p(°> in the form m+st (8.156 CHAPTER 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. We write the coefficients "(m.E0\ (the latter in a restricted domain).V0)4>m = 0.111) s then (see below) 0(1 = E (m. Then if ]e < \E\ and \v . Considering (frm+s we have.
m + r)(rn + r.m) E J^ ( m ~ Em+S) ' • fa. We observe that the coeffcients of expansions (8. (8.m)4>m in (8.Em+S) = 0 (second approximation). (8.^ (Em . /01_.m) 1^1^ IW .m + s)(m + s.113) must vanish.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .m + s)(m + s.119) follow a definite pattern and can therefore be constructed in a systematic way.116) m ^ v (m. . (8.n\ ^119) + . . V ^ V ^ {rn. We then find altogether ^  l^m  Em+S) + y^y^ SJ. . the coefficient of the term (m. (8.T? S The first expansion determines the solution 4> (apart from an overall normalization constant). (8. Eq.7 DingleMiiller Perturbation Method and so (m.„N This procedure can be repeated indefinitely.m) + > ^ . .119) is effectively the secular equation).TP \(T? .m) — 0 (first approximation).m + s)(m + s.Q (m.m + s)(m + s.. Thus (m.102) to order (1) of the perturbation (v — e). (8. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.110) and are therefore related to the inverse of a . (8. The factors (Em — Em+S) in the denominators result from the right hand side of Eq. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m. .118).e.8.m) — —.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.117) „ .
(8. 2 .[q(llq2 + 1) + 2(33g 2 . The eigenvalue expansion obtained is an asymptotic expansion. S.120) n4 . does not occur in any of the later contributions. for the investigation of the behaviour of the late terms of the perturbation series. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion.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. . has been treated by R. ^The potential A r 2 / ( 1 + gr2). i. for instance. W. TimeIndependent Perturbation Theory Green's function. This means the solution is not normalized. since it will be used extensively in later chapters.423) + Z(2720 9 3 . in which q = in + 3. An explicit application of the method (not including matching) seems suitable at this point.I2q + 64) + 256Z 3 (41 g . as will be seen later. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . n = 0 . J.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . . One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. 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. The following Example 8.9) + 4096J4] + 0 ( l / g 3 ) .2). as already mentioned after Eq.e.[ 4 ( 8 5 g 4 + 2q2 . .* Example 8. this is — so to speak — the price paid for obtaining a clear systematics which is essential. (f>m in the above. This can in fact be done and will be dealt with later.e. . Kaushal [146]. § H. as will be seen in later chapters.6q + 1)1 + 24(5? . The result is the following expansion.158 CHAPTER 8. and is typical of a large number of cases.38).2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. Miiller [211]. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. 1 . (8. i.1)( 2 + 64J3] — . for instance. Other applications can be found in the literature.
159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. Sees.123) The first approximation (note the last expression as convenient notation) ^ . multiply the equation by onehalf and set h = —a/g. (8.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . . a .a + l)if>(a + 1) + (a.2) where 1/ .2 '(l + l) . b. . a)ip(a) + (a. .. „2 J„2JL._ „!2 ^ ( " « 2 r r 2 > .b.2a2 A. 2 The solution XQ(S) = $ ( a .8. We now insert this equation into Eq.1). Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . Therefore in the general case we may set k2 + g2 = ga(2l + q) . . (8. \2 .b.z2 is a normalizable function if a = —n for n = 0 . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. 11.0(z) = zl+1ez2^fa. Setting q = in + 3 implies k2 + g2 ga(2l + q). (fc2 + <?2) 4ga and b = I \—.• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a.l)ip(a . 2 . (. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. S) is a confluent hypergeometric function. . 3\ 2 V ' 2. 1 . The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a. .V ! 4 I z S *! V2 ipq(z). The solution i.122) where A is of order 1/g. i/> with Dq = d2 . (8. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.121). 1 1(1 + 1) 1 .6 and 16.123) the contribution fl<°> Ah + .
120). where [a. a .a]i = .Z. a + r) + S m _ i ( a . a) = 1.a + 2]i P . a + r) satisfy the following recurrence relation: Smi(a. a ] x + L _ 2 J 1 [ « .l . As a check on the usefulness of the perturbation method employed here. fa. So (a. a + j)i/>(a + m). . fa.a + j}i+1^(a + j). _ J [a . The coefficients S m ( a . ( 5 m ( a .e. the latter determining to that order the eigenvalue.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps".^ 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 + l)(a + r + l .i ^(i)=£V+i i=0 J2 j = (i+2). 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 + 1) = a = In general CHAPTER 8.a—lli. when j = 0. If we now evaluate the various coefficients in the last expansion we obtain the result (8. . (8. a + i)(a + r.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except.a — 211 .a]2+L ^ J1[a + 2 .1) = a . „ .160 where (a. a + r) = 0 for r > m. 0 = h[a. In fact. a ] i + 0(/i3) [a. TimeIndependent Perturbation Theory (g .2. Now we make the following important observation that Dqip(a + j) = jip{a + j). a]i + h fa. Hence the next order contribution to tp^ becomes oo i+2 r [a a . a + 111 r [a. a) = b . This equation shows that a term fiip(a + j) on the right hand side of Eq. a + i) = 0 for i ^ 0. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . one could now use the RayleighSchrodinger method and rederive the result (8.120). of course. a + r — l)(a + r — l . a + r) with the boundary conditions Sn(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.2a = I + q.b = (g + 3) . a ] i + L ' ^ J 1 [ a + l . (a. i.S m (a.a]1=0.3). (a.j^0 ' + jU+1 J ^(a + j) with h[a. o + r) + S m _ i ( a .
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. 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. We therefore begin with the appropriate classical considerations with a view to generalization. proceeds along similar lines. 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. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets.Chapter 9 The Density Matrix and Polarization Phenomena 9.1 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. the generalization to many particles.e. Regrettably this has to be left out here in view of lack of space. light. and this means electrodynamics. 9. i. 161 . although not exactly straightforward. as we shall see in Chapter 27.
1 The vector potential ellipse.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0.e. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.3) (9.i). (9. i.6) . 9.1) Fig.t) = 3ftA.2) (9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.1. This is the equation of an ellipse as indicated in Fig.162 CHAPTER 9. In general the constant Ao is complex. (9. 9. It follows therefore that A(r. t) given by A(r. t) = A 0J R cos(cji) + A 0 / sin(u. (9. The Density Matrix and Polarization Phenomena potential A(r. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.4) (9. A 0 = A0R + iA0I.
A« we can construct only the following scalar products with this invariance: (A(°\A(°)). ie. Rae [234].e. (AW. determined with the help of a polarizer. for instance. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. AW = £ fc=l. which leads to an outgoing wave which we give the label "1". (9.2) the vector potential A is completely determined by its two (say (x. i. (9. . Here we are interested in observable properties of light waves — like.k (910) We give the initial wave the label "0" and subject this to an experiment.A«).e. I=<l2x2>:=E40)*l*40)i. we recall. (9.y)) components A i . a calcite crystal or a film of nitrocellulose.A2 is the sum of the moduli. (A(°). 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. Observable quantities can only be those which are invariant under translations and rotations.t in which k and ui remain unchanged.A 2 .A(°)). From the vectors A(°).A i = 0 = k . the fields E and B are observables. We set therefore A 0 = Ai + A 2 . i.AW).g.9.e. as e. the vector potential A. the intensity of a wave. (AW.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. however.2 *i*4 0) (98) with coefficients F^. the direction of polarization. represented by a filter "F". i. 17. The intensity 7 of a light wave with amplitude coefficients A\' . (9. Then their connection is given by Eq. See for instance A. p.7) We consider now measurements with polarized light.11) 'This means we consider the polarization.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9. the connection must be linear. In classical electrodynamics. with k . not.8). A2.
To this end we observe that as a consequence of Eq.12) i.13) the quantity F must be an hermitian matrix and thus representative of an observable.k i (9.e.A*FikAk. if (F) is an observable quantity. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.16) i.e. (9.17) . i. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment.A*kF£A* = E and hence F = Fl A p i tkAk. We write the observable quantity as (F):=Y. (9.k (9. (9. (F)=Tr(Fp). i. the connection between the initial polarization and the final polarization.164 CHAPTER 9. Thus.k l f c 4 ° \ (9. (9.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%.13): j2^FikAk = Y^AiF*kAi = Y.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik).e. Thus (F) = J2FikPki = ^(Fp)a. (A(°). (9.15) in order to rewrite Eq. it must be real and hence (F) = <F>*. i. We use definition (9. (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments. Then according to Eq. pki = AkA*.A( 1 )) = ^ ^ F i. so that (F) can be looked at as the expectation value of an observable F.13).
is given by a single polarized wave.19). . The socalled pure case.d "»=(o ! ) " In the case of an admixture with equal portions of 50%.18) The vector A describes a particular ray of light. But light. b = 0 and o = 0. 165 (9.15) with A\ = a.21b) This state is still a pure state. "incoherently.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). like that of a bulb. i.22) For a — 1. we have (9 19) ' In this case there is no preferential direction.(AAt) = ( £ £ ) .* For a lightwave travelling in the direction of z. For a 45° polarized wave we have 1 . b = 1 we obtain the expressions (9. <»^> (9. i.e. A2 = b as </*) .9. i.21a) follows from Eq. 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. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. that of a pure state. mutually orthogonal polarization directions x and y described by the matrices (see also below) * . it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. In the present case of a polarized beam of light we have two possible. This admixture is described by introducing matrices p. (9. is in general unpolarized. (9.e. The density matrix for the pure state (9.( J o ) .e.
Equation (9.3 Schrodinger and Heisenberg Pictures As in Chapter 2 we now consider on the purely classical basis the (analogues of the) Schrodinger and Heisenberg picture descriptions.166 and CHAPTER 9. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.Z^ 3. l l\ " i °\ ) ' ^925^ 9. We set in analogy to Eq.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. Then according to Eq.™ V 0 n kmAm / ^^jPim^mk' . 1 = . in which the intensity I = Tr(p) is not conserved (e. as a consequence of absorption). (9. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . (9. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics.15) Pik — A i A k . (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".23) 1 . (9. b = —=.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.8) A(1)=E^40)> k R = R(<*). Pi35° = \ \ 2 ? )• (9.
9.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation. i. the observable is transformed.e. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .34) .31). the observable F remaining unchanged. in the state functional) in analogy to the time dependence of W(p. here this contains the dependence on a. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. (9.33) i.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.e. (9.t) in Chapter 2 in what we described as the Schrodinger picture there. i. and p(0' remains unchanged.q.4 The Liouville Equation 167 pW = RpMRl (9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations.pW). We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)). in other words without creation or annihilation of systems.30) Ti{FRp{®I$) = Tr(F.e.29) where (i. In the case of Eq. the dependence on a is contained in F' = R^(a)FR{a).31) (9. The corresponding condition here is that no absorption of the wave takes place. (9.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. (9.9.e. In the case of Eq. (9.28) (9. that the initial intensity Jo is equal to the final state intensity I\.32) (9. (9. on the other hand.
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.p(°\ao)]da.35) Thus the matrix R has to be unitary.37) (9. (9.F}. (9. We set therefore M^:=iH. i.36) (9.40) Equations (9.B]. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.2) has to be taken into account.B}:=i[A. (9. Eq. j£ = i[H. (9.p].40) can be considered as quasiequations of motion.21) and (2. CHAPTER 9. the matrix M is antiHermitian: Mf = .32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.39) If the dependence on a is not contained in p^1' but in F' (cf. (9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.39) and (9. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.M . In view of Eq.e. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. = Tr(R^Rp^). (9. (9.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.e. Comparison of these equations with Eqs.46) suggests to define Poisson brackets of matrices by the correspondence {A. . (2.168 i. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9.
Heisenberg and interaction pictures. To insure that the 169 .1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general.Chapter 10 Quantum Theory: The General Formalism 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.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. Thus in particular we establish the uncertainty relation for observables in general. (10. and consider timedependent perturbation theory. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u). 10. the Schrodinger. we establish their Heisenberg equations of motion.
however. q). z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. (10.Qj] = 0. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i.170 CHAPTER 10. \pi. and assume that A = A(p.2. such as for instance of the angular momentum operators Li.Pj] = ihSlj.y. The phase space coordinates qi.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.Pj] = 0.2) (F(A)) = {u\u) <"lf. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1.q). [qi. E. 10.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. We can describe a system as one with a classical analogue if the Hamilton operator of the quantum system is obtained from the Hamilton function of classical mechanics by the correspondence •^i • Qi. (10. where i = 1. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. .g. we assume this now. Pi • Pi. classically qiPi = PiQi.2. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. Here. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.1 U n c e r t a i n t y relation for observables A.4) *Even in those cases where this is possible the operator correspondence may not be unique.3 represent the three Cartesian coordinate directions of x.* a conspicuous exception being for instance a spin system. which clearly does not hold in the case of operators. y. Such a correspondence with classical mechanics is not always possible.Pi are called fundamental observables. Naturally one can also write (10. we must have (u\u) = 1.B] = ih.
(10.11) (A2) = (A2) . Ai = = = ((A2) . (3.0)1/2 = AA.13) AB = AB = (B2)1/2.(A)2 = (AA)2.10) . using Eq.7b). Then [A.{A)2)1'2 {{A2) .. (10.6) B:=B(B).12) the Schwarz inequality (cf. (10. (10.B\u) G IK as vectors) we obtain (for an application see Example 10. Applying to the expression (10. (4.2(A)2 . we have A2 = A2 ..5)) (for A\u).2A(A) + (A)2. AB defined by the mean square deviation AC=((C2)(C)2)1/2. so that (A) = (A) .8) and. Since A = A— (A).2A(A)}  {A)2)1'2 (10. B}= ih..12) where \u) 6 "K is a ketvector representing the state of the system. Thus AA = AA = ( i 2 ) V 2 .(A)2)1'2 (1 °= a) ((A2 + (A)2 .9a)..7a) (B) = 0.10. = \\A\u)\\2\\B\u)\\2 (10.(A) = 0. C = A.2 States and Observables then their uncertainties AA.9) ((A2) + (A)2 . 111 (10.1) (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). Eqs. (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u).B. (A{A))(B{B)){B{B))(A(A)) (10.B] = ABBA > ^h. (10. (10.7b) = = [A.
for a complex constant c. (AB + BA) This means (AB + BA) is real.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). A\u)=cB\u). (10. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. . The condition (a) is satisfied when. (10.17) as had to be shown. since A and B are hermitian. and hence for condition (b) 0 = = tSee Sec. 4.13) (AA)2{AB)2 i.3.172 CHAPTER 10. and (b) when (AB + BA) = 0.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*.15) as (u\AB\u) = R + iI. (10. (10. Zi + BA)\ + ^ih.18) >R2 + I2. We can therefore rewrite Eq. (10. Zi Zi (10. and with Eq. AAAB>I=h. The equaltosign applies in the case when (a) it applies in the Schwarz inequality. (10.e.16) I = h.4). Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. From this we obtain (uiJBu) 2 = i22 + / 2 . R= (AB + BA).14) Hence (u\AB\u) = /^(AB where. (10. Zi (10.
(10. Ly are not "sharp". m\lm)). Example 10. = 0. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. What does the relation imply for states u) = \lm). such that Lz is "sharp".z. for which one component of L is "sharp"? Solution: Applying Eq.e.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li.i relation holds: (ALi) 2 (AL.13) to operators Li — Li.) 2 > = x. {u\LiLj\u) = R + iQ. 9ftc = 0.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. i. + LjLi\u).e. Thus in this case Lx. (10. For these we have in the onedimensional case the relation [p. For states \u). h2(Lz)2. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0. The observables like angular momentum are functions of p and q. i. 9 = h{u\eiikLk\u). ALy = (u\L±\u) ^ 0.e. in view of Eq. Separating the product into symmetric and antisymmetric parts. . (A{A))\u)=i(Zc)(B(B))\u).10. (ALy)2{ALz)2 > ~h2(Lx)2.7a). ALX 10.y.e. = 5R2 + G 2 > 9 2 = hi2(Lk)2. q) = ih. implying {Lx)=0. (Lx±iLy) ^ 0. i. the following uncertainty h2{Lk)2.Lj] = (LiLj + LjLi) + ih£ijkLk. (Ly)=0. (Lz)2^0. i. (Ly)2=0.20) or. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.3 OneDimensional Systems 173 Thus c = — c*. (ALZ)2(ALX)2 > \h2 = (Ly)2.
Using [q. . and (c) the spectrum is not degenerate.p] = ^ihp ~ Proceeding in this way we obtain [q. Eq. i.qn] = ih^(qn).23) = ihg(q2) (10. Q2] = [p. Then (see below) we can derive for A the following commutator equations [q. q]q + # > v] = ~2ihq and \p.i h ^ ^ . We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). as remarked earlier.pn]=ih^(pn).p)] = .p2] = [q.BC] = [A. Similarly we obtain \p. they do not have a common system of eigenvectors.p}p + p[q.34d)) [A.e. i.21a) Equations (10.B]C + B[A.174 CHAPTER 10.p)]=ih^^and \p.22) d ih—(p2). Next we have a closer look at the operator q.A(q. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).A(q.21b) (10. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable.e. (10. (b) its spectrum is necessarily continuous.21a) and (10.p) can be expanded as a power series in q and p. A — A(p.q). We obtain these equations as follows. (10. (3.p] = ih we obtain therefore [q. This is why the operator p requires a representation in the space of eigenvectors of q.21b) now follow with the assumption that A(q. We have (cf.C\.
Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.28) i. 10. U] = ih^= aU. and a a cnumber.e. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). = eipa/h = Uia)'1 = U{a). an operator.10.27) (10.f^ ^ J—oo dke ikx —ikx' (10. 10. (10. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.26) it Here p is an observable. > We begin with (a).1 The delta function for K — oo.21a) A by U.25) It is shown in Fig.3 OneDimensional Systems 6K(x) 175 Fig. (10.e. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. we obtain [q. .24) lim SK(x — x') — S(x — x'). U is unitary.1 how the delta function arises in the limit K —> oo. i. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. Replacing in Eq.
that with U(a) = e~ipa/h = U. q any other observable F(p. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).29) This means: U\x) is eigenvector of q with eigenvalue x + a. this means: With a unitary transformation (which leaves the matrix elements unaffected * and hence the physically observable quantities) one can pass to any arbitrary eigenvalue in the domain (—00.00). (10. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator. the spectrum is not degenerate.00). Thus all these values belong to the spectrum of the operator q. 10.1 T h e translation operator U(a) We saw above.q)U^U\ilj).3. we can see that also p possesses only a continuous spectrum (from —00 to 00).176 i.e.q)\i>) = (i>\U^UA(p.31) . or qU\x) = U{q + a)\x) = (x + a)U\x). q) can be constructed representing a dynamical variable. or qU(a)\x') = (x' + a)U(a)\x'). cf. ^Observe that {iji\A{p. (10. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). Eq. which means that the spectrum of this operator is a continuum. CHAPTER 10. Evidently every eigenvalue has only one eigenvector. Since this holds for any arbitrary value of a in (—00.29).e. (10. b a cnumber). i. WU = 1.30) (10. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). 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.
38) h 6(x' . since q\x') — x'\x'). i.34) {x"\U\a)U{a)\x') c*(a.x")c(a.32) Since U is unitary.35) (10. (10. (10.x" . i. (10.33) (10. for instance.e. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). we have q\x' + a) = (x' + a)\x' + a).a).x") U{x'\p\x" % pz.3 OneDimensional Systems 177 Moreover.e. \x') = U(x')\0).x')5(x" = (x" + a\c*c\x' + a) x'). i.e. We choose the phase such that U(a)\0) = \a). the operator U(a) acts to shift the value x' by the amount a.e.e) ~ 5{x' .5(x> .37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . U(a)\x') = c\x' + a). (10.x" .x') = = i.x" .hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). The operator is therefore called translation operator.32) we see that \x' + a)ocU(a)\x'). (10. offdiagonal elelemts of the {x}representation of the momentum operator. (10. In the expression of Eq. it follows that 5{x" .37) With the help of the latter expression we can calculate. (x'\p\x").x')\ = l.e) (x \p\x ) = . It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .x") .36) i.39) .e. \c{a.31) and (10. (10.10. Comparing Eqs./ (10.
to) is called the time development operator. since 5(x) = .4.tQ) = eiH(tt<Mh.27) and (5.4i) We assume first of all.t0). (10.40) 10. . In this case the classical Hamilton function does not depend explicitly on t. For the way the state of the system develops in the course of time we make the following postulate: Postulate: A linear operator U(t. Quantum Theory: The General Formalism We mention in passing that.L fdpeipx. Differentiating U(t. i.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. it follows that 5'(x) is an odd function of x.45) (10. isolated from any measuring apparatus which would disturb the system and hence its behaviour in the course of time) and let \tp(t)) be its state vector at a later time t > to (with no interference in between). 6'{x) = !.43) (10. The operator U(t. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E. 5. H\uE(to))=E\uE(to)).4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. (5. This is in conformity with our earlier postulate in Sec. (io. i.° ° d(p) (ip) e"« = S'(x).e.44) (10. See Eqs.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .t0) = HU(t.e.to) exists which is such that^ m)) = u(t. a conservative system.e. so that U{t.35).178 CHAPTER 10.f dpipeipx. . to) with respect to t we obtain ihjtU(t.to)m0)).x") = (x'\p\x")*. We introduce another postulate: Postulate: The time dependence of the vector \uE(t)) is given by the relation \uE{t)) = eiH^t0^h\uB{tQ)). (10.
Ufa.t) = 1.z / dt'H(t')U(t'.tiM*i.4 Equations of Motion 179 We now demonstrate that Eq.*l)^(*l)> = I7(*2.49) .t0)U(t0.10.t0)\^(t0)). i. (10.to) can be expressed as an integral as the solution of Eq.t)l]U(t.to). t0) = E/(t 2 .to) (differentiating we obtain immediately Eq. (10. t0) = H(t)U{t.t)U(t. Evidently U(t.41) we obtain ^(t 2 )> = = J7(t2. = U(t0.*l)^(*l.t 0 )^(*0)> U(t2. (10. From Eq. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.t). (10.e. (10.48): [/(Mo) = 1 .t) and therefore [U(t.t) = l^eH{t). Then with U(t + e.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. (10.t0) or ih—U{t.45) has very general validity. But obviously (cf.48)).44)) we have U(t.t0) [U(t + e.to) = 1.47) e.t0)]1 Let us set for small values of e: U(t + e. The operator H is d efined by this relation. t0) (10. Eq.48) with U(to. (10. Hence U(t.t) = l.
Another such description is the Heisenberg picture.41).53) . t)\t/>(t)) =U~ and since H is hermitian.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^).(£ + dt)) = U(t + dt. Since the only measurable quantities are moduli of scalar products. Eq. i. (io. 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. (10. it follows that U(t + dt. and these remain unchanged under unitary transformations.30).e. 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. (5.180 CHAPTER 10. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations.51) Since </. 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. the Schrodinger equation ju(t. the instant at which a measurement is performed on the system. Now let \tp(t)) be the state of the sytem at time t. The eigenvectors of these observables are also constant in time.5o) ihjtm)) = Hm)). Then \Mt)) = u(t. (10.t0) m0)). whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.t0)\ii>s{to)) (10.e. Comparison with the equation postulated earlier. >(*)> = i. The operator U(t.t) = l % ^Hdt) \il>(t)) Hdt (10. the predictions of different descriptions are identical. requires the identification H(t) = H= Hamilton operator.
. (10.e.HH]+ihU^U. we consider only cases without explicit time dependence of As. However the associated observables are timedependent.57) and rf[As. As(q. 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.p) in the Schrodinger picture dAs/dt = 0.10. Thus.55) and using Eq. (10. subject to a continuous change. but here. for instance in cases of nonequilibrium or if part of a system disappears through absorption.55) Thus AH is explicitly timedependent.59) This equation is called Heisenberg equation of motion.H]U = = = U]ASHU AHHH — U^HASU . Differentiating Eq. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.54) Observables transform correspondingly with a similarity transformation. even if As is not timedependent.58) HHAH (U*ASU)(U^HU) [AH. (10.(tf HU)(rfASU) (10.e.H]U + iW]^U. in order to obtain the Heisenberg picture (description).4 Equations of Motion and \^H) = U\t. = so that ihjtAH or ihjtAH = = [AH.HH]. Here H is the Hamilton operator in the Schrodinger picture.p. as a rule. 181 (10. so that a matrix element (i. t0)AsU(t. (10. It does not always have to be the case that As does not contain an explicit time dependence. {AH. (10. For As(q.HH].48).56) U^[As.i). i. a scalar product) remains unchanged: AH(t) = U\t. we have to take into account the contribution dAs/dt.t0)\ips(t)) = \ips(to)) = constant in time.With explicit time dependence.e. In the Heisenberg picture it is HH = U]HU. if ^ = 0. i. t0).
Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. We also observe that the momentum pi is conserved when \pi. In many cases the Schrodinger picture is more amenable for explicit calculations.21a) and (10. . Finally we add a comment concerning conserved quantities.21b). 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.e.13) to derive the uncertainty relation for energy and time. This holds also in the case of the Schrodinger picture. In our treatment of representations. 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.HH}=0. A. [CH. since in the transition from one to the other the commutator remains unchanged.H] = 0. For an extensive discussion see B.61) Thus conserved quantities commute with the Hamilton operator. An observable CH is called a conserved quantity. (10. In the treatment of the pictures we used unitary transformations of operators (and vectors). Orfanopoulos [224]. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. Example 10.Pi): and where the expressions on the right follow from Eqs. in the case of the fundamental dynamical variables qi.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. We end with a word of caution.Pi we have in the Heisenberg picture (replacing in the above AH by qi. i. we used in essence unitary transformations of matrices. We see that formally one obtains Hamilton's equations of classical mechanics.182 CHAPTER 10. if ffCH = 0. provided the Poisson brackets in the latter are replaced by commutators. (10. In particular. Solution: We have for an observable A (A) = (i>H\AH\TpH).
)\2 ( = \{i. this is practically the same as at time t for times t with \t — t'\ < T.15): (AAf(AH)2 Using Eq. a position or momentum observation is independent of t. (10. H\tl>) = {H(H))\f).68) To put it The eigenvalues a of A (i.H]) = 0. (10. of A\<t>) = a<£)) are then called "good quantum numbers".H}\f)\2.13) and (10. > (10.e. (10. We construct the following vectors using Eq. in a different form: The eigenvalues of an observable are called "good" if [A. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .63) so that with Eqs. This means that for a physical variable A(q.66) This means. i. is displaced by A A in the interval A T = TA. 2 d{A)/dt " and. and TA O C (10. (10.65) and rA AA d{A)/dt' . (10. Let \ip) = \"4>H) represent the state of the system. the "centre of mass" of the statistical distribution of A.HH]) Hence we obtain by ih(A). if we set AH = AE we obtain TAAE (10.e.10. we can replace ([AH. then if a measurement is made at time t'. AW = {A(A»V>. since \IPH) is timeindependent. (10.Such a characteristic time interval can be defined for any dynamical variable. at ^±AH>h.e.\HAM\2) \(nA.p) we then have 0=^M) = U[A.64) = 0.2 > h. .4 Equations of Motion Then. dt in. (A). If r is the smallest such characteristic interval of time. H] = 0.62) and dAs/dt > \WAH\i.e. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t.7a). i.67) 1 d(A)/dt is oo and AE = 0. 183 dV ' \ dt >"•"»» + ( £ ) • (10.62) The Hamiltonian H does not depend on time. i.
have around EQ the form ipEofct) = i. But the number of radioactive (thus excited) nuclei decreases exponentially.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. i.70) by From Eqs.69): oo poo p / oo *PEo(x. the particle density diminishes in the course of time (t > 0).70) we can determine the function fEo{E) integration with respect to t. the wave function ipEo(x.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . for instance. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. (10. i ) = J dE'^El^)eiEltlhfEo{E'). Recall for instance radioactive decay. (x)/ E o (E')5(E .t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x . First we obtain from Eq.)exp(iEt/h)). (10. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei.H ) ( £ . t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. (10. the wave function ^ ( x . states with sharp energy (AE = 0).. This means. (10.70) we obtain: oo / oo TPEO (x.E1) 2irhfEo{E)4>E(x).Eo(x)eiEot/he'^2h9(t).(x)fEo(E')e«EE'Wh 2irh f dE'^E. Quantum Theory: The General Formalism 10. This means that in the case of decaying particles. as is observed. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. /oo (10.e.72) = V'EoW 7/2h + 0i(EE )/h  oo .t) — ipE(x.71) From Eq. (10. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".tyEt/hdt = = = / J—oo dt J dEJrl>E.69) and (10.184 CHAPTER 10. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.
whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. every unitary transformation of states and operators defines a possible "picture"). The state with lifetime r — h/'y is not a discrete state.e.e. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula.t0). ihU(t.74) Thus the solution of this equation is the main problem.e. £/ = £/W(Mo)^'(Mo). of course. 7 > 0. (C/(0) unitary).6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. W)) = U(t. We saw previously that if the state of a system is known at time to. (10.g. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x).e. i. Let U^°\t. and is called "interaction picture" or "Dime picture". i. ^(*o))> then \xp(t)) for t > t0 follows from Eq. (10. (14. that** EQ. to) be an approximate but unitary solution of this equation. They are called "resonance states".75) (10. 185 i. Considered as a function of E.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). (10.e. where U is the solution of Eq. (10.105) and (20.toM{to)).41).10. 10. i.45).to) = HU(t. i. We obtained the result (10.11). Eqs.45) can be solved only approximately (in actual fact. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state.76) **For specific applications see e. This description contains effectively parts of the other two pictures and is particularly useful when Eq. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. . (10.
186 Then CHAPTER 10. .82) (10. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.e. varies only slowly with t).e. i. with Hamiltonian H = H^ + H'.81) But [/(°) is unitary.83) ih^U^ ishermitian. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. £/(°)£/(°)t = 1. (10. with Eq.e. i. H(°Ht) = ih^uW.78) (10.h^U> = at ( io. (10. [/' has only a weak ^dependence (i.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. (10.e. = HU<®U'. Quantum Theory: The General Formalism ih^(uWu') i. i.e.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).e.e. Now let H^(t).t0) 0 = l.79) ^'~°i. In addition we have (10.
90) = U^H'\^s(t)).85) = where we used Eq.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. U<®\Mt)) = \Mt)).e.78) at (10 85) (10. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. (10. (10. (10. (10. lfe(*)}.85) multiplied by U') on the right of Eq. 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^.80). (10. We also set H^U^H'UW.10. U^H'U^U' (10. and obtain from Eq. i.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. (10. (10. (10. (10.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). H = H{®{t) + H'.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).U^HU^U' H'fi'.89) where ^s(i)) and As are state and observable in the Schrodinger picture. (10.84) U^ where H' is also hermitian. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.92) .78).86) at = U^HU^U'.
93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. (10. Since H'j is assumed to be small (i. (10.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. i.e.94) _Hf)Al 1 + lhdAl+AlHf). the vector varies with time (different from the Heisenberg picture).92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .H\0)}.94) Then we have (normally with dAs/dt dt (io=8o) — 0. cf.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).e. comments after Eq.55)) at + imm<Ms_u(o) at dt _umH(. (10.89). (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)).e. (10. a perturbation so that H ~ H(°)). dAs/dt=0) . i. (10.89)^0. A^t) = U^AsUi0). (10. near zero and \tpi(t)) is almost constant in time. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. i. we differentiate with respect to time the operator defined by Eq. ot i. (10.e.90) and replacing the left hand side of this equation by the right hand side of Eq. (with ihjtAI(t) = [AI.60b) with iff0) instead of H.e. ^ (0) l^(*)>.e. we have i. (10.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10. so that with Eq. In order to obtain the equation of motion of an interaction picture observable Ai.60a) or (10.188 CHAPTER 10. H'j = UWH'UW.
the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. We obtain the operator U^ from Eq.94) Hi = uW(t.100) (10.80) as E/(°)(t. (10. H' = H'(t). Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).93).10. Since . ^ \<Pn){Vn\ = 1n (10. ih±\Mt))=Hi{t)\Mt)). (10.102) We assume that before the perturbation H' is switched on at time to.99) (1098) Instead of the original Schrodinger equation we now solve Eq. where according to (10.e. i.87) and (10. (10.7 TimeDependent Perturbation Theory 189 10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. (10.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. i.t0)H'uW(t.e.t0). H0\ipn) = En\(fn). (10.98).e. i. We assume again that the spectrum and the eigenvectors of Ho are known. {ipn\<Pm) = Snm. (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.i0) = e<Ho(tto)/ft. We have the Hamiltonian H = H0 + H'.97) The equation to solve is Eq.
102) and truncate the series after the first perturbation contribution. 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. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„. (10.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).104 ) Here we insert Eq.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. (10.106) We insert this expression into Eq.^ V ™ ) . (10.108) . the time dependence of \ips } can be separated as for stationary states.103) for m — n and use Eq.190 CHAPTER 10. Hence we set ^> = £°«(*)l4 0 ) >n. n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon of the perturbation H'.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 . assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^).107) Jto With this we obtain from Eq. (10. we can write l 4 0 ) ) m = elHot/hWm) =e . (10. Quantum Theory: The General Formalism HQ is independent of time.e. (10. i. (10.
if in the scattering of a particle off some target its momentum p changes to p'. We assume that the perturbation is switched on at some time to.109) 10.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . where the momenta of the aparticles form a continuum. A further example is provided by adecay.109) the transition probability Wmn(t) = jp Jo .En (10. t o = 0. for instance.112) for a^0. (10. (10. or if by emission of a photon with continuously variable momentum an atom passes from one state into another. < KO?t .10. (10. » — < (10.111) Different from Eq.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. Thus we set H'(t) = H'0(t).Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. (10. This is the case.110) With this we obtain from Eq. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.Em)t' (<Pn\H'(t')\pm) = Wnm(t).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^.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .Em)t} \{tfn\H'\ipv h 2h(En . but that otherwise the perturbation is timeindependent.
Em).ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.— = o (En 2h En 2h5{En .Em)t} r i 7 TTT.192 CHAPTER 10. we would have En — Em ^ 0. so t h a t Wnm would be zero.E7m)t} = (10. though sufficiently small. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . .io. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. > 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. (10. and no other state would be nearby.2. 10.2 The delta function for £ — oo. Hence lim St(a) = 5(a).114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. 10. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. Otherwise. Fig. if the energies belonged to the discrete spectrum.
118) where 8e is the separation of states around En which in the case of the continuum goes to zero.116) The transition rate V is the transition probability per unit time.* In view of the usefulness of Eq. Schwabl [246]. Example 10. i.113). pp. sin 2 (tx) ax i—( * According to F. If we are concerned with pure scattering. so that initial and final states belong to the continuum. for instance from a potential. f See E. then Eq.r Z x 1 ^ ) = 4>{Q).8 Transitions into the Continuum 193 as equal. p.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . In the case of the functional (10. (10. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. Let p(En)dEn be the number of states in the interval dEn. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. in the article of W. Fermi gave this formula the name "golden rule" .113) this is* hm . Pauli [227].10. 142.117) provides the outgoing particle current (see Example 10. (10. Fermi [92].3: Application of test functions to formula (10.113) Use the method of test functions to verify in the sense of distribution theory the result (10. We evaluate the integral 0(0) = 1. (10. 266.117) in many applications. (10. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. if p(En) varies only weakly with En. . F Jo .4). Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). 75.115) and (10. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e .117) were derived by Pauli in 1928. Then. _ 2irh AE > > Se.W .e./ t>oc TZ J_00 dx .
./ cfc a n p 2 +fc 2 .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<).b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . . . * ^ = . Then ^ .194 CHAPTER 10.e " 1 * 1 = 2 lim — . 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). K = k 0 . with n .a 2 ' i that for a = b = t . S.4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. 491. The ingoing particle flux is therefore v / L 3 .. E = h2k2/2fi. .117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n .<p). ... ^o n . 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. Jin § m = See I. with v the velocity of the particles. Quantum Theory: The General Formalism with the help of Tables of Integrals.. = ( — ) dk= 2ir (—  2irJ k2dkdfl. = 0 . p = 1 we obtain I = . x . formula 3. iknx . p. 77^ + .„I. 2 . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . . and is obtained as ih.( ^ V V • < ih 2 _ i k n .^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . E x a m p l e 10.947. . Hence the result is 0(0) as expected. Then p(Ek)dEk Since Ek — h2k2/2[i. . § We have r Jo \p\x dx sin ax sin bx  V. 2 T / ^ M. and hence . Gradshteyn and I. . .? 2TT / L \3k2dkdUn 1 /V(x). V +(ab)2 a _x 2p6 7 m —^ '. M. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. Ryzhik.k. 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.ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.tan p 2 + a 2 . dEk = h2kdk//j. L \3k2dkdUn It follows that T. . 1 .
(10. see the discussion after Eq.t)+ / n a{E. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>).t) = ^On(t)^(x. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. For a possible application of this result. Thus we set ^(x.10.x) = S(EE'). specifically in the case of the Coulomb potential./continuum = <San(£)</4°)(x.69).122a) (10. 193. 1 rfx^°)*(x)^)(x) = 8mn. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx. (21. pp.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn). We expand the wave function V( x .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.121) .120) In the case of Eq. (10.x)^(E'.122b) S e e also L. 10. 148. by which we mean without leaving the Schrodinger picture. Schiff [243].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. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known.t)^°\E^t)dE (10.^ We start from the Hamiltonian H = HQ + H'. (10. we present another derivation using the usual method of timedependent perturbation theory. (10.9 General TimeDependent Method For a better understanding of the "golden rule" derived above.^°>(E. .t).
.122a) and (10. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). Observe that these conditions do not contradict Eqs.128) . into dip (10. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. (10. dnptMrnW = { ^ _ En) (10.x.x.127) i n c a g e Qf c o n t i n u o u g £.t) = eiEtlhy(E... / *. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10. i.E').i) = e. (10. f 6kn in case of discrete tp's.„.t)rJ><®(E'.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. .126) + H')<pn^)eiE^h}..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.„„.129) . (10. The second term on the left of this equation and the first term on the right cancel out.120).122b). .iJ5 »*/Vn(x). Next we use the orthogonality relations . ^0)(E. (10. .x.123) We now insert these expressions into Eq. (10.x).t)dx. (10.196 CHAPTER 10.125) where an(t) — dan(t)/dt. t) = San(t)<pn(x)eiE^h... (10.127) we obtain from Eq. V>(x. = e^EE'^h5(E Thus with Eq. We multiply this equation from the left by </?£(x) and integrate over all space. since f rJ>(°>(E. .E') = 5{E .e.
134) . Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. let's say am / 0. Comparing coefficients of the same powers of A on both sides.130) This result should be compared with the Schrodinger equation (10. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. In the first place we replace in Eq. SH'kn(ty^E^l^\t) (10.130) H'kn by \H'kn and at the end we allow A to tend to 1. 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. i.10. 0. (10. since it is equivalent to the latter. (10.130) along with the parameter A in front of H'kn. (10. i.120).9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. they are fixed by the initial conditions. (10. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. SH'kn(t)e^~E^ha^(t). We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. We restrict ourselves here to the two lowest order equations. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• .132) These equations can be integrated successively.133) The numbers ak = const. and we set ak°}=5km or ak0) = 8(EkEm).e.131) and insert this into Eq. (10. to perturbation theory in the lowest order.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] . Thus first we have ak0) = const. In order to solve Eq.130) we develop a perturbation theory. determine the state of the system at time t — 0.e. (10.
137) . (10.132). we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'.ml2sin2{^frn (En . so that ihakl\t = oo) = 0. (10. (10. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). In this case we obtain from Eq. The amplitude / B o r n is a Fourier transform in spatial coordinates. What is the physical significance of Eq.135) and (10. i.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 > 0.136) we see that both "amplitudes" are Fourier transforms and look similar. ifm{kl) = f J —oo (10. i. ) ' X V(x')dx'. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . 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. Equation (10. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. (10.198 CHAPTER 10. ip) in the context of timeindependent perturbation theory.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.111).Emy in agreement with Eq. ipm) into the state k. (10.k .e. Integrating the second of the equations (10.e. (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.136) Comparing Eqs.135) We put the additional constant of integration equal to zero.e.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. H'^y^"^^/hd£.
r=  r . such as the electromagnetic field (the quantization in these contexts is often called "second quantization". the position coordinate and the mass of particle i with i — 1. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1.2.1 Introductory Remarks The quasiparticle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e.i be respectively the momentum. 11.y. ( 1L1 ) We let pi. The Hamilton operator is then given by w= jt + ji_^M. Let r : = (x.2 Separation of Variables. Z<i = Z) y(r) = _ M ^ . Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory. meaning the inclusion of the creation and annihilation of field quanta). Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.m. Both of these important potentials are singled out from a large number of cases by the fact that in their case the Schrodinger equation can be solved explicitly and completely in closed form. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.Ti.Chapter 11 The Coulomb Interaction 11. 199 (1L2) .
_d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc.5) The momenta P . y.200 CHAPTER 11. (11. P = pi + p 2 . so that as expected according to Eq. so that as expected we obtain the expression for p in Eq. .3) for the individual position coordinates we obtain ri=R+ •^ —r.2/ mi+m2 The mass mo is usually described as the reduced mass. mi + m 2 va\ (11.4). Z) and r = (x. 2Pip2 m\m2 pip2 mi + m 2 .3) and centre of mass momentum P and the relative velocity p/mo by setting r» . p are such that as operators in the position space representation or {R. Y. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. (11. where m 0 = mim 2 . r} representation they have the differential operator forms Setting R = (X. (11. r = rir2 mi + m 2 (11. mi + m 2 iTi'2 r2 = R ^ —r. (H4) \mi m.U7&. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. /'Pi p = m0 P2 A v. Solving Eqs. Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' .4) P = p i + p 2 .
13) * and = M ' 1V1 P = 0. P on the one hand and those in r. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. [Ri. Angular so that Momentum 201 ^ ^ . (11.r2) to (R. p = W. Pj (which in the following we write again ri.= ^.e. r = —. (1114) m0 We observe that the equations of motion in R.11.pj. The canonical quantization must always be performed in Cartesian coordinates. p on the other are completely separated.z). for which we postulate the commutator relations* [ri. (11.8) and so W =777 v + 7 T .Ri.j = x. (11. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii.Pj). pj. with M := mi + vn^R = ^ P.+ P. we mean n = rx s x.Pj} = ih6ij.pj]=iMij. etc. r) is a canonical transformation.12) (11.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. (11.2 It follows that p2 2 P2 n2 n2 n2 (H. Hence Hamilton's equations apply and we obtain.y. (11.il) ""Op' i. . The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.2 Separation of Variables. Ri. (i.+ ^(r) 2(mi + m 2 ) 2m 0 (H.pj.15) In accordance with Eq. 2(mi + m<i) 2mo 2m\ 2m. The equations of the relative motion have the form of the equations of a single particle of mass mo moving in a potential V(r).Ri.10) we set H = Hcm + H.16) Although we write ri.+ ^2.10) One can convince oneself that the transformation from (ri. Pj over to operators ?i.
Thus the twoparticle problem is practically reduced to an effective oneparticle problem.23) y — r sin 9 sin tp. = rcos9. —n < tp < ir. i. it is reasonable to go to spherical coordinates r.Q 2M' + V(r).22) If. (11. 9. r ) = J E t o t a i*(R.21a) L*+v{r)] ip(r) = Ei/j(r) (11. (11. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .21b) + E.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. 0 < 9 < ir. ip: x z = r sin 9 cos ip. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).20) r2MAR_ $(R) = ER*(R). where Hcm$(R) = and Hip(r) = with ^total — ER (11.24) . (11.17) (11.r).202 CHAPTER 11. The Coulomb Interaction nc H = 2m. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. 2mo (11. 0 < r < oo. V(r) = V(r).r) = *(R)V(r).19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R.e. (11.
26) However.11.pr]=ih. 2 1 9 / 1 9 . (11. (11. that the above expression follows from l:=rxp=ift(rxV).e.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.2.27) We can now write (see verification below) p V = h2Ar^ where 2.29) . where (r x p)j = ^eijkrjpk. We have I2 = (r x p) • (r x p).31) . (11. i( i.2.5r = . = —1. = 0.)In Example 11. otherwise.2 Separation of Variables. is the operator of the relative angular momentum.32) (11. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r.2.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. (11. (11. = (p2 + ^ V (r + 0).k a clockwise permutation of 1.3.12 — I2. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. with tijk — + 1 . _ h2 \ .3.j.i ' . k an anticlockwise permutation of 1.j. i.33) i. .1 it is shown that pr is hermitian provided lim n/>(r) = 0.28) . k = 1._0di})\ x We now demonstrate that 1. n ( . if i. j . (l + . i—>0 (1L25) (11.3.
k ~ rjpkrkPj) (11. (11.k Using Eq.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .(r • p ) 2 + 3ih(r • p) . * ' S Since r =xz we have (cf. (11. This means that l2 = so that 9 9 1 (11.35) j. Eq.39) + ih).38) + y* + z . (11.40) (11. ih6jkrjpk (11.34) W i t h this we obtain — keeping in mind the ordering of r and p — I 2 = 5 Z eijkrjPkeij'k>rj>Pk' = ^{rjPkrjPk i.ihr • p r2p2(rp)2 < + ih(rp).j.27) we obtain r{prr)pr = r(rpr . The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.42) (11. (11. (11.ihr • p ) . r— = r • — or or (11.44) .36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .ih)pr and x 2 2 9 9 2 2 Q O (11.41) o = r pr.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.ih5jk)pk (11.37) .43) r2(p2p2).ih) — rpr(rpr With Eq.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. (11.
<p)*?(ri. ip 6 X>Pr C 7i. 9 € [0.<p))ril>(r.<P) = Etl>(r.. Pri/Oi V0.(r.e. Angular Momentum 205 Comparison of Eqs. H 1 2mo pi + K) +V(r).e.Q ' 2mor2 with the condition (11.<p)drdna " r r=0 J \ll§r~(r4'(r.1 Separation of variables We deduce from the expressions of H and l 2 that [H. S o l u t i o n : Set d£2s = sin. reR3.l2} = 0. 6.. 11. i. <p)*Pri>(r.6l. <p)*rip(r} 0. 9. prip) = = I <j>{r. <p) d °° . i.46) (11. <p)* .[ i J With partial integration with respect to r this becomes (<l>. 0. ¥ p)r 2 drdn s = (p r 0.<p) (11. h f l d —(r<l>*(r. r := r : lim rip{r) = o i .d. Then {<j>. r—>0 il>(r. <p)r2dfls = .ip) = (<£.[ 4>(r. (11.<p G [0.e'iP^} .48) . 6.2.d_ r dr the operator pr is hermitian.11.2TT].26).<p))drdns.e lim rip(r) = 0.47) Example 11.! /').2 Separation of Variables. (11.6. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.28) and (11. <p))r2drdns % J r or r<t>(r.e. pr = .1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).Pril>) J / r= dQs rcj>(r.7r]. or J .45) oe Thus in Eq. (pr4>.30) I2 is to be identified with the square of the angular momentum operator.e.— (riP(r.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11. /'(r.6d0dip.e. 8. (11.
i. zpx] + [zpy.2: Verify that [h.53) Either of these operators is the hermitian conjugate of the other. These operators here play a role analogous to that of the quasiparticle operators . ly] = [yPz ~ zpy.50) Thus the components of 1 do not commute pairwise.l2] = 0.e.e. We therefore first search for a complete system of eigenfunctions of l 2 . so that [lz. z we have altogether [lx. z]px + py[z. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.lx] = ihly. 2 2 2 = 0. i. However. 11. for any two of the components there is no common complete system of eigenfunctions. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.l2} and correspondingly [lx.lz]=ihlx. 2 . (11. ?3 = z). (11.49) By cyclic permutation of x.51) (11.lx or l . y.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. g i = x.206 CHAPTER 11. [lz.xpz] = [ypz.ypx) = ihlz. we have [h. (11. [ly. 3 .pz]x = ih(xpy . Example 11. (11. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.ly]=ihlz. /z> *xJ = l <i\}ylx + tx'j/Jj [h. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. they are incompatible variables. xpz] = y\pz. H and l 2 possess a common system of basis eigenvectors.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .52) This means that l . [lyA = 0.3 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. The Coulomb Interaction i. zpx .q2 = V.lz (or l .e. Since 1 = r x p .
* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq.59) = 9) h(l0 r)\lr). Hence we must have Z+I0=0.r) := (Z_)rZ) so that lz\l .62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. With the help of Eq. (11.Y\l) = (lQr)h(lY\l).52): 0 = [ l 2 . assuming a finite dimensional representation space. .Z_] = [ l 2 . y . and with Eq.e. l\l) is eigenvector of lz with eigenvalue (IQ — l)h.r) (11 = {l02)h{l)2\l) (11.3 Representation of Rotation Group 207 A. (11.54) On the other hand we obtain from Eqs.+ ll.lh)\l) = {l0 . We write or define: / .54): lzl\l) = {llz .54) l 2 = \{l+l~ + 11+) + ll = l+l. A^ in the case of the linear harmonic oscillator. Similarly we have lg{l)2\l) and more generally lz(l. In the present context the operators are called shift operators for reasons which will be seen below. [i+M = 2%iz.11. (11.51).).49) we obtain the following relations: [iz. (11. (11.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i).57) i.i+] = m+. (11. [ig. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.61) But lo is by definition the largest eigenvalue of lz.60) Similarly for l+ with the help of Eq. (11. Therefore (lo + l)h cannot be an eigenvalue of lz.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. (n.58) {l0 : max.l)hl\l). (11.i] = m. (11. i + ] = [l2. (11.h~lz(1156) (11. (11. since lz is hermitian.
..67) i (11 60) = But according to Eqs.56) CHAPTER 11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i).Z_] = 0.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.55)) we have 12Z_Z> = l. l\l .n). Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.y\i) l)h2\lr).54) (n = 57) (11..63) + l ) ^ 2 . (11. i. (11.From Eq. The Coulomb Interaction (l+l.64) i.\lr) terminates at some number (let us say) r = n (n a positive integer).e.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l . (11.n)}h2\ln). (11..64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11. i. (11. (11.h)\ln) {(lQnf(lQ. IQ(IQ 1 0 (11. Eq.n) (1 = 60) {l)n+1\l) = 0.e. (11.60) and (11. (cf. (11. i.\ll).208 Thus for l 2 we obtain: V\l) ' = 11.e. [12. \l). l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .l2\l) = l0(l0 + l)h2l\l).68) .66) It then follows from the second of relations (11.e.+ 11+) + l2z\l) = i+i.
11. y/(l0±m + l)(l0^m)h. lQ = .m) with — lo<m<lo2 (11.. The normalization factor in Eq. \l — n) are eigenvectors of lz with eigenvalues loh..1). A.  U o .n)} = (l0 .m) = = = = y/lo(lo + 1) — m(m +l)\l. (11.m).{lo .67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . In the following we consider primarily the case of integral values of IQ... —loh.n) 2 .. we skip this here. § See e..l)h.2. Brink and G. We therefore write the 2/o + 1 orthonormal vectors: \l... (11..1).70) they satisfy the eigenvalue equations (with m = l0Jo l. Satchler [38].l.m)\l.m). According to Eqs. (11.m±l\l±\l. 24.m) = = mh.69) Prom this it follows that IQ must be either half integral or integral.m .m + 1)(Z0 + m)\l.. Messiah [195].m) : Mo).65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. lz\l. \l — n) into one other.. lo = l0n.. R. The operators l+.m / + l)h . We write these* for integral values of IQ: \l. *D. 17.m) l\l. (IQ — l)h. i.n ..l_. Vol.l)h \/(lo + m + l)(l0 .m) = IQ(IO + l)h2\l.3 Representation of Rotation Group so that with Eq.l0): l2Z.m).. The dimension of the representation space is therefore n + 1 = 21Q + 1. where lz\l. II.. M.70) According to Eq. The nonvanishing matrix elements of 1 = (l+. The basis vectors \l). m + l)h.lz transform the vectors \l).m\lz\l. The phases are chosen such that§ l+\l.. (11.e. p.. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.m) = mh\l.m) (l.m y/lo(lo + 1) — (m — l)m\l.m) — mh\l. y (lo .n){l0 .63) and (11. .g. lz) are then: (l. As in the case of the harmonic oscillator we introduce normalized basis vectors.71) can be established in analogy to the method used in the case of the harmonic oscillator. p. (11.Z 0 ).
i.. The Coulomb Interaction Z+Mo> = 0.With the ansatz Y(9. $ oc eim*. as remarked earlier. dp Without prior knowledge about the integral nature of IQ and m we can show that indeed these have to assume the integral values derived above. 11. for the simple "rotor". —lo + 1 . ip) = e(9)$((p). . .<p).<p) = i0{io + lzY£(0.2.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. where IQ = 0.. and m = —IQ.e. We . d2 sin2 9 dp2 _' (11. We now perform the analogous procedure for the angular momentum or.73) 1 d d sm0 06 \Sm9dd) + (11.77) $((p) = m2<b(y). (11. In these coordinates we obtain: h l± so that h2 = xpv . with i2Y£{e.yp.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).l0)=0. . as above. l_\l.1. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip).72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r.76) d2 (11. as this is sometimes described.. mhY£(9. (11.75) the variables contained in the expression (11. /o.. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . 9. .210 Then.<p) = i)n2Y^{e^). ip. CHAPTER 11. (11.4 Angular Momentum:Angular Representation In the case of the harmonic oscillator we arrived at the position or configuration space dependent wave functions by considering the position space representation of states.
y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—. (11. so that im(ip±2ir) The functions Ylm(8.79) . is given by (2Z + l ) ( Z .i». These functions are defined in such a way that they are normalized to unity on the unit sphere. Uniqueness implies immediately that m has to be integral.80) y? Furthermore Y°l = V^fcosO). (11.r o ) ! YT{e.11.* ? ^ .76) and (11.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}. m > 0. / 3 / 5 J — casO. (11. and that their phases are such that Yt0(0.V>)Yir'(8. (11.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.n>).ypx = tfrK(11.(5cos 03cos0).82) .<p) Jo = 6mm'5w. orthonormalized system of eigenfunctions. We mention in passing that in spherical coordinates d lz = xpy .4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs. This is the case if one demands that l 2 and lz possess a common complete.t»'\Y7n((l'.(p) are called spherical harmonics.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.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. (11.^ = *(n . The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9.0) is real and positive.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.\ = W .
89) Inserting this expansion into Eq. Setting y = (lx2)m>2Pr(x) m .x2)Pf . the designation s. in the replacement ip —> if + 2K. For the solution we use the ansatz of a power series.87) we obtain the equation (1 . f.. respectively.x2)P[' .1(1 + l)]xK+i = 0. Instead of the values I = 0..e. (11.2xy' + 1(1 + \)y = 0.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .m = 0.1. i=0 (re lowest power > 0).83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 . is also in use.86) y = 0. (11..d.m(m + l ) ] i f = 0.x2)y" .90) . of Ytm.e.212 CHAPTER 11. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. (11.. i.2(m + l ) x i f + [1(1 + 1) . i. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .85) (11.l)xK+i~2 i . oo (11.h. i.e. (11. It is instructive to pursue the appropriate arguments. 22 (11.g. exp(imip).2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.p. Consider the equation for Pi(x). we would now have to search for those values of I for which the solutions are square integrable.88) y = ^aiXi+K. i.. (1 . If we did not know yet that I = IQ = an integer > 0. (11.2.88) we obtain for every value of x ^2 ai(K + 0( K + * ..x2)y" .e.2xP[ + 1(1 + l)Pi = 0.
For the solution in accordance with (11. (11. Murphy [190].1.Qr2 with (cf. For \x\ < 1 we can obtain from this the coefficients of a convergent series. Eq. ai(re + !)« = 0.92) This is a recurrence relation.91) from which we deduce that K = 0. (11. they have to be polynomials (recall the orthogonalization procedure of E. first line of Eq. We see that this happens precisely when I is a positive or negative integer or zero.94) from the left by r) h2 d2 h2 2 + H! + \)IL.93) where Xl(r) is the solution of the radial equation (cf. l 2 and lz are the solutions of the Schrodinger equation of the form (11.E Xl(r) =0 (11. . Schmidt).46)) p. .96) + V(r) . M. p. We have therefore (multiplying Eq. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . Margenau and G.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0. these are of little interest to us here. > 1. but also those for a different series for a.26)) yi(r = 0) = 0.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. l(l + l)h2 _2mo + 2m. (11.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.94) ^For further details see H.2 + 2mo dr 2mor V{T)E Viir) = 0 (11. 93.11.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11.*" 11. when the series terminates after a finite number of terms.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. This is precisely the case. (11. However. (11. Eq.1(1 + 1)]. (11.78) to be square integrable.
Z\ = Z2 = 1 in Eq. s = 1 + 1. The wave function of the relative motion is the solution ^(r) of the equation A 2mo Here •4>{r)=Ylm{eM and Vl(r) (11.1)+1(1+ 1) = 0. (11. (11.e. i. an analogous consideration applies to hydrogenlike atoms like.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. and substitute this into Eq. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential.1)). for instance. 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. Obviously the irregular solution has to be rejected. Equating the coefficient of 1/r2 to zero.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. r m0 = meMp me + Mv (11.1.96).99) e ip(r) = Eip(r).e. He + . Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. (11.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. L i + + . In the first place we investigate the behaviour of yi in the neighbourhood of the origin. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. nor the condition of normalizability for l ^ 0.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. we obtain the socalled "indicial equation!'' s(s . which behaves like 1/r1 in approaching r = 0. We now consider the hydrogen atom with V(r) — —e2/r (i.102) . oo) of a onedimensional space. The Coulomb Interaction (11. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). since it satisfies neither the condition yi(0) = 0. and others.100) with (11.214 with CHAPTER 11.
(11. Oberhettinger [181].105) we obtain the equation for V[(x).529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11. (11.+if>x/2 vi(x).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.a * = 0.v) vi(x) = 0. (11.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). T This equation is known as Whittaker's equation.11.108) This equation has the form of a confluent x$" + (6 .103) °y moe2 ( = 0. behaves there like xl+1) extended to an exponentially decreasing branch at infinity.6. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . p.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.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0.g. (11. hypergeometric (11. (11. *The factor 47reo appears in SI units. It remains to determine the function vi(x) for a complete determination of the solution. We can see from Eq.107) equation.e. i. Magnus and F.106) W i t h this ansatz we separate from the solution the behaviour around the origin as well as t h a t at infinity. (11.88. see e.e. W.105) 11.6 11.104) W i t h these substitutions we can rewrite Eq. (11.e. We set therefore Vi X J. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. . for E < 0 they > decrease exponentially. Inserting the ansatz into Eq. i.x ) $ ' .
02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . The function T(z) is defined by the integral /co T(z) := / eHzldt. more precisely. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases.1/2)!.(0_i)!(6 + n l)!n!' (11. In general the function $ is an infinite series and behaves at x — oo like » x{l+l+u)ex_ Obviously such a series is useless in our case. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. (11. and hence the *Some writers. it is convenient to use only the factorial notation. the factorial for an arbitrary argument.112) The gamma function is a nonvanishing analytic. However. with many of these around in some calculation. .z) = ^f^.! ) ! = imlFI' c a l led inversion or reflection formula.J. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 . 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. particularly pure mathematicians.216 CHAPTER 11.e. and for both cases. (5)r(n + l ) = n ! .z(zl)\ = z\. (4) r ( i ) = V5F. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (2) r ( z ) r ( l . (11. ^ ( 2 ^ ) ! = 2 2z z!(z . called dupZication formula. (z)!(z .106).109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. meromorphic function. (11. i.
(11. 1=0 (11.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. i. and so to the number of different values of m. 2 .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. One defines as the principal quantum number the number n.» • » • » . in our case: a = 0. which is thus seen to be a consequence of demanding the square integrability of the wave function.e. . for z. which are also described as its nodes.2. i. The quantum number n' is called radial quantum number.e.101) to (11.1) + n = n2. Eqs.113) This is a quantization condition.1.v) T(l + lv) _ ~ K sin{7rQ/ .11.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. = n = Z + l + n'. 1 . n' = 0 .104)) 1 n =v = K.e.. must be a polynomial. i. With (cf.I) r(vl p)sm{n(i/ I . d = 1. .. .1) + n = n(n . .V))T(v . (11.110) as follows: T(l + 1 + p . if the series terminates for certain values of u. This is achieved.. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). . 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 .6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. Thus the infinite series must break off somewhere. the orbital or azimuthal quantum number I can assume the values 0. (11. For every value En.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ .. every n.
'.1 = 0 (called sharp series.. and from n. 11. respectively.. 2 .\ I = 0 or 1. . 11. from an initial state n > 2. hence s). from n. from an initial state n > 2. I = 0.4. One describes as Balmer series the final state n = 2. np. hence p). relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11... hence d). 0. I = 0 to n = 2. . ' O n e describes as Lyman series the final state with n = 1.1. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. stand for I = where n denotes the principal quantum number and s.1. (c) The nucleus was treated as pointlike and not as something with structure. ne. / = 1. nd. .§ One also writes n2l+1l.'. and hence the fine structure of the energy levels is excluded.l = 2 to n = 2.p.1 = 1 (called diffuse series. .218 CHAPTER 11...1 = 0 (called principal series. 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. The spectrum of the hydrogen atom thus has the form shown schematically in Fig. (b) The spin of the electron was not taken into account. The following spectroscopic description of states is customary: nl —> ns.1 Hydrogen atom energy levels.d... nf . .1 = 1 to n = 2.
so that one always has to check the definition. M. 2 . for instance. and I. . (11. With the help of Eq.115) is effectively the associated Laguerre polynomial normally written Lk(z). (11. . Magnus and F.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.118a) where Lk. p.2. . • ( . Vol.z) = (r + k)\ Lkr(z).115) ' =0 (ri .114b) yields the binding energy of the states. Gradshteyn and I. however.414(3).6. Messiah [195]. "1'f„+11" P] „. l . several different definitions are in use. Unfortunately. see. .= t .p)\(2l + 1 + p)\ The spectrum (11.r f .2 (11.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). Each can be recognized by reference to the generating function. • (11. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me.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. 84.2l { + 2.r = 0 . We mention here three different definitions used in the literature. Oberhettinger [181]. p. For clarity and later use we denote the function used there by its normal form L^.x). (u. I. 3 . Unfortunately the definition mostly used in mathematical literature differs. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr).6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. W. (11. • fc. Ryzhik [122]. a + 2:*) = t v . 844. k + l. in fact. the polynomial contained in y\ of Eq. 11. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. Appendix B.11. formula 7.1.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. .117) The function *Lk(z) is a polynomial of degree r. S. Apart from factors. We distinguish between different definitions in use.
which we write ^Lk.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.„M„ z"*L.z . 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. L2(z) = l2z + z2.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1.164. (11. and Ll(z) = l.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. Messiah [195].220 CHAPTER 11. ^ F o r this definition and the following equations we refer to I.1. Appendix B. is defined by^ dk (—'\\k(r^\2 (11. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. 163. L\(z) = 2z.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.d z—^z + {k + l~z)—+r Lkr{z) = 0. It is said that this is the definition preferred in applied mathematics.122) "See A.121) In Example 11. Sneddon [255]..3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function.3): e2t/(lt) (1 . Ll(z) = 33z + ±z2. **I.** The associated Laguerre polynomial here. Example 11. There at some points n + 1 is misprinted n + 1. L1(z) = l . . N. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . p. 162 . r r Z r=0 v tr k k (11.. N.„\*Kk. I. Vol. . Sneddon [255]. pp.2. .
8): (11.120)) it)k& zt/{lt) ~*P p=k fc (11. (11. *L\(z) = . ^L2(z) = 2 . Example 11.4 + 2z.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 . .119) derive the generating function of associated Laguerre polynomials (11. Solution: We consider the following equation of the form of Eq.z .121).107) implies k = 11 + 1 and r = n + I.120).. (11. ^L\(z) = .3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq. The limits will be determined later.1 8 + 18z 3z2. (11.124c) Example 11. (11. (11.124b) Comparing Eqs. (11. N d rk 0.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. Then.1 .% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.4z + . i L1(z) = l . Jfpd>2s{p))eptdp.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..122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z).124a) (i*) pi The orthonormality condition is (cf.. and partial integration. '"''Comparison with Eq.118a) and (11. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.119): ty"(t) + (a + 1 . and *L\(z) = . and obtain with this the normalization condition (11..
It follows that 1 d i I _i. j p = (P 0 "V?.q) and P~ t ( .p ) « + i q c + i ( i . Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. 2TT_ Jq=0qb+l(lq)a+l 1 . Case (2) implies the exponentially increasing solution ~ e + t .. agreeing with Eq. since for n > b : (b — n)\ —• oo. (11. (2) from —1 to oo.a = 21 + 2.. . Then dp = dq/(lq)2. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo.p + 1 = 1/(1 . n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). and (4) around p = 0 if b > 0 and integral. (3) around p = — 1 if a + b is a negative integer. 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.118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . t > 0. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. 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. with requirement of a finite solution. the only possibility is the last. p and one has (as polynomial of degree 6. (11. Thus in the case of Eq.119) with a and b as there.222 CHAPTER 11.] from the above equation: C(p) := p(p + l)s(p)ept = 0.p ) o + i 7 t = 0 .) 2 J J p 6 + i ( l . this is the rejected irregular solution. We demand (to be considered later for integration contours) the following condition. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* .124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. case (3) similarly. which removes [.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./ ( l — <?) —— = G(t)." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt.^ . s bK> (lq)°+i (11.( a + l ) p s ( p ) + 6s(p) dp 0. Hence the expression in square brackets vanishes.
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.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11. and 0 if cj=b. I.(x).?). and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = .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. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as .11.125a) and the factor a~ 3 ' 2 with a = Bohr radius.—i. (11.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. so that I = (c + a)! 1 f . The wave function of the state (n. 11.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. (11. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! . where Fnl(x) = yi(x) (11. is introduced for dimensional reasons. Clearly only the second form is sensible.112). Jo (11.111). m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.6. .127) . Eq. Then Nni is a dimensionless normalization constant and (note that we use the associated Laguerre polynomials as in the book of Sneddon [255]) Fnl(x) = xlex^LZ.
(11.3fc + 2). thus on the average the electron is the farther away from the proton or nucleus the larger n is. 2 (11.. This behaviour is indicated in Fig.6(n + Z)[(2Z + ! ) .(l) n+l. = 0) we have ) r /u =.1 + 1) (11. from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix ()r2dr 2 ^i + /„. a (r)is = ~a.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .224 CHAPTER 11. (11. + . The Coulomb Interaction One finds in particular (see also Example 11. For the ground state (n = 1.7.129) 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 .v (6n2  6nk + A:2 + 6n .(2Z + 1) + 1} = ^[Zn21(1 + 1)].3(2Z + 1) + 2} 2 {2(n + 0 .2.131) .130) We see that the mean value of r is the larger the larger n is. 11. and for the explicit derivation of the second case below Example 11.8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .! ] + (21 + l ) 2 .2Z .
Sneddon [255]. with n +1 replaced by n and 21 + 1 by k. Problem 5.J n 1s 2s 3s *r Fig.11.( n .2 Hydrogen atom wave functions. or (r ) ~ n a 2 2 ^ [ 3 n 2 .j . In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .20. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.e. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . In the case in which I assumes its maximal value n — 1..(r) = x W 2 n + 1 =.2 I 3/2 Kl r 225 \2 1 R„.6 The Discrete Spectrum of the Coulomb Potential r.133) for large r. v ^ V2n + 1 (11.2 1 A r = V (^2) . 11. \f{r ) ~ n a. / ! J — . = l a nl n N'nll rFnl. 173/174. pp. i.^ We obtain therefore for the quadratic deviation (r) .134) *These are taken from I.l ) n ] =  [ 2 n 2 + n] an[ n + (11.132) (11. S. we have and tp becomes particularly simple.
i = 0.4: The relativistic eigenenergy using cylindrical coordinates Using the formula for the relativistic total energy E of a particle of mass mo and charge e moving in the Coulomb potential V = ~Ze/r (with E numerically equal to the corresponding Hamiltonian H for a conservative system).r. r. (c) The nucleus was treated as pointlike and not as something with structure. pJ = —— and j> Pidqi = nth. the electron remains practically localized within a spherical surface of radius (r). we have pg = const.226 CHAPTER 11. for these values of I it is not distributed with uniform probability over the spherical surface. p2 = p2 + — pg • eV. we consider in Example 11. However. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". in which periods are obtained from the BohrSommerfeldWilson quantization condition. Solution: We have H = yjm%c4 + pIn cylindrical coordinates.4 the relativistic equation in cylindrical coordinates.6. Example 11. \/v vT 7 v^ See also Example 14. the states Ytl are predominantly concentrated in the xyplane). (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. The Coulomb Interaction or TT = ^ = (11. Since 6 is a cyclic coordinate. i. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). This means. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. and separating this in cylindrical coordinates r. Thus. = ngh. 6.135) This expression becomes very small for large values of n. .e. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.3.— > where ampler is the length of the semimajor axis of the elliptic orbit. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . but rather like classically in a plane (i. In order to obtain an impression of relativistic corrections in a simple context.e. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r.
The deeper reason for this are symmetry properties of the problem. Whenever there is degeneracy. 11. where the constants are identified by comparison. ( v c + i ) . (p).136b) £ = r ( l + cos#) = r + z. Schiff [243]..11. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. it is possible to separate the Schrodinger equation in other coordinates. which correspond to a new choice of polar axis.2. 86.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. z=^rj).V . (11.T). Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . p. <p = <p.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 £. 11. .(p . y=y/%qsanp.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9. where r = ^/x1 + y2 + z2.136a) (11. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r.* See L.9. y — 0: V ^ c o s tp. 2 2 Therefore the energy is no longer a function of n = nr + ng. r\ = r ( l — cos#) = r — z.6. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. where ip = 0. which are related to Cartesian coordinates in the following way.3. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137. as also indicated in Fig. ip are the spherical coordinates.
228 CHAPTER 11.3 Parabolas £ = const. for uniqueness of the wave function. 11. there is really only one basic equation to solve here.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. (11.22) becomes — as we show in Example 11. 2 . . a separation constant. 1 . Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . with zaxis as axis of rotation. In terms of the coordinates (11. .3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. The Coulomb Interaction Fig. and r\ = const. = 0. We rewrite the first of these equations with the help of the relation M (w ^  m + e) . d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o.136b) the Schrodinger equation of relative motion (11. . so t h a t the variables can be separated.
2 . we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". i.105)) Hence we write I = (m — l ) / 2 .11. . ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) .i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 .e.± ( m + 1) = 0 . .6. . . 2 . Adding now / .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. E x a m p l e 11. " .125b) (dividing by p 1 / 2 ) £ oc e . We leave the calculation of the degree of degeneracy to Example 11. . To obtain a closer analogy with the separation in the spherical polar case. 1 . . T h e n by comparison with the spherical polar separation the answer is as in Eqs. 1 . Eq.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. .o. n 2 = n" .114b). p = /?£ and cr = /^r?. m0Z1Z2e2 ^ 2 = n (say). . from Cartesian to parabolic coordinates. as well as the Laplacian A = V and the Schrodinger equation. . . In the spherical polar case we had (cf. (11. .' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).106) and (11. ds2 = dx2 + dy2 + dz2. (11.
E E n l „ . W.2 .e.. z = (£ — rf)/2. + l 4 {^^)drl f i + v \ .1) ~ . Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). Similarly we proceed with dx and dy.g. . 2 _ = (9id02 + (gvdvy + /„ .1 in this expansion is seen to be n. The Coulomb Interaction Solution: We begin with z of Eq.e.l ) ( n . for instance. Eq. (11. 2 2V? V^ One now performs the square of this and obtains similarly dy2. i. last chapter. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n .6: Calculation of degree of degeneracy Show that the degree of degeneracy of the wave functions in terms of parabolic coordinates agrees with that obtained earlier for spherical polar coordinates. .2) and summing over m.230 CHAPTER 11.2 = n — 1. i. 2 + ^dV . see Eq. " .g<p 9 \ . i. Thus dz2 = .l)a + (n .136a). Thus. We then sum all possibilities and select the coefficient of ojn~1 in the result. and attach a factor to to each term introduced in the wave function. {g^Y. we have to find the number of values of the quantum numbers n\.114c). </. 712 and m which give n. Solution: For the wave functions ip = £(£)J\f(r])&(ip).* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . N2 where g^dS. \— < / — cos ipdr] — \/£r] sin ipdtp. dx = d{ y £77 cos ip) = .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. Beware! For m = 0 there is only one mstate.e. all cross terms cancel in the sum of 7 the squares. n 2 (1UJ)2 The coefficient of a .. / — cos ipdl. when m = 0 = 77. the total number of wave functions we have is the number n with m = 0 plus twice the number with m ^ 0 starting from m = 1. We use the method of selector variables. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. for m / .2)d} = (n.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . (11.2d? dr)}. We consider first the case m = 0. Thus we want to obtain the coefficient of wn~1 in £2 .tx2 . Oberhettinger [181]. 0 there are two.e. / J„^2 . = e±im^e("+^'2{par'2L^{p)L^ (a). Observing that the highest possible value of m is n — 1 (i. as t h e two signs of t h e exponential indicate. . (n _ !) n = n2 §See e. n the double sum ]T]2 is TI — m.136a). Magnus and F. > . are elements of length. (11. Here we have (see preceding text) 711+77.2)d/2. and so dz = (d£ — drj)/2. In the case m ^ O w e have n\ + 77. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.[df + dr]2 . etc.
(11.a). (11. with z = r cos 9. I.128) if one makes there the substitutions n —> n\ (or 712) + m.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.•* n (l) = / Jo It follows that .127) and setting n = r + k. With the help of the wave functions we derived above. eF{z). we can relate the expressions Iq to those of Eq. L ni\ ny.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. with the connection » (11. and determine the average value of v for a given state. if we expressed the perturbation in spherical polar coordinates.e. which will then give the change in energy of that state to the first order (i.e. fc — m. (11. . .6).) = ^eF(p .™(p)] 2 By contour integration (see Examples 11.11.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. where F is the electric field. 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. = j (2ni+m+l).128). Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). i.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz.drid<fi oc (p + o)dpda.8 and 11.g. Thus e. with Z\ —> Ze. {p — a) — = dxe~*xkvLkn (z)] 2 . with Eq. Ikr+k. . we would have to use degenerate perturbation theory and obtain the same result.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—.1 (cf.124c). i. In fact.—r~n = 3(mn 2 ). with RayleighSchrodinger perturbation theory). . Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. .e. s 0/ ..6 The Discrete Spectrum of the Coulomb Potential Example 11. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. these expressions agree with those of Eq. Z2 —> e. Sec. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. 8.r. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . However. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in.
9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. tt„:_ (1 . (11.124d) of the alternative integral representation obtained in Example 11.( 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)!. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. dt = ds/a.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 .x ( n + m)! ( n .p ) ( l .n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .g )(l.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l .128).1 ) ! (n .a. s = at. _ „ . 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 by one power of t. In Example 11.p ) r + 2 [(lp)(lg)]"l+1(lpg).3).^~T^——^ ) = coefficient of (pn . L™(t) = coefficient of qn in G(t) e tq/(lq) (1 .. .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. the integral arising here becomes with Eq.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11. (11.1)! n\ This verifies the expression I™t in Example 11.9 )« Solution: Using the generating function for both associated Laguerre polynomials. the integral / is the coefficient of pnqn in the expression _ / .9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11. and is a special case of Example 11.232 CHAPTER 11.b.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t . The Coulomb Interaction E x a m p l e 11.p ) ( l .pK — — —L .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. E x a m p l e 11.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n .9 we consider integrals with an arbitrary power of t.3: /"OO I := Jo et[L^{t)]2tm+1dt. (n + r . c positive integers.
= c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .e. .e. where (a+**£)! c j n Jo = .?(t)t a e.a _ 1 = [1 + a ( l .q)(l . i.p ) ] .(i) = i ! ( . ° ^ + a ( l . (a) j / = 1. In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.f ].q(p + a  The coefficient of q in this expression is 1 [l + a ( l p)]a+l where f 9 / p + a — ap\c (c + a)! _ ac(c + a)\g(p) \ l + aapj c\a\ c!a!(l + a)a+c+1' [l + ^ P ] " [1T^P]Q+C+1 y>^ d(a + c + C .1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0. Using the contour integral representation (11. we have 2r = 6 + c = even. c + b — 2r = l (i.124d) of an associated Laguerre polynomial. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . (lp)(lq) + a(l q)(l .e.a t [Lg(i).Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. . c + 6 — 2r = 0 (i. 7£.11.p) .a f + c)\(vr)\(l + ayrP „ • ^0^r(cr)\r\(a •• Ha) a\ f dp ac(c + a)\ ^ =%*??* c!a!(l + a ) ^ c + l 9 ( p ) = g J c+*"«a> where with /3 = a + c + l + b (observe that for a = 0 a term with c + b — 2r = 0 remains) „_ = 2 ^ 2 ^ K «.r ) ! a ' .111) of the gamma or factorial function. i.L. — (cr)!r!(6r)! Kv. we have to have y + (c + b — 2r) = 1. so that = (q+fc±£) .™( r > (1 + a)P r £^0y^w Jc+6_2r = .e.w(r) = and / a ) = r\(P ~ r .p ) 0 + l 9 c + l ( l _ g ) a + l ^ a ) ' where j(a) contains the integration with respect to t which may be evaluated with the help of the integral representation (11. c + b odd).(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. c + b even) and/or j / = 0.l ) * x coefficient of a* in the expansion of 7 ( a ) .^ ( l .p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .
~* 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 . 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. the standard principles of nonrelativistic scattering theory.1%^ 11. Taylor [267]. 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. ^A very readable source to consult. however parallel to the Coulomb case. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). t We have the Schrodinger equation of relative motion of particles with charges Z\e. is the work of J.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16.^)(. R. b\ V cited in Example 11. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. 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. In view of the slow falloff of the potential at infinity. . also with regard to earlier literature.)!' f^)(.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.7. Since such a rigorous treatment is beyond the scope of this text. . + (!±<)«.
87 under Kummer functions. (11. (11. (11. . A = const.z). „ az 1. b..+ ^  + . writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.108) and (11..138) becomes 235 ^A + k2 . (11. where ^. pp.141) Inserting this expression into Eq.142) This equation is of the form of a hypergeometric equation.e. that we encountered previously. with the asymptotic expansions (11. i.109) we can therefore write the solution ipr = Aeikz$(i>y. .146) See. Our question is now: How does ipr behave for r — oo.z) + W2(a. for instance. Oberhettinger [181]. (11. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. z = rcos<9.143) and v = ikrj.e.140) The simple Coulomb potential permits particularly simple solutions.M) = l + . In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr)..144) *(a. W. Magnus and F.b.7 The Continuous Spectrum of the Coulomb Potential Then Eq.140) and using (11.b. what is the > behaviour of 3>(a.11. 86 . a(a +1) z2 ] (11.^ W ) = 0. for r — oo) • ipr = Aeikzf(r .z)). (11.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). . i.. According to Eqs. (11. v = ik(r .z).136b). (11.b.z) = Wi(a.
a) (z) na)"~ r(la) T(ba) n\ ' (11.150) ^(rz)]1"^ 1 + 0 T(«7) i.152) Since z = r c o s # . J\") ci(kr~/\n2kr) (11.^l*7 l + O (11.153) The asymptotic solution represents t h e stationary scattering state (E = const. = r cos 0. (11. if tbr were „ikr ibr ex el"z + f{6) r (11. > A = Aeikz[W1(i1.b.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.7 ln*(rz)) k(r .154) .l.i 7 ) (11.kz+l^ktrz)) C .ik(rz))} =^ + ^d.l.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].z) + la)T(n + b. (11. a comparison of Eq. (11. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. with 2. i.z)T(l . The Coulomb Interaction W2(a.e.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) .) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.148) .151) with Eq. Without t h e logarithmic phase factors.(11.e.149) (11.236 n\Zzab CHAPTER ^T(n '£ fo 11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.
11.4 Variation of the observed differential cross section with 0. The quantity /(#) 2 would then be a measure for the scattering in the direction 9.11.n . that we encounter here is a characteristic of the Coulomb potential.. (11.155) Since the gradient is proportional to the configuration space representation of the momentum. z — — oo.156) . As a consequence of the masslessness of the photon. also called phase anomaly. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. the current density is * h . 11. do dQ K 12 K Fig. We define as current density the quantity h J := 2imo [^(v^)v(v^n.Q hk mo v0. (11. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain.7 The Continuous Spectrum of the Coulomb Potential 237 the wave could be looked at as the sum of an incoming plane wave exp(ikz) and an outgoing scattered wave. The problem of the logarithmic phase.7. 2im. we see that this expression is the exact analogue of a current density in classical considerations. In other words. For the incoming wave ipi = exp(ikz).1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. the effective range of the Coulomb potential is infinite.
238 CHAPTER 11.152) as 1. i. is called scattering amplitude. 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. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later).152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. 'screened ZlZ<2e2 cr/r0 ° . for 0 —> 0. In Eq. the square 7 2 ). This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . We recognize the result (11.154) we chose the prefactor of the incoming wave in (11.e. That this formula retains its validity here in quantum mechanics is something of a coincidence. (11. (11. We make a few more important observations. thereby screening itself off from the surroundings. This implies that scattering takes place for the attractive as well as the repulsive potential.159) as the Rutherford formula which can also be derived purely classically. We see that a depends only on the modulus of the potential (i.e. which yields the scattering cross section.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. in the current this would in any case contribute only something of order 1/r. In the context of quantum electrodynamics one refers to "vacuum polarization" and relates this to the idea that a charged particle like the electron polarizes the otherwise neutral vacuum by attracting virtual charges of opposite polarity and repelling virtual charges of the same polarity. Correspondingly the other part of (11. In reality this divergence does not occur.153): 4fc2 s i r  The expression /(#). since in actual fact the Coulomb potential occurs only in a screened form.
whose maximum moves with the particle velocity.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . The wave packet is a superposition of waves ip(x. (k). and which provides the uncertainties arising in quantum mechanics.t).8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. i.165) ^ ( x ) = £ fe Vk(x). 2mo V + F(x) k (11. (11. t0) = e. t) = U(t.166) Let the incoming free wave packet at time t > t$.^ ^ % ( x ) . of continuum wave functions. and if H is timeindependent one has the stationary wave function ^(x. (11163) (11.4. i.2 fr2 .t). t 0 ) .x° ' r j .8 Scattering of a Wave Packet 239 This screening of the Coulomb potential insures that the scattering cross section in forward direction is actually finite as observed experimentally and indicated in Fig. t 0 ). The particle velocity v is v = — . 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.i i f (''°>/Vk(x.161) where Ak0(k) ~ A(k — ko) and ^ k (x. / dk 7^340(k)4(xxo.t) = e .e. (11.164) Ek~Eko + {kk0)v0h. t 0 )^k(x. m0 m0 and V.11. which in the case of free motion (zero potential) are simply plane waves. 11.e. 2 (11162) where Vk(x) = ^ ( x . be given by ^ f dk AL. i. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . the wave packet in the absence of a scattering potential. of the equation r fi. 11. v0 = .e.yJ ^ j A k o W 727)3/2 e 0.
165) ko • (k . can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.iB *(**°)/ R °'tj + 1 pikr e * .169) The wave packet in the presence of the potential is obtained from the free form (11.170) A ko (k) e . i. The Coulomb Interaction In Example 11.10 we show that the solution V'k(x) of Eq.163) we obtain </>ko(x .x 0 ). (11.i) dk (2TT (11. k = kez.k 0 ) 2 ~ Jk20 + 2k0 • (k . ip) could be carried along but we ignore it.169) and (11.ko) k0 (11. (11. Eq. (2TT)3/2 eik* + —fv(P.^ ( * .167) by replacing the plane wave exp(zkx) by this scattered wave.xo.167) we have ^k0(xx0.166). (11.161) and (11. ^k(xxo) ^k(x) ~ = e. the stationary scattering wave. From Eqs.e. (11.t) = J ^ A k o ( k ) e . this solution > has the following asymptotic behaviour in which f(6.ip) is the scattering amplitude with respect to the incoming plane wave.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.x °Vk(x).173) . (11. With Eqs. Then with k = ^(ko + k .t o ) / V k ( x .240 CHAPTER 11.154)) that for x — oo.e.172) h h k0 {rv0(tt0)} (11.k0) ~ k0 + and the expansion of E^ in Eq.<p) + Ol Akr l (11.171) The phase of fk(0.ik . i.
r ' ) = 47r<5(r .r'). (11. and determine the Green's function G(r.y)eiko'Xoe'[fcor£.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . (A + fc2)t/. t) f.*0ko (x . T h e scattering amplitude is t h e same as t h a t given by Eq. Solution: We define the Green's function G ( r . *. however.to) has t h e form of t h e wave packet in radial direction as indicated in Fig. .8 Scattering of a Wave Packet so that with Eqs. i. with the energy of the maximum momentum of t h e wave packet. E x a m p l e 11.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r.e.x 0 .174) r /k0(^. t) .^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.x 0 . U(T) = ^ V ( r ) .^y /k0^.r')V(rW). '. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.5.167). (11.11. Here ^ (x' — xo.r')l/(r>fc(r').fco(tto)/ft]^0o)(x/xo. the solution can be written iMr) : (2TT)3/2 ^dr'G(r.z Fig.to). Adding a suitably normalized solution of the free equation. 11.171). r ' ) by the equation (A + fc2)G(r.5 Scattering of a wave packet with scattering centre O.154).(«)/ ipko(x . (11.10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.r ' ) . A particular solution of the Schrodinger equation is — J G(r. 11.
e > 0 small. displaced slightly away from the real axis as indicated in Fig. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r .6 The contour C+ in the upper half of the k'p\ane. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k).fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. 11.=k+ie/2k 1 d f. 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. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . which is such that the contribution along the infinite semicircle vanishes.fc ) oo dfc' 1 d nik r 7rr dr / . dk' ik'*r 2TT2 fc'2 . The Coulomb Interaction We find a set of Green's functions G(r. . Thus we consider the contour integral taken around the contour C+.Rek' kfc/2K Fig.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. These are the socalled outgoing and ingoing Green's functions. one obtains another Green's function given by G_(r)= r eikr. Imk ./2 dk'd(—cos 6)dip iki.242 CHAPTER 11. Hence we write G(r)G+(r).6. 11. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r .
Then yW (11.9 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.£).145)) F(a.176) " 0—A ~ (I + 1 + iy)<t>i = 0.178) (11. i. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7.11. Thus we proceed as above. (11.140) (previously we solved this only for the bound state problem) § . e. 21 + 2.e. b.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). 2ikr)] [Wi{l + 1+11.z2 27fc 1(1 + 1)' yi = 0.2l + 2. Eq.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij.g. b.0 + W2(l + 1 + iy.21 + 2. (11. If we proceed from the radial Schrodinger equation. whose continuation to infinity is again given by (cf. under the stated conditions 2/07 = Z\Zie2 = —Mo. Before we introduce the general form of the socalled partial wave expansion. (11.100) — 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. where Mo is a parameter in Chapter 16.e. (11. z). i. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. (11. Eq. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. 2ikr) s For comparison with calculations in Chapter 16 note that for h = c = 1 together with 2mo = 1 and hence E = fc2 the quantity 7 here contains a factor k. z) = W^a. Then £^<fc + (2l + 2 £ = ~2ikr. b.2Z + 2. z) + W2(a. .9 Scattering Phase and Partial Waves 243 11.
(11.i f c r + i 7 In 2/cr+i7r(i+l)/2 + We set T(l + 1 + 17) = r(z + i + i 7 ) ' r(z + I .244 CHAPTER 11.l e*"" + T(l + 1 + H ) T(l + 1 .n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 .j syyll iy il( _ . (2kryr{i" V 7 .183) r(Z + 1 + 17) := Reie. Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5. .ry) = e"^r(Z + 1 + ry). Then e2i5< = Reie Re~ J2i9 9 = Si. T(2l + 2) (2fcr)*T« (Alliy ikr .ii)' (11. The Coulomb Interaction Using Eqs.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .181) so that eldlT{l + 1 .«7) „ .180) ^ ' (11.i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.146) and (11.' . +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .
(11. However. 245 (11. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .1. (11. kr) the regular solution with the following asymptotic behaviour: r 0 (11.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr.e. In the present quantum mechanical considerations I is.187) contains almost the entire physical information with regard to the Coulomb potential. i. One defines as regular Coulomb wave Fi(j..185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element.e.11. .184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. where the continuum of the spectrum starts. That the energy E appears in Eq. (11. a positive integer. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type. with Eq. i.139) n' = 0.e.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7).114a) reveals). considering I as a function of E extended into the complex plane.. as we saw. For instance the poles of S are given by Z + l + z7 = . g = e«* = rSi + 1+ *V*ri (11.2.e.n ' . i. considering I —• an(E).188) in the form y/E. i. ._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq.
11. In these cases." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. (11. Singh [252]. (11.7.189) where as before the incoming wave in the direction of z is ikz ipin = e (11.7 A Regge trajectory of the Coulomb potential.191) . We shall return to these in Chapter 16 in the consideration of screened Coulomb potentials or Yukawa potentials and indicate other important aspects of these. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. For the Coulomb potential a typical such trajectory is indicated in Fig. we obtain the partial wave expansion of the scattering amplitude as follows. we obtain trajectories which are known as Regge trajectories. The Coulomb Interaction and plotting these for different values of n. 11.246 CHAPTER 11. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt. and thus in general.
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.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.~ J > Z + l)Lil*e~ikr . The mathematical expression can be found in Tables of Special Functions (there cf.^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY.i f c r .r ^ V ^ Wostf). 21 . The incoming plane wave can be reexpressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase. . pp.11.g.22. Magnus and F. W. r e.195) with the scattering matrix element S(l. (11.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg . (11. Oberhettinger [181].^2l + l^S^k) lik (11.k)=me2iS^k\ in which r\i^\ indicates absorption. 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 ' .193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave.eikr\P^cos9). (11. There the expansion is given in terms of Bessel functions Jv which have to be reexpressed in terms of the asymptotic behaviour (1 2) of Hankel functions H„ ' .192) Thus the incoming and outgoing spherical waves have a definite relative phase. The potential disturbs this phase relationship (by delay or absorption or redirection).196) "See e.
This page is intentionally left blank .
e. therefore. Similarly difficult is. i.g. It is very instructive. Razavy [236]. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. in general. can be found in any traditional text on quantum mechanics. It is not surprising. in texts on quantum mechanics is rare. however. Such a vital quantum mechanical phenomenon is socalled tunneling. namely the instanton (or more generally pseudoparticle) method. These newer methods will be presented in detail in later chapters. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. For various general aspects and applications of tunneling we refer to other literature. as — for instance — for trigonometric or double well potentials. that quantum mechanically a particle may have a small probability of being in a spatial domain. states of a finite lifetime. that the explicit derivation of such quantities in nontrivial contexts.e. In fact.* *See e. one of the main and now wellknown methods to perform such calculations. i.Chapter 12 Q u a n t u m Mechanical Tunneling 12. M. 249 .1 Introductory Remarks One of the main objectives of this text is the presentation of methods of calculation of typically quantum mechanical effects which are generally not so easily derivable in nontrivial contexts but are important as basic phenomena in standard examples. to explore first much simpler models in order to acquire an impression of the type of results to be expected. which a particle in classical mechanics would never be able to intrude. which in one form or another. was developed only in the last two to three decades. the computation of finite lifetimes of a quantum mechanical state.
t).5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.t) that the one particle concerned is somewhere in space at time t. (12.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj.t)\2 =1 (12. i. ihr(x. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.t) = HiP(x.4) (12.t) (12.. t) was developed by Born in 1926.250 CHAPTER 12.1) as the probability described by ijj(x. r (12. so that p = T/>2 is the probability density for a spatial measurement at time t. According to Eq. M^ Then dt + Vj 1=0.e.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. .t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x. Quantum Mechanical Tunneling 12.
the problem is analogous to that of the scattering process from a potential in a onedimensional case. higher dimensional delta potentials do not have properties like that in one dimension. Nonetheless. 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). we can do this also here.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. the equation can be integrated at every point. From this follows that both p and j have to be finite and continuous everywhere. The Schrodinger equation for this case is* ip" + [A. However. (12. where „ mpVo V(x) =+V0S(x).33. even for discontinuous potentials. Delta potentials in more than one dimension do not permit bound states and scattering. This effect is known as "tunnelincp. 1 0 ~x o a= f See L. 'Maybe the reader dislikes being confronted again with the onedimensional case. If ^>(x) and V^>(x) are known at every point x. finite and singlevalued at every point x in order to insure that V( x ) i s a unique representative of the state of a system. This case plays a special role.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. 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. Tarrach [119].6) .12. pp. 2m E 2 V0 > 0. Such a region is for instance the domain where V is larger than the total energy E. Thus it is possible that the probability for the particle to be in such a domain is nonzero. as shown by P.2a6(x)]il> = 0. although in general small. as we mentioned above. t For the singular delta function potential a modification of these demands is necessary. which is the case we consider here. Schiff [243]. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. ~hr> k = n^ (12 7)  . It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. 12.2 . 29 . If there is no absorption. 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. Gosdzinsky and R.
a — ik £Qr < { eikx a eikx •t" .12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—.ik(C .8) (12.' : It follows that ik . Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).11b) 2aC.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo. The continuity conditions at x = 0 are therefore for V : and according to Eq. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.252 CHAPTER 12.ikB) = 2aC or 2a(l + B).2 a J 5{x)4>(x)dx = 0. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .[^']o.: (i2iQ) Ce for x > 0. 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. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .12a) (12. 2 (12. i.e 1 + B = C.ikC = i. e Q (12.9) for ij.11a) ikC .1) .(ik . except in domains of the potential V.e. (12. (12.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes. i. .e. i. [<//]0+ .e.= 2a^(0). a 2 — a 2 a — ik a2 + k2 ' .13) ~A R= for x > °(12. (12. (12.
17) da#(s) 2 = 1. i.e. = . f Ce~KX c. at k2 = —a2. .) and V>'(0+) . poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. Thus ^ = Ce.ip'(0) = . R. oo (12. i.12.a l s l. Omnes and M.e «x = J \ CeKX for for x > 0.i 8 ) .g.K 2 < 0. 'Note that the physical pole has negative imaginary part.K C . K = a. See e.2 a C . .e. (12.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. possesses simple poles at a ± ik = 0.2 = . We write the solution ^ so that ^(0+) = ^ ( 0 .2 o ^ ( 0 ) . the quantity T.19) Thus the delta function potential supports exactly one bound state. as coefficient of the outgoing wave. 2m 0 (12. Proissart [223] or other books on scattering theory like R. z<0. ( 1 2 .16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. or h2 E = —a2. when normalized to 1. Newton [219].(K)C = . and C follows from the normalization: OO A.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0.
p. (12. x€[a.a]. (12. 58. Fermi [92]. 12. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.21) 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. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0. .22) + De Feikx + Ge~ikx *We follow here largely F. or in the lectures of E.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. \x\ < a.254 CHAPTER 12. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. 12. 55.1 The square well potential. x G [a. We want to consider the case of stationary states for E > 0. The latter contains also specific applications in nuclear physics.1.20) for for \x\ > a. Schwabl [246]. Quantum Mechanical Tunneling 12. p. V(x) V=E a v=y Fig. —a]. oo].
ikGe~ika.24) yields F ^ G. ).26) K^gifoa—i/ca fc.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. (12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca. M(a) .ikBeika = inCeiKa + DiKe~iKa. ( V C n vD ).27) v ' Similarly Eq. 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . Fikeika .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.f )e" t\ K\0—tKa—ika . .e. (12. (12.12.. (12. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12.iKa ? ikAe~ika and .28) Prom these matrix equations we obtain ( B ) = MWM(fl)_1 ( G ) ' where M(a)1 = v ^12'29) i / [i + n i K\eiKa+ika *)e (i t\ K\„iKa—ika fcje \ ] J fl2 30) • J ' 2 \ ("I — K'Jgita+Jfca Q _i_ K\p~iKaika \ . fc Af(a) = . (12. I (12.
xe[a. = < Ce~lKX + DeiKX .32) [cos2«.a .i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq.34) into Eq.31) we have A AT(a) = )MFo B ^r)sin2na . (12.a].^esin2«a]e 2lfca \r\ sin 2/ta Thus if we now set G = 0 in order to have only an outgoing wave proportional to exp(ikx) to the right of the well.35) with with with x G [oo.—2ika [cos 2/ta + \e sin 2«a]e~ (12.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.e.e. Quantum Mechanical Tunneling Hence with the complex quantities IK (12. i.a ] .37) . (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes. (12. . (12. we obtain A = F( cos 2na — .34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .oo].e sin 2na IJlika e B = r)F sm2na.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12. i.256 CHAPTER 12. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12. \R\2 = T(E)^sm2Ka (12.(1 + e 2 ) sin2 2na 1 . x € [a.4T(£)e ifci.
p.43) Thus we set k — —i ' H . 2\k K. an equation of the type f~1f = g19. Dwight [81]. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. 83. i. We see in particular from Eq.35) and using Eq.e. (12. in the domain of A imaginary and K real. (12. (12.40) we obtain cot(na) — tan(/«j) — i K IK . (12.12.38) as expected.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. Li ( — H— I sin 2na = 0.44) .e.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. formula 406. B. which is satisfied if either / = g or / = — g'1. nor when both parameters are purely imaginary.e. (12. (12. cot(fca) = i— or tan(fta) = K K . for VQ < E < 0. (12. .4 Scattering from a Potential Well On the other hand according to Eq.31): 1\T(E)\2 257 1 .12.41) i. However. (12.39) 2 i. they have solutions. and k real.42) These equations have no solution for K. 2m0\E\ h? ' (12. i.e.
+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n .l .l ) n . Eq.c o . T(E)e2ika = — r . ± l . and is itself infinite there. i.20). (12. (12. We now return to the scattering states. pp. This can be understood. (12.e. (12. . from Eq.46). (12.258 and obtain the equations CHAPTER 12. . (12.fl = ( . ± 2 .35)) for E > 0 sin2Ka = 0.e. . k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. (12. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.49) Taylor expansion of the denominator in powers of (E — ER) yields . i. F. n = 0 .45) These are the equations that one obtains from the connection relations for the case of Eq.65.42) which then leads to bound states with associated even and odd wave functions. 2m0\E\ Ktan(«a) = — y——^— . The remaining wave part becomes exponentially > decreasing. „ .48) We expand T{E) around these energy values. A = 0. Then according to Eq. t&n(2Ka)ER = 0. 64 . (12. 2Ka = nn. Schwabl [246]. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . Also.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. (12. = 2/rn (EER).e. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. 2m0\E\ and Kcot(Ka) = y—^—L. t2 2 (12. The amplitude of the transmitted wave assumes its maximum value where (cf.' We observe that T(E) has poles at those values of E which supply binding energies.50) See e. i. around the values given by Eq. At their positions cos(2/«z)£.34).47) These states in the continuum are called resonances. Quantum Mechanical Tunneling .g. .
The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i.' ' 5a \V / Fig. 12. s \ V0 /• ~\ 'X . = 0.12.2 The PenneyKronig potential.^ l / . 2a<x<2a.2. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.' ''' 5a 3a \ a ~~~. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. We restrict ourselves to the period — la < x < 2a. A potential of this type is known as a PenneyKronig potential.52) Here ER is the energy of the resonance and 2/F the lifetime of the state. T(E) has poles at E= ERi: (12. V(x) = 9(x + a)VQ6(ax). 12.e. 12. This result is physically very plausible. .51) i.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. V(x) x=2a x= =2a .5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. . a 3 a. These are precisely the resonance states.
. ..53) *w«w^( « j)'*^{i i ^ .Vo) n.2 < a . the differential operator is even in x.1.e. . Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.55) The Hamilton operator of the problem. We set .e There is the additional boundary condition (for Vo = oo) ^ ( .260 CHAPTER 12. 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. (12. E^En = ^ ^ t n = 0.2. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo..e. HV^(x) = H.2 = W h 2 _„2 = 2mo(E . (12. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). In view of the periodicity we have ^(x) = ^(x + 4a). i. and hence for = Eip(x).a ) = 0 = ^(a). This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. and we choose (12.56) 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. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall.. = EVip(x).. i. with V as parity operator.
(12. so that (W meaning Wronskian) W[tl>+.60) The boundary conditions that ip± have to satisfy.Be Ka (12.59) Asink(x + 3a) with x & [~2a. V+(0) = const. We therefore have different boundary conditions.il>]?0.2a. 12.. At x = ± a we now have to connect ip and ip' of neighbouring domains. VV(°) = °.63) = KBe~ F = . We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . = Ka ±Asin(k2a). (12. xe[a. ^kAcos{k2a).61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka]. This means that i>±(x) = <  = ±1. We demand these for even and odd wave functions ip±(x) which we can clearly construct from ip(x) (as examples see the dotted lines in Fig.a]. that these are such that for VQ — oo they approach those of the previous case.2). (12.58) D sin k(3a — x) with xE[a. This wave function is no longer restricted to the interior of a box. We > obtain even and odd functions ip± by setting ^ = ±1. which. — a]..62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] .12.a]. of course. BeKX±Be~KX with xe[a.2a]. We expect. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. ±Asink(3a — x) with x€[a.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a. (12. (12. however. — a].2a]. are ^_(0) = 0. we shall not exploit here. The central region of the solution is the classically forbidden domain. V'(O) = const.
262 CHAPTER 12. (with reinsertion of nir/2a for k) 2ak ~ nir . (12.e. For large values of Vb. (12.e.68) .64) we obtain ~[(2ak k . 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.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . say A. k = fc0dd. F ^odd (12. we set tanhx = l .2 a .66) Inserting these expansions into Eq.67) The value of k belonging to the upper sign.nir) + OU2ak . i. = AA. i.nir)2}} = 1 ± 2e~2Ka. 12. where Akoce'2™. and is — as we see — less than that belonging to the lower sign case. for VQ — oo..mr)2}. is the one belonging to the even solution. Taking the ratio of the equations in both cases.— (1 ± 2e~2A + O (—).2 e . In general one now rewrites the equations in matrix form. However.+ • • • 1! cos^(n7r) 2ak .e. (12. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo.3 Level degeneracy removed by tunneling. Quantum Mechanical Tunneling Fig. in the present case it is easier to continue with the above equations. (12. It is plausible therefore to use appropriate expansions. i.e.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.nir + 0[{2ak . + 0 ( e . the odd solution with. say. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—. i.65) and we expand tan(2afc) about nir.even.
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. .. Solution: The wave functions i/>o>Vv in regions V = 0. y(x) = sin and y(xo) = 1.4. exp .12.1: A car's quantum mechanical probability to pass a bump A very slowly moving car of mass mo (kinetic energy almost zero) encounters on the road between x = a and x = b. Example 12.. Fermi [92].4 A car meeting a bump in the road. 0. ^v + ^[E. .^ 0 = °' Hence in dominant order •00 — exp . w(x0) •• = = mogy(x).70) xo = (a + b) 2' . V ^ 0 are given by the equations ^o + " J . . '.2 dxy/V(x) (12. 57. but quantum mechanically there is a finite probability for this to succeed. 1pV 2ro 0 4>o '. (12.e.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. as indicated in Fig. i. . 12. Compute this probability. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig.V{x)\i. 12. b > a.3. 12. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. Classically the car is unable to overcome the bump. exp r rb rb / •J a 2mo (V(x) . 2mn.v = 0. p. 2m0E „ 2mnE .** " W ~^7~ Fig. and is a reformulated version of a problem set in lectures of Fermi.
Ryzhik [122].17)). Jahnke and F. **For instance E.384.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. (12. one evaluates an approximate probability of exp[6. p. n I .055 X 1 0 _ 3 4 J s . 369 and 8. ^ ) = 2 ^ . which give /•TT/2 dx sin'M .2. and assuming a length of b — a = 100 m of the base of the bump.72) Assuming a mass of 1 ton of the car (i. We find™ (with ( . formulas 3. M.l . 950. : 2 ^ B ( ^ .621.e.4 x 10 3 9 ].2 . we obtain ( . We can now evaluate the probability. 1000 kg) and using the following values of the natural constants: g = 9.9191 ~ 1.1.**. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. . p.71) The integral can be looked up in Tables of Integrals. dx sin ' x •• Looking up the value of the factorial in Tables.y) is the beta function (see Eq.1.807 m s .264 we have ro / dxyfV(x) Ja CHAPTER 12. h = 1.1 / 2 ) ! = v^r) TT/2 /. Gradshteyn and I. (15. 14. S. Emde [143]. p.
Finally.2 The Freely Falling Particle: Quantization Under quantization of the freely falling particle we understand here the solution of the Schrodinger equation for the linear potential. This is the quantum mechanical version of the problem of Galilei in one space dimension. * A highly abridged treatment of this problem is given. Then considering the probability distribution determined by the squared modulus of the wave function. for instance. 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.Chapter 13 Linear Potentials 13. 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. the last case allows only a discrete spectrum. Siissmann [266]. Thus. we consider the corresponding case of a particle above the flat surface of the Earth. 13. we consider the linear potential in three space dimensions that forbids the particle to escape. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves.3.1 Introductory Remarks We distinguish here between three different types of linear potentials. 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. whereas the first case permits only a continuous spectrum. in the book of G. This is the problem of Galilei in quantum mechanics. each of which propagates with the classical momentum and energy of a Galileian particle. in Chapter 15. Example 14. 265 . 144. p.* In Chapter 14.
We consider here the case of Eq.t).= l V 27T Joo dxelkxip(x. (13. g > 0.266 CHAPTER 13. The present case is an exception and a good example of its applicability.4b) . 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.= l V 27T J_oo 1 dkeikxijj{k. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.t) = c\x\ + V0.t) = . Recall the classical treatment of the freely falling particle.1) where g is the acceleration due to gravity and we can put Vo = 0.1). (13.2) which is the case with a discrete spectrum.e.2.4a) f°° ${k. and yields rriQX = mog.t) = . V(x) = —mogx. (13.1 S u p e r p o s i t i o n of de Broglie waves We consider a particle of mass mo falling freely in the timeindependent homogeneous field of the gravitational force F(x. We see " that the case g < 0 can be related to the reflection x — —x. from d_ m x2 0 dx + V(x) 0.3) In general the momentum representation is of little importance.t). (13.t) = m0g with potential V(x. The equation of motion follows for instance from the derivative of the constant energy. 13. c> 0.t).t) = mogx + V0. (13. Therefore we use the Fourier transforms il){x. (13. i. x > 0. A different > "linear potential' is the socalled "confinement potentiaF V(x.
13.6) and wish to solve it.t) dk nikx k=oo '2K (im0g)il)(k. Applying the total derivative d/dt to this function.t) the expression (13. d \ ~f1 dkr[ k + m0gt ^r't (13. we obtain the differential operator on the left of Eq.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.t)im0g^k.3) yields us now by inserting for ip(x.7) applied to every quantity in Eq.6) the argument k is then replaced by this. (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13.e.t).d m +imo9 .6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.8) With this result and the substitution (13.e. multiplication by a test function and subsequent integration). (13. i. (13.< (13. • fd ~(' m0gt %h . 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. To this end we consider the following substitution in Eq. We rewrite this equation in the following form th (13.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq. (13.7) In the wave function on the left hand side of Eq. Equation (13. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i. (13.t).t) dketkxiP(k.t) h2k2 d ~J(k.t) m0gx /2TT m0gxip(x.6).9) .
(13.6) to the solution (13.6) is then $(k. t ) .1 we verify t h a t this solution indeed solves Eq. Thus we have first (the second term arising from the upper integration limit. (13.10) k by m0gt back : k —> k h The solution of Eq.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. We have thus obtained the solution of the transformed equation.11).1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. i.t).t ) ' U ( f c .11) in a neater form. Our next step is to reexpress the solution (13. we have to reverse the substitutions and replace in (13. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k. On the other hand the second operator expression implies dtj>o imog^—ip(k.gh [ d t ' j f c + ^ ( t ' . We define the wave number kt as kt. (13. Solution: We apply the two operator expressions on the left of Eq.6). E x a m p l e 13. (13.= k rmt. (13.t In Example 13. the following result h 2moi tp[k\ ^—. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.268 CHAPTER 13.11) solves the partial differential equation (13.6). We observe t h a t t h e solution is actually a function of kt — k — mogt/h. To obtain the solution of the original equation (13.6). Thus addition of both indeed leaves the expression on the right hand side of Eq.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc.t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.e.11) *[k?f.6).10) Note the amplitude function ipo(k) in front.i) .12) .
17) = J_/ C X ) h2 [It(k) . m0g (13.t) defined by Eq.11) can be written j'A ^..19a) .2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.o ut = 2mo —kf 2mo A.} fA ' _ _tkt 2 k+ t) 2 k .14) (13. m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13. we obtain rp(x.I0(k)].16) Next we evaluate the Fourier transform of ip(k.= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13. . (13.4a).11).t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . (13.t) into this integral and recalling the property (13.t) = .2„2+2 mfaH 3^3+3 .— [It(k) . Inserting ip(k.13.gfci + 2 g2m0 2/i 2 ' (13.15) we obtain / = . we have ip(x. ./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.18) h =k mogt p h S' h . we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .13) rn0_t+2 + mlg2t3 .+^f (13. _ k t  h2 3 Inserting here the explicit expression for fct. 27T ^ .o o 2 —  exp iktx h mZgH' .
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.t) as 1 I00 ip(x. 7 u>t'dt' (13.v = kh/mo.19b) h3 3 we can rewrite ip(x.t) =y= dkip0(kt)) exp i< fctx V27T J . (13.19a) exactly as for a falling particle. (13. Linear Potentials f dt'. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .The associated kinetic energy of the particle is mv+ = huJf.2. (13. fm0gt\ \ 2h ) (13. 13. Such a decomposition is possible only for homogeneous force fields.21) In other words.ut> = Jo t2 g2m0 i 3 k t . 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.t)\2 for t — oo (distant future or past).t) as a decomposition into de Broglie waves. These wave vectors vary in accordance with Eq.o o Jo This result can be interpreted as follows. the relation (13.270 and CHAPTER 13.20) expresses the wave function tp(x.20) (13.18) into a • different form in order to be able to perform the integration with respect to k.23) .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 . To this end we convert Eq.
/m0gt\ \ 2ft J ° ) I m09H3 ft J 6ih m0g2t3 _ m0x2" 2Aih 2hti m gt IT / dk^oik t) exp oo ht k2mo« ~fti 2ft~] (13.28) (for a verification replace a by a + ifi.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ .18) thus becomes ip(x.QX 2 ^ 0 7 (m.26) with the formula 1/2 (13. (13. 2h h 24ift ' mogt^ r m.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics./oo dkip0(kt) exp l x ixI exp oo .t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. (13.t) 1 exp r .29) . may be put in front of the integral. Then ip(x./m 0 a.2 The Freely Falling Particle: Quantization The integral (13. It follows that ip(x. TOO#A ft and. integrate and then take the limit fi —> 0). ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. m0xcl ht m0gt 2h gt O v: °° 2hti (13.24) We now set fit / 2m 0 V" 0 V m0a. (13. rnogt m0g2t3 tX^r~.13. fi > 0.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.t) = 1 1 oo 271 27T /.26) d£e ie £ kt^ m0' 0. being now independent of £. We evaluate the Fresnel integral in Eq.e.
(13. Linear Potentials Replacing here x by the large time value x c i. For stationary states we set xl){x.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.1) has only a continuous spectrum. xc\). (13.e.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 .t) ~ W —exp i V irlt m ktxd J cot.dA 4>o(h (13.31) and obtain from Eq.1 A turning point at x = —XQ.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. 1/2  ±^(2mlg/h^x^ / 2i (13. These are incoming and outgoing waves. 13. kt and ut 13.33) oo for E <C \mogx\. this expression can be brought into the following form: if *\ 0 ip{x.t) = eiEtlhct>{x) (13.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0. For x exponentially increasing and decreasing solutions are possible. V(x)=m 0 gx V=E Fig.272 CHAPTER 13. .
t). so that ih dtp{k.e. 13. cf.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. E)dk.1. but also d 00 dt Eijj(k.6) that also for 4>{k. at which E — V = 0.13.0 0 dkelkxE^(k.4b).e. H2k3 4>{k) — cexp Ek (13. i.3 The TimeIndependent Schrodinger Equation 273 Classically the particle coming in from the right with energy E cannot penetrate into the potential barrier since (where V > E) with conservation of energy E — V = p2/2m.e. probability. its momentum p would be imaginary there.e. 1 f01 dketlcx^{k.e. 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. The Fourier transform is given by the relations (13. t). However quantum mechanically this is possible with a small. i. (13. exponentially decreasing. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation.t) = dt (13. Akx (13. i.t).34b) It therefore follows from Eq. i.35) It follows that /W tm0g \ 2m0 i. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function.34a) Then d &k 1 f°° ih—il>(x. J \ „ (13.36) m0g \ 6m0 with c = const.4a) and (13. at which the classical particle bounces back. we . The point — XQ.e. is reflected. i. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. t) = — = / V27T J . which is not possible for real particles. and hence is continuous and extends from —00 to +00. t) = Eil)[x. Fig. and we can establish for the resulting continuum solutions the orthonormality and completeness relations.00].
(1340) i.£') = <*(#£')• 1 27rm0g' = /o . Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). (13. / <>~ikx VZ7T J (13.37) (eikxdk. i.36) into (13.39) dk4>*E{k)4>E. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).L dxeikx(f>E(x).41) J—oo Next we insert (13.38).L [ dkeikxMk). / dE(j)E(x)(t>E(x') = ^ s .36) (with c replaced by (13. (13.40)) into (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 .e.M *K Joo Setting dt cos ( ts n Jo st). the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').38) we have for <j)E{k): OO /"OO / oo (13. </>E(x) = .39) we obtain dkcc*exp m0g It follows t h a t cc = &(# . h? h2 . (13. In a similar way we can verify the validity of the completeness relation...(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.a : ' ) .43) y=2^x+2^E.e. V 2lT J Mk) = .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. apart from a constant phase which we choose to be zero. 13.
For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. the function 4>E{X) i and setting s y = — = 1 ./ 2m 0 E (13.^x + ^ E (13.^ . (13. .13.3 The TimeIndependent we can write <J>E{%)'4>E{X) = Schrodinger Equation 275 2ir^/m0g J dkexp h2 yk — k '2m20gV~ 3 W i t h the substitution k = [it we change the variable to t. / s M .43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x) s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) r 2^rJ 2W dxe^Qe^^eU*3*3) _^Pi(Sy~tz)ei(s3t3) S(y = j£j**w I ^LJt{yz) 2TT _ We will encounter the Airy function Ai(s) again in Chapter 14 in the matching of exponential W K B solutions to periodic W K B solutions.. (13.J * .46) This follows since with Eq. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.45) One can now derive normalization and completeness integrals for the Airy functions.44) is then expressed in terms of the Airy function: . whereas the other is of exponential type. so t h a t the integral becomes 4>E{X) = 2vr ^/mQg / n oo dtexp oo h? * 2m£g y \ 7 5 * " 3 M' Choosing /2mfo\1/3 A= ^ . and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. and we shall see t h a t one branch has periodic behaviour.
.e.2 The contour C for s < 0 with ends in the angular domains (a) and (b).r 3 sin 30 ).51b) . i. . (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . (13.f dzJ*W 2TT JC with z = reie. o (13.49) (13.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 .. Linear Potentials 13. (13. (13.\zz.. (13.276 CHAPTER 13. i.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.TT < 30 < 0.4 The Saddle Point or Stationary Phase Method The integrand of Eq.r 3 cos 30 J + i ( sr sin 9 . 13.TT < 6 < 0 for r > oo.43) contains the phase (with x — z): > *(*) := sz . or n <6 <TT o for r > oo. We want the integral to exist and so desire that lim e**^ = 0. Thus we must have sin30<O.e.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.
Inserting (13.48).e.57) $"(z s ) = 2% s. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. at which the phase becomes stationary.z3s and V^s + . Such a piece can be found in the neighbourhood of the socalled saddle point zs. i.( \ / ^ ) 3 2z.e. oo] as indicated in Fig. provided its ends at infinity satisfy these conditions.51a) and (13. Then Ze TeC : V^se±i7r/2 = ±i\fzrs.45) exists for the contour C. Then 6S = ±7r/2.54) where a is the new variable with —oo < a < oo. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.S )V2 (13. (13.55) into (13. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.2. (13. (13.53) Since we shall have the contour C in the lower half plane. (13. 13. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(. and the main contribution to the integral comes from the a region around the saddle point. (13. Then the ends of the contour lie in the domains specified by Eqs.52) = exp(±i7r) Let s be real and s < 0. where the derivative of the phase function vanishes. 0 = &(z8) = s~zt s = za „2„2i6>.54) i ¥~s)3/2' (13.13. —i ${zs) = szs .4 The Method of Stationary Phase 277 The integral (13.51b). we choose zs = iy/^s. and we set with a real: z := zs + a (S)V4' (13.58) .55) s < 0. so that exp(±2i9s) — 1 and — s — rl. i.56) 2n(s)^Jc But I daefr"?*"^.
2 we show that for s — oo: > Ai(s) 1 .42) into an equation with a more appealing form.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.s ) V 4 exp \(s) 3/2 (13. S.32) therefore becomes dz2 + zd) = 0.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). According to Eqs.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._ 1 .44) we can convert Eq. as one can see from the following equation^ ri' + Ffz •2^2. cos .56) decreases exponentially around a — 0 for real values of a.45) 4>E(X) ~ Ai(z).2/32u = 0 (13..491. Ryzhik [122]. V 2h2 (13. (13.60) The reason for describing the point zs as a saddle point is that the integral in Eq. M. Sec. > so that Ai(s) 2 v ^ ( . (13.44) and (13. (13. (13.64) tSee I. For (—s) — oo the integral converges towards T/TC. With the help of the substitutions (13. . Gradshteyn and I. This is an important aspect in the WKB method to be considered in Chapter 14. V^s / 4 V3 (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. (13. 8.59) In Example 13.278 CHAPTER 13.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). 2 s 3/2 . where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}.
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. (13. P Introducing a new variable a by setting p = ai\a\. Example 13. cos  . we obtain f° Joo dp=—[a+ i\a\]da. Then.s 3 / 2  . Eq.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).51a) and (13.^ e ^ / V2 4 [°° Jo dae'2^. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2._ . . the conditions (13. 27T Ic Jc 3 This time s is real and positive.z ± ) 2 * " ( z ± ) .59) and (13. with *"(z) = 2z. changing to the variable p.4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. It follows that for s —> oo: Ai(s) ~ . Choosing the contour C to lie along the real axis. 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 . kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = . Thus there are two symmetrically placed real saddle points on the real axis. * " ( z ± ) = =F2Vi.13..51b) are satisfied.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.
~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq.3.3: 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.Be 2 Fig.111). (11. 13. ^/2KJW1. Jo Jo \ . . the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. since f'(zo) = 0 and (z — ZQ)2 = —x2. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. and hence integrating as described above through the saddle point indicated in Fig. Linear Potentials E x a m p l e 13. i.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). Thus here / ( z ) = — z + n l n z . Then.e. 13. Solution: Briefly.*.280 CHAPTER 13. z o = n . n!= / e~zzndz= / e~z+nlnzdz. / " ( z o ) = —n/z2 = —1/n.3 The saddle point at z = n + ix.
Jeffreys [144].§ 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. Froman and P. A critical overview of the history of the method. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i.317. A. It goes without saying that such comparisons are very instructive. named after its first promulgators in quantum mechanics: Wentzel. O.Chapter 14 Classical Limit and W K B Method 14.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. of misunderstandings and other aspects may be found in the monograph of Dingle.295. Wentzel [282]. ' T h e most prominent earlier expounder is H.g. The WKB result is therefore frequently described as the semiclassical approximation. Dingle [70]. however. the linkage of solutions above a turning point to those below. where *G. Kramers and Brillouin. for misunderstandings pp. N. 281 .e. ^ The central problem of the method is the topic of "connection formulas". Kramers [153] and L. § See e. i. 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. Froman [99]. 316 .* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. *R. will there be that of a coupling parameter. For the history see pp. been developed previously by others in different contexts.e.* The method had. Brillouin [37]. H. 292 . and the role played by h in our considerations here. B.
It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.282 CHAPTER 14.* We illustrate the use of this condition by application to several examples. 14.1) (14.g. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. for which the observation of position and momentum is independent of the time at which the observation takes place. the function S being called phase function.t) = eiEt/h7P(r) and \V(r. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T). Classical Limit and WKB Method E — V changes sign). (14.3) .t)\2 = \iP(r)\2.V(r)]^(r) = 0. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. the energy has a definite value E with V(r. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E .2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. In this equation we set. A(r) and S(r) being real functions of r.7. the derivation in Example 18. For stationary states ^/. V>(r) = A(r)eiSWn.
1) and separating real and imaginary parts.14.13) .10) m 0 ^ .t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.t)}2.6) 2mo dt now with A also a function of t. (14.10) with a continuity equation by defining as probability density p:=*** = [A(r. (14. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.+ VAVS + AAS = 0. h h (14. (14. We can identify Eq.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq. These equations are equivalent to the Schrodinger equation.4) (14. We set in this equation. V(r. (14.t)eiS{r't)/h. (14. h2 ih— = H$ A + V(r) * .5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation.12) VA A (14.V(Y) + and 2VA • VS + AAS = 0.* VAVS+AAS + V(r)A. (14.Q A 2i 8A 1 (14.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *. we obtain the following two equations: E .7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.
16) BA 1 moA—.J = 0. + h2 AA V = — ^ .19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. ie. (14. Next we investigate the significance of Eq.21) .284 CHAPTER 14. (14.14) A2VS.9) as 9S 1 mo^ 2 + T. (14.15) and (14. (14.01. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.20) With this we can rewrite Eq. we have mo—(A2) + V(A2)VS + A2AS = 0. (14. .17) With Eqs. mo <™> (14. (14. m0 But now (14.10) we obtain = — [V(A2) • VS + A2AS}. Since J = — A2VS = PVS. TTIQ (14.9) for the phase function S. the implications of this equation are also those of the equation of continuity. Since we obtained this equation from Eq. (14.n. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. (14.+ AVA • VS + A2AS = 0. (14.18) ot ot This is the equation of continuity that we encountered already earlier..16) this becomes m 0 1 rr + m o V .10). ^ + V • J = 0.
e. that the quantum theory implies an essential singularity at H = 0 (i. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. we obtain ^ V 5 +  v ( v v ) + V V = 0.2) for the probability amplitude. The particles have no interaction with each other.23) (14. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. with the help of Eq. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. i.22) (14.1/h. a singularity different from that of a pole). and hence''' ^(m 0 v) + m 0 (v.e. This expression with h in the denominator shows. so that V ( v • v) = 2(v • V ) v + 2v x rotv. We deduced this equation solely from the phase S in the limit h — 0. d — = (14. V • J = 0. the wave function \I/ describes effectively a fluid of particles of mass mo. which for very small values of h yield the dominant contributions.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. In the first • place we require the ansatz (14.24) d 1 dr d •— = d Lv _ . where. they are asymptotic expansions. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). dt 2 Constructing the gradient of this equation. r o t v = ( l / m o ) r o t g r a d S = 0. However. This result is very general. (14. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0. The equation of continuity then describes the stationary flow of a fluid.V ) v + V 7 = 0. This means. 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. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. (14.20).14. . Such series are typical for expansions in the neighbourhood of an essential singularity.
i. Sec.5)) and 2 ^ f dx ax + All=0. the method is widely familar under the abbreviations of the names of Wentzel.30) ax *For details see R. z (14. Kramers and Brillouin. Messiah [195].26) We observe already here that an expansion in rising powers of k. 318. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).3 14. As remarked at the beginning. .3. p. is an asymptotic expansion.In A(x).4) and (14. similar to our earlier procedure with A and S real functions. We consider the onedimensional timeindependent Schrodinger equation y" + ^. such as "LiouvilleGreen method' and "phase integral method'. or h2/2mo(E — V). We set.2. (14.27) (14. There are also other more specific descriptions. i = eiW(x^h. 6. (14. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large.e.§ y = A(x)eiS^'h where W(x) = S(x) + . Classical Limit and WKB Method 14. B.$ 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. Dingle [70].286 CHAPTER 14.[EV(x)}y = 0. (14.28) The calculations we performed at the beginning imply the following equations (compare with Eqs.1 The W K B Method T h e a p p r o x i m a t e W K B solutions Without exaggeration one can say that (apart from numerical methods) the method we now describe is the only one that permits us to solve a differential equation of the second order with coefficients which are nontrivial functions of the independent variable over a considerable interval of the variable. §We follow here partially A. The expansion in powers of h2.
.29) becomes (note that this is still exact!) (S')2 . Eq.32) is therefore (S'0)2 + 2H2S'QS[ .35) .34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .3 The WKB Method 287 Equation (14.35) can be integrated. (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 .2mQ(E .14.10)).V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .32) s" = sz + h2s'( + •••. 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. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.28). (14.\ In S'.36) Equation (14.2rn0(E V) = h2 lS'A 2S'0 (14.V) and (14. (14. Since the expression (14. + (14.33) 3 (S")2 .31) \a.30) corresponds to the equation of continuity (i. Li This expression can be substituted into Eq.A = const. (14.e. 3 / 5 . (14.4 (S>)2 1 S'" 2 S> (14.
y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13).27) we obtain now with Eq.39) where a and /3 are constants. Eq. .' For the expansion (14.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .19).I In S' + In c h 2 2 2 exp ^{S0 + h Sx) .E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. E > V(x).33) to make sense as an asymptotic series.and So = ±h / — — A J k(x (14. (14.41)? We note first that the expansion in powers of h2 does not converge. Dingle [70].288 CHAPTER 14. it is reasonable to set (cf. (1.40) rx dx v "7 W)\. i.37) For Eq. B. we expect that in any case So » ft2Si. Classical Limit and WKB Method has the form of a momentum.e. 1 S e e R.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). (14. Chapter XIII.41) We now ask: What are the conditions of validity of Eqs. and instead is asymptotic which we shall not establish in detail here. p = h/X) k{x) :Then S'n +. y/2mo(V(x) . (14.38) h y/2mo(EV)' (14. The solution is seen to be periodic provided k(x) is real.39) and (14. (14. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. (14.
45a) . Since k(xy 1/52\S'0 (14. we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14. E > V(x).1 / 2 ]" [§0SS). Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .14. (14.V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.3/2 s&T + 3 {So Il« we have (cf.36).43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A. Eq.42) [(^).
In terms of X the original Schrodinger equation is y" + = y = 0.290 CHAPTER 14.2 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). and hence (with UIQ = 1/2.V(x) = V(a) . at which E — V{x) changes sign? We note first that in this case near x = a: E . around the classical turning point. In a similar way we have for E < V(x): l'(x) <C 1. (14.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. (14. In the neighbourhood of E = V(x).46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. h = 1) .a)V'{a). i. *2 A(z) = V / 2m 0 (£ . How do the solutions of Eq.V{x) ~ Or . V(x)) (14. 14. Classical Limit and WKB Method we can rewrite the inequality (14.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. Roughly speaking one therefore requires E to be large in both cases.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . the WKB approximation is in general invalid (see below).e.48) The (dominant) WKB approximations are exact solutions of this equation. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. (14.48) behave in the neighbourhood of such a point called "turning point'.3.
14.3 after stating them first here.e. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. in the domain E < V the WKB solutions are exponentially increasing or decreasing. 14.48). On one side. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . 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. (14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a.1 there are WKB solutions on either side of the turning point and some distance away from it. 14. As indicated in Fig.1. since this requires different approximations there. V=E Fig. 14.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. The transition is provided by matching relations.1 The regions around the turning point at x = a. We assume we have a turning point as indicated in Fig. 14. in the domain E > V the WKB solutions are oscillatory. and on the other side. i.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x.3. We shall not derive these conditions here in detail but rather make them plausible in Sec.1. (14. which connect one domain with the other. 14.
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. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. This means the asymptotic part suffices only in t h e exponentially decreasing case. has the same asymptotic behaviour in the domain i « a a s Ay\.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. as in Fig. however with the limits of integration x. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right.H— J =^ v/exp + m (14.1) is given by the following relation with the + sign in the argument of the exponential on the right and no factor of 2: VACOS I / —. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . Thus. (14.52) For a potential rising from left to right.50) x S> a. one could simply replace everywhere /•a / pa • • • b y / Jx J x .292 CHAPTER 14. in order to make the formulas independent of whether the potential is rising or falling. 14.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. Note also the extra factor 2 in yi. (14. a interchanged. 14.
y" + Toy = o. 1 ~ (x . 2 1 X (14. (14.3.2 The overlap regions around the turning point at x = o. 14. the equation V approximately linear around turning point V=E Airy solution WKB . the equation becomes approximately y" = (xa)xiy. Therefore we consider now this original equation in the domain around x = a. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.48).53) Since at x ~ a.54) .51) — always with the decreasing exponential on the left — is valid in both directions.46).Airy overlap* Fig. the matching relations have to be altered accordingly.3 The WKB Method 293 In that case Eq. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. whose exact solutions are the dominant WKB approximations.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. 14. If higher order contributions are to be taken into account. (14. as we saw. (14.e.a). is singular at a turning point x = a. i.14.
J (X (i4 58)  Replacing in Eqs.a)X\/3.57b) UV2 3 11 by J dx . (14.60).Bi(—z) for \z\ — oo. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. Cf. > for a. we have r x * f x ' /2(x a)i/2d*=^(x .59). (14. A second solution is written Bi.57b) To the same degree of approximation as the equation y" = (x — a)xiy. or M.294 CHAPTER 14. In the following we require the asymptotic behaviour of the solutions Ai(—z). Stegun [1]. . (13.55) the equation becomes the Airy differential equation d2y dz2 zy.1 / 4 by \^. — a ^> 0.^ 4 I 7T x — a <C 0. (13. 448. B.4. (14.a)3/2 ^ /2 =^ 3/2 o Q. This is given by the following expressions. A. In particular we encountered a solution written Ai.5960. i I 772 \3 + > for ia>0. as > may be seen by referring back to Eqs. With the substitution z = (x . (14. formulas 10. (14.56) We have discussed some aspects of solutions of this equation in Chapter 13. p. R. Dingle [70]. p.291. Abramowitz and I.57a) and (14. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant.
we enter the domain of validity of a WKB solution.3 The WKB Method and 295 •. These solutions. by Ai(z). In this way different asymptotic branches of one and the same solution are matched across a turning point.14. are only valid in a small interval around the turning point at x = a as indicated in Fig.3 The anharmonic potential well.l v/2m0(EV(x))dx 7r = (2n + l).2. 14.l.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.52).1 to the quartic oscillator. Example 14. e. In the direction to the left as indicated in Fig.2. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig.(zf2 by J we obtain the matching relationsX (14.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = .51) and (14. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. As an illustration of the use of the WKB solutions we apply these in Example 14. n = 0. In this sense we have now verified these relations. 14.59a) . and hence must be proportional (in view of the uniqueness of the solution there). In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). 14. however. The latter domain is a small region where the solutions from either direction overlap. We summarize what we have achieved. As we approach a limit of this interval in going away from the turning point. (14.g..
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.3. (14. (14. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). In order to see this we consider r dx TT\ _ Jx T " 4 J ~ COS / r" dx_ _ Via ^""1 ^~4J" COS Ua I + 4 ~ i ( . (14. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. We argued earlier that the WKB method is suitable in the case of large values of E.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 . 2 n = 0. b are the two turning points. . 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. Classical Limit and WKB Method where x = a.(x) = c ' v Acos I / — I for a < x < b. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. The condition for this is ya{x)=yb{x) for xe(a. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. Hence we search for solutions which are exponentially decreasing in the far regions. (14.59b) Solution: We consider an anharmonic potential well as depicted in Fig. Sometimes this distinction becomes imprecise. 14. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.e.44) this implies large values of n. According to Eq.296 CHAPTER 14.60) Using the matching relation (14.b)..51). for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. Then around these points the potential would be well approximated by a linear potential (as we saw above). . Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator.y o dx TT T provided 6 _ 4 /. Let E = V(x) at x = a and x = b. 2 .l. (14. Here we are interested in the discrete states (the only ones here). regions 1 and 3 in Fig. as in the case of the harmonic oscillator itself. 1 . In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a. /n = s u ^ TJcos^ya TJ+cos^a T J .3. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. 14. i. . ^ A = (2n + l ) .. .^ .2. j/i. . n = 0 . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. extends only over a length of very few wavelengths and is therefore valid for n small.
The corrected form* is. In this socalled "old quantum theortf (i. Sommerfeld and W. This relation is sometimes wrong. B. and knows of no published form or script.e. 2 .x number of turning points TTH.4 Three cases with different pairs of zeros. 2. A.1. Dingle.4 BohrSommerfeldWilson Condition 297 14. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. .64) (14. ri2 I Jqi :pdq n + 1 — . equivalently. n = 0.63a) or. Wilson by supplementing classical mechanics by Planck's discretization. (14.63b) *The author learned this in lectures (1956) of R.1. . . . with n — 0. Bohr. 14.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N. (14. (c) V Fig. pdq = n + 1 — .14. .x number of turning points h. 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.
In quantum mechanics we have dq2 + Ki/j(q) = 0. (14. where n —. Before we continue we recall from Eq. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h.4(a). We have the case of trigonometric solutions. this is an exponential zero. i. In this > case the wave function for g > go is . together with q — — oo.e.e. if /(g) is negative. as illustrated in Fig. / • n an integer > 0. this means dqyf—f = (n + l)7r. (14.65) We use the knowledge of these solutions to find the eigenvalues of the equation.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. Since sine and cosine have zeros spaced at intervals of TT. 14. this is a trigonometrical zero. Case (a): The case of two trigonometrical zeros. the case of tp(q) — 0 at q = gi. 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.66) J<1\ Thus in this case the socalled old quantum theory is correct. 14.4(b).298 CHAPTER 14. i. where /(g) is positive. where /(g) is negative. 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.e.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. i. i. 0(g) ex .67) Case (b): Next we consider. There are three cases. Jqi ^ fpdq J =2 pdq = (n + l)h.
3. (14. and • • for the wave function to be exponentially decreasing to the right of q'0.63a) or (14. E x a m p l e 14. 14. or Pdq (n + l ) " 2 h. For the wave function to be exponentially decreasing to the left of qo and using Eq. this is the case of two turning points. Hence we have pdq= i n +  . As illustrations we consider Examples 14. (14. Solution: As in the case of the quartic potential. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.e.63b) to obtain the eigenenergies of the quantized harmonic oscillator. i.e. (14.2: The harmonic oscillator Use the relation (14. the sines must be proportional.69) pdq = (n + l ) " 4 h.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1).2 and 14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r.4(c).63b). Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT.14. ri\ / J do y/f(q)dq (n + 1) 1" (14. Thus in this case the socalled old quantum theory is wrong.68) (14. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq .70) We can therefore summarize the results in the form of Eq. i. Therefore their arguments differ only by (n + l)ir.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa .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. Since both cases refer to the same region qo < q < q'0.
71) For m 0 = l x 1 0 _ 3 k g . as illustrated in Fig.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. so that E = hu{n + 1/2). 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 a trigonometrical zero is the condition of absolutely no penetrability beyond it. 14.5. Calculate its quantized energy. .300 With potential V(q) = 2ix2mov2q2.5 A particle above a flat Earth. 4 / x CHAPTER 14. where g is the acceleration due to gravity.V(q))dq The result is E/u. permit this although with rapidly diminishing probability. :: V=m0gq Fig.807 m s . The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . and h = 6. one obtains E ~ 2. Example 14. Thus there is one turning point at E = mogq.2 3 J x (n + 3 / 4 ) 2 / 3 . g = 9. whereas a turning point does. 14. t Hence we have the case of the integral J> pdq (n + 1 )  Inserting the potential this becomes fqo=E/™0g 2 / dq^2mo(E Jq: Jq=0 Evaluation of the integral yields the energy 1/3 2/3 — mo m) • (n + 1) En = ( — m0g2h2 n+ • (14.63 x 10~ 3 4 J s .28 x 1 0 . in principle.2 .
= — <*— = h(n+\. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' . d dE V)={n+)n. Chapter 16) one obtains the quantization relation (for Mo = const.5: Period of oscillation between two turning points Use the BohrSommerfeldWilson quantization rule to obtain for the period T of oscillation of a particle of mass mo between two turning points the relation T 1 m0 f dx d ( — = . Solution: The proof is contained in Example 18.( 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. (14. E x a m p l e 14.5 Further Examples 301 14. In the case of screened Coulomb potentials (cf. and u> is the oscillator frequency in either well.5 Further Examples E x a m p l e 14. 1 .75) which verifies the formula immediately. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V).2.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.14. 9 .l.72) \\2dxp=Qdx^{Eh Hence.f V ft h JXl h 2TT (14. with p = modx/dt. (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 £ = .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 . ra 2/ = 0.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.73) Jx! rp2a dx I^<E 2 v) = ^ n — = .7.
6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). (11. where 1 2 • 1 2 y = Q. The q term in this equation can be removed by transforming the equation to a cubic in y.77) Determine (a) the turning points.302 and T CHAPTER 14. 14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.133)). which determine the imaginary part of the energy. Example 14. (14.78) (a) We determine first the three turning points at qi = <?i. where .<?2i<?3 for E > 0 given by E — V = 0.a .79) .6 The cubic potential. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. i. q = y+ a .6. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.y2)(y . Fig. (14.~ \ (14.e. The equation E — V = 0 then becomes after a few steps of algebra y3 ~ Py + Q = (y P = ^ a 4 and Q = 2 ^ yi)(y . Eq.y3) = 0. 14.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.
B y r d a n d M . gi ~ y {V3e} < 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 . ' P . 91 < 93 < 9 < 92(14.78)) c o s 30 •• 4P3 ( ' \ aeb2} .93 2a2 [ .= s i n t 3 v 3 (14.9 2 [ .{ V 3 e } > 0. p . 80 a n d 361.ui=K(k).6.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ .q)(q .\tan0^ v/3. 14.14. 92.83) T h e i n t e g r a l m a y n o w b e e v a l u a t e d w i t h t h e h e l p of T a b l e s of I n t e g r a l s . f o r m u l a s 236. a n d t h e a n g l e 0 = 0° t o E = £ m a x .a 2 [ l .9 l ) f f / du s n 2 u en2 d n 2 u . we h a v e qx = .k 2 2tan# \/3 + tan! .0^6O°.< 1. 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.04. W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . we set 0 = 60° — e.120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . M i l n e .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.§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. 78: sin 120° = v ^ A c o s 120° = . tan60°=%/3. p .81) m a 91 . D w i g h t [81].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 . V cose' + sin 8 (14.84) 92 . i. ui=sn 1 \l— — = l. (14.2/3 i n t o q = y + a 2 / 3 . q3 ~  .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.g i < 1. (l_fc' 2 a + fe'4)l/4. 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 .120°)] = . 0<6O°. 6 = 0 i m p l i e s E m a x of E q . T h e r e f o r e E = 0 c o r r e s p o n d s t o fc2 = 1. (14.T h o m s o n [196]. p .c o s O + cos(6» .85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. .2 c o s ( 0 + 1 2 0 ° ) ] .e. (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. B . g 2 ~y{3}>0.gi = 1. e > 0. 212. 92 = a2[l .9 3 — = = 2 " 3 " [ . D . (14. 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. M .79) a n d t h e n s e t t i n g m o = 1. 37.cos 0 + cos(6> + 120°)] = a2 92 .' : ± 6 0 ° w h e n E = 0. p .g i ) ( g 2 .80) I n s e r t i n g t h e r o o t s yi. H e r e cos(<9 ± 120°) = . 93.c o s ( 0 + 12O°) + c o s ( 0 .08. (14. F r i e d m a n [40]. V3 + t a n 0 2 .3/2. _ \ / 3 — tan(9 2 V92 .93).2 c o s ( 0 . § T o h e l p we cite H. q3 = a2[\ .2 c o s 0 ] .1 / 2 . F .
the integrand of the integral across the well is effectively the inverse of this momentum. E(u) the incomplete elliptic integral of the second kind.^ V3 / V V3 y a / c o s e J. V?2 . formula 233. (24.87) Inserting G(fc) together with the expressions for the prefactors into (14.91) a2[cos0^] y/3tanfl ? L .00. 0 (14.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . V 93 . Thus Iwen=lgF(ip. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . . D.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . 72. 2 2 . Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —.32)). 1+72 = and J o s 0 + ^ = (1 . . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d .86) 15 tan. as in (14.) .k).sm t %/3 £\. Friedman [40]. F.91 F(TT/2. p./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. G(fc) =86aS (ITS£ li[fe. E(u = 0) = 0. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . sinV = q3 qi ~ = 1.2(fc2 . (14.85). E(u = K(k)) = E(k). A:') = # ( * ' ) • (14.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. The WKB Method elliptic integral of the first kind. Eq.fc'2 + fe'4)"1/4. (14.qz) (14. (14. We can evaluate this integral again with the use of Tables of Integrals.92) ''P. g= 2 .304 CHAPTER 14. Byrd and M.?)(?3 . 91 < q < q3 < 92 bJc.~ y ^ ) a< [ COS 5 + .= —^ = k a2[cos0+^] v ^ + tanfl V=".^ + ^ ) (l + fc'2)2 .S .91 . Whereas the integrand of the above barrier integral is effectively a momentum.84) we obtain Carrier = 2a 2 sin6> 2 b( . with K(k = 1) = oo. Next we evaluate the required integral across the well from 91 to 53 at energy E.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.
4 2n + 1 fc' = 4 — ^ — . we obtain E a6b2k'4(l + k'2).l. 2a 5 b K(k) + 2 E(k). 32 (14. (14. we obtain 15 . (14.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq..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 (14.63b). . To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). to represent a physical decay rate. S.93) where w is the harmonic oscillator frequency in the well.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.2. (14. a°b n = 0. formulas 8.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . Gradshteyn and I.14.113.3 and 8. we obtain > .5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. pp.3.** However.80). (14.96) Expanding the coefficients of the elliptic integrals in Eq. . M. so that even in the case of the ground state the zero point energy will contribute to the prefactor. Ryzhik [122].114.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14.85) (14. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. 905.88) in rising powers of fc' .! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . 906. 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. With q = 1/z the integral (14. Thus one obtains from (14. (14. and hence at E = 0.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1.
the result agrees with the ground state path integral result (24..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.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.e x p [ ..178)) call the Furry factor set equal to one.j . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . i. i. 2n + 1 — / 26a5b \ . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . _ exp[2/barrier] >.96) for the imaginary part of the energy as in the Breit. also Eq. (20. this has not yet been done.e.fc'2 + fc'4)1/4 * ± t ^ .„.100) 'barrier = ~0°b In [ ^ — j .— (14.ba / 25a5b\n+2 ±l_l eT5 a f >. (14.> tf = = 2/wfiii = ±tv8a°b——P. n ~ V27T . _ T T • (14..e. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.Wigner formula (10.306 CHAPTER 14. .2 7 b a r r i e r ] Z Z7T where (cf. It follows that (for one complete orbit back to the original turning point) .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. so that we obtain 4 . We observe that with \j2je as 1.. 15 a .73). 8 Bfc .102) . . (18.101) (14./^~^l^ab ^aub .( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . V (14. 7 ~ —.J + 0(VoJb).7 . (0) i hw En = E\> . The WKB Method For k' —• 0 the first and the third terms dominate.
[233] and H. This potential became widely popular in the spectroscopy of elementary particles. the Coulomb potential. indications from the nonabelian generalization of quantized Maxwell theory. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. Quigg and J. led to a spectrum which contains only scattering states.1 Introductory Remarks The particular linear potential we considered in Chapter 13. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. in fact. [231]. L. 307 . Rosner [230]. D. Yan [82]. T. that this is indeed the case. K. Lane and T. which permits only bound states. C. Kinoshita. at least indirect. See particularly E. B. K. L. the classification of the various states of nucleons and mesons and other particles.Chapter 15 Power Potentials 15. The study of this potential.* 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. along with inclusion of angular momentum and spin effects and their interactions. Eichten.g. Thacker. Rosner [269]. Quigg and J. i. after the realization that quarks as their constituents might not exist as free particles and.e. There are numerous. now known as quantum chromodynamics. Gottfried. 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. C. describing free fall under gravity. [232].M.
£ ] * ( r ) = 0. This force maintains this value irrespective of how far apart the particles are separated. ~ (I + l)rl (15. m\m<i M= : .u ~ r i + 1 (/ + l)R(r).1.e. f dr{u(r)}'2 1.m2. and u'(0) > [u(r)/r]0. 15. i. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). v > 1. Power Potentials general power potentials. A > 0. In order to have a clear starting point. and the normalization [ dr\y(r)\2 = 1. (15.1) i. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15.4>)ru{r) = Ylm(9.2) tt(r) = Ylm{9. mi + m2 and r is the relative coordinate.3) with the boundary conditions''' u(0) = 0. These particles therefore cannot be separated by any finite amount of energy. We leave consideration of the logarithmic potential to some remarks and Example 15. h2 2n V 2 * ( r ) + [V(r) .308 CHAPTER 15.2 The force The Power Potential F = .4) Instead of considering the linear potential. R(r) = ^ . and so u'(r) . For a central potential V(r) we write (15. (15.V 7 = const. we return to the Schrodinger equation in three dimensions. (15. where \x is the reduced mass of the two particles of masses mi.r acting between two particles is constant and directed towards the origin (of the relative coordinate).e.5) u'(0) u[r) r=0 = R(0).6) = (I + l)u(r)/r = R ~ rl. </>)R(r).
and the turning points are at a — — b := — xc. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality. (15. (15. (15.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.V(r)) = (2nl) +1 N.8b) This condition implies that we have only one pair of turning points..63a). We now want to find a quantization condition for the threedimensional problem. . (15. d5.2.1.63b): 2 and V(0) = 0 (15. . so that with Eq.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!). which has been reduced to an effective onedimensional problem in the polar coordinate r. This means that we transcribe Eq. 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. or f l H " ^ ) ! .8a) ^ JO = (2Ar + l ) £ .9) into the threedimensional case in the following form: / Jo C dry/2n(E . We therefore proceed as follows.63b) for the onedimensional Schrodinger equation. (14. We also assume that V'(x) > 0.8a) we can write the quantization condition (14. x > 0. N = 0..2. (15.. (15. odd integer ^=(nll)vh.7) k(x) = h/y/2m0(EV(x)).15. In order to to obtain the . Then k(x) = k(x).10) Here rc is the turning point and n — 1.4) in the threedimensional case."MA. is to be interpreted as the principal quantum number in three dimensions. 3 . .
15) I ' = ^ 1 (( 1) B{x. (15. (1516) (15. (15. E A fv.11) rc = (15.13) dt .y) defined by (xl)\(yl)\ y ^ (x + y .13) we obtain _ X_ „ _ so that A£_ (15. . (15. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2. Now.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .310 CHAPTER 15. c (nj\irh. dr = .17) The integral on the right is the integral representation of the beta function B(x. _ —vr" *dr. s.14) Prom Eqs.'(1*)1/2= J1F\lt)y1dt.12) and (15. Power Potentials eigenvalues E —• En for the radial potential (15.\rvc = 0 .12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.
t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. The case v = 4 is that of the pure anharmonic oscillator also discussed by M.1)!.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. One can convince oneself now. $T(z) = {z.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/.1 Approach to square well with v —> oo.55). T(z + 1) = z\.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. (12. for instance simply graphically as indicated in Fig. 15. ^ '"Compare with Eq. Weinstein [281].1. . 15.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. (15. and n — — ]7r/i. i r ( i ) = r ( l + i ) = ( i ) ! .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 ) .15.
25) "Comparison with Eq.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). 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). 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 = 0.2) implies A s fj.> r ( 2 ) A V 2 in \)*\W \it^T& (15. (13. (15. discussion at the beginning of this section). we obtain E„.oj2/2.1 + \)hx> = {An • l)£fiw. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15.2.21) (4nl).23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. For the following we set in Eq. the case of the linear potential. The potential V(x) = X\x\ (15.g — —2. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15. (6.41) for the . Physically it does not make sense for the probability amplitude to have a discontinuity there.oj = ^/2X/JX. (6. the harmonic oscillator. since these are the ones we are interested in (cf.mo = \. eigenvalues (JV + h)tkj — (2n . Power Potentials In the case of v = 2. we obtain En = (n .312 CHAPTER 15. > and with Eq. x > 0. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 . In the case v = 1.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.61) h2 = I. 15.
e.60) applies.29) . W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. Eq. Ai(s) 1 2v^(s)V4 exp x —> +oo.e. this wave function has the required exponentially decreasing behaviour at infinity.59) applies. In the domain —a < x < + a . dz2 According to Eqs.15. (13.e. (14. \(s?» (15.56) and (14. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . i.x)\x=0 = 0.e. (15. Ai(s) "v^ G ' cos s 3/2 _ t (15.27) i. 15.2 Behaviour of an odd wave function at the origin. E > V or s > 0. (15.x).2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. s = E. i.28) In particular at x — 0. i.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . i.26) For z = s := E — x —> —oo.e. and for s —> oo the trigonometric behaviour of Eq. (13.
1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 . Rosner. (15. The expression (15.^ = !(2nl).314 CHAPTER 15.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. to obtain the zeros of the Airy function numerically. Quigg and J.28) is really only valid for s large and s — oo.. (2/3)E3/2 or 3 / 1 2/3 E — En — . (15.51716 6.En from Ai(£„) = 0 2. the eigenvalues En are determined by the zeros of the Airy function.32025 4.00852 11.29) implies that cos (  ^ . It is interesting to note that in the dominant approximation both methods agree.1 demonstrates the quality of the WKB approximation by comparison with the exact values.00767 11. . 201.30) ! ^ / 2 .31) (an odd function has an odd number of zeros!).02265 10.08181 5. n = l.04017 11. [231].82878 1 2 3 4 5 6 7 8 9 10 2. i.03914 11. Table 15.e.e. It is possible.78445 7.82814 i.52056 6.33811 4. p.93602 12. so that n — 1/4 ~ n. with permission from Elsevier. L. Power Potentials Table 15. Table 2.78671 7.e.2.93528 12. but for large > values of E Eq. *This Table is reprinted from C. Co. i.. Quantum Mechanics with Applications to Quarkonium.94249 9.02137 10. (15. = im — (7r/4).08795 5.94413 9.= ) = «... copyright of NorthHolland Publ.3. of course.
cos2()^— / 1 fr° 2 > w . In order to be comparable with experimental data these investigations. van Royen and V. i. The resulting masses agree in most cases very well with those extracted from experimental observations.** r ( # ^ e e ) o c *(0) 2 . . in Chapters 24 and 26. i. e.3 The ThreeDimensional Wave Function 315 15. We indicate briefly here — without entering into further details — how these quantities can be calculated.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. See also J. The decay widths T. *Note that this makes N a W K B normalization constant. D. ee. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (.g.• •) by a mean value.g.e.e. In general these quarkantiquark pairs.15. of course. which are inversely proportional to the lifetimes. Jackson [140]. (14. the charmanticharm meson ^ . P. We saw previously (see for instance Eq.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. r <53) 1 3  dr cos 2 () (15. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions.e. can be shown to be proportional to the modulussquared of the particle wave function at the origin. Weisskopf [239]. have only a finite lifetime and decay into other particles such as electronpositron pairs. We shall encounter WKB normalization constants at numerous points in later chapters.34a) **The most frequently quoted reference for this result is R. F. e.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.
dE a—— / dn J0 so t h a t with Eq. A = (15.38) N : sin V ^ Jr dr' 7T (15. (15.316 In accordance with t h e relation CHAPTER 15.1 )n .39) in / ..2) and (15. i.e. (15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.37) (15.2.l N n . & (15.rcu w X{r)dr = Kh .40) VWY ..35) N2 . From Eq.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.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .34b) VMEvy Variation or differentiation of Eq.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .36) \i dEn irh2 dn According to Eqs. Power Potentials this averaging implies a numerical factor like 1/2.\ \ * n = l. (15. so t h a t approximately l N2 / Jo k(r)dr ..32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. (15. (15.
W. Kaushal and H. W. .g.23) for the energy levels of the linear potential. .3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. § S.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. (15. MiillerKirsten [147]. Bose and H. we can investigate the dependence of the decay rates on n (note that these would be decays into particles different from those bound by the linear potential — a consideration we cannot enter into more detail here). (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. MiillerKirsten [29]. . K. E. S.2. 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.15. (15.44) (15. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. Hite and H. The case of arbitrary positive power potentials as above has been considered by R. J. K.38) implies 317 (15. J.l ) n . 3 . .37) we obtain: 1 (2M^)1/2 47r ft. W.1 ^ ( 2 ^ „ ) 1 / 4 . G. MiillerKirsten [28].41) I*(0)2=_L^)^. 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. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. Hence Eq. and hence with Eq. we have ^'(0) = ( . for various potentials: (a) For a confinement potential of power v. Itt^pocn2*"1^"^.* Using the result (15. S. Bose. J. the expression (15. for instance.
318 (c) and for the Coulomb potential^ CHAPTER 15. W. (15... Lifshitz [157].2.. Bose and H.5.I l n f ^ p . Landau and E." 1 11 L .^ ( 3 ^ + l)55p^«(3^l) + . (15. *=^f. as well as Expanding U(z) about its extremum at ZQ = —1/2 and using the perturbation method of Dingle and Muller derive the expansions N)2 = ^^ 3 . and there Eq. . K.^ +•••. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . M. (14. E' = E + gln(r0). 7 With the substitutions r = exp(z — c).j . q = 2n + l.75). = (I + 1/2) 2 .L2 h4 = 4/3exp[(2a — /3)/f3].g ^£+(<*Pl»riy = 0. n = 0. Solution: Details can be found in the literature. MiillerKirsten [29].133).46) and the classical Kepler period of Eq.e.L 2 + U{z)]<f>. D. Eq = i q . S. U(z) =  dz zf3e2a/?e2*. Example 15.c = —a//3.46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14. (11. i. ip = exp[(z — c)/2].1.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). . {S = ^ > 0..— oo < z < oo. J. . = 1(1 + 1). rewrite the equation as ^ z + [ . Power Potentials *(0) 2 oc . in agreement with Eq.
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. In reality. This implies. It is clear that for ro — oo the > potential becomes of Coulomb type. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units.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 .Chapter 16 Screened Coulomb Potentials 16. and in fact. A very similar potential is the Yukawa potential V{r) = g2 — . in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. it is. so that like charges are repelled and unlike charges are attracted by each other. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. Historically this potential arose with the realization 319 pVr . that no matter how far the particle is scattered away from the source of the Coulomb potential. it will always notice the latter's presence. This effect sounds unphysical. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. The result is a screening of the Coulomb potential which thus attributes it a finite range ro.
who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.2 ) or (p = hk) 1 /" .1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. (16. in fact the parameter [i represents the mass of such a spinless meson. (16. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. 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. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. i. however. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. Heisenberg [133].ii k . and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. from the relation Pi + P 2 + M2c2 = 0. the mathematical expression with an exponential is not so easy to handle analytically.320 CHAPTER 16.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. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. Since. 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 . all coefficients Mj are real and independent of the energy E = k2.* The relativistic equation of motion of such a meson when free is the KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x.e.1) We encountered Green's functions earlier. . and that.3) where for the real potentials we consider here. one frequently resorts in calculations to the expansion of the potential in rising powers of r.
however. i. £+**PVV t/.{l. Cheng and T. Mandelstam [185]. .4) where E = k2 and h = c = 1 = 2m. Squires [258].e. i. bound state eigenenergies and the 5matrix. physical) values of angular momentum as functions of energy E.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.k.3) have been considered by various authors. S. Lovelace and D. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential. . i. C.k.3) in the radial Schrodinger equation for the partial wave ip(l. Basically this argument amounts to an argument similar to that used in the case of the Coulomb potential where the integer n arose from the requirement that the wave function be normalizable. (16. Potentials expandable as in Eq. l+n+l = ^P.r)=0. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H.r). for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. to be of the order of 1/k. Bethe and T.e. Omnes and M. Froissart [223]. A.e. S S.5) where K — ik and A n is an expansion in descending powers of K. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. (16.16. J. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes.1. if>oo. See also the other references below. Frautschi [97] and E. (16. T. Kinoshita [21]. f H.e. 2 . "These were first considered by C. Masson [180].Q.1 in the above. . Wu [47]. .mo being the reduced mass of the system. only for the cases of n = 0.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. 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. 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. They are usually written I = an(E). (16. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.§ 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.
Muller [201] and [202]. W.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.)/j. one can assume an expansion of the potential V(r) in ascending powers of r.322 CHAPTER 16.19).6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. J. 16. (16. Muller [204]. J. 'In the case of the Coulomb potential (cf.** The energy is later obtained by reversion of the resulting series. Screened Coulomb Potentials as a function of the energy.ndp. Bottino. < const. H. for all n. o rV(r) I dpp\V{peie)\ regular at r = 0. J. In the following we follow mainly the last two of these references.k. W. Regge [31] and E.z) Then x is a solution of the equation VaX= (16.z) = ez2/2zl+1X(l. tf H . Section 11. M. Under these conditions the S'matrix is meromorphic (i. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. Longoni and T. as may be seen from Eqs. A. Squires [258]. .9) the cut starts at E = k2 = 0.e. 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. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. W. (16. *See in particular A. 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. starting with the power r _ 1 of the Coulomb potential. Schilcher [203].15) and (16. Proceeding now as in the case of the Coulomb potential we change the variable of Eq.k.4) to z = —IKv and set TP(l.t Considering such superpositions of Yukawa potentials with /•oo / a(/j. J. < oo for all 0 < vr/2. in the present case at fc2 = 0 or k2 — Mi = 0. Miiller and K.
10): Sm(a. a) —b — 2a.b. anharmonic and cosine potentials have been derived in L. which means that the hypergeometric series breaks off after a finite number of terms.10) zm$(a)= ^ j=m Sm(a.j) = (a + j l.11a) The coefficients Sm(a. W.1). The function <&(a. and all Sm(a.11b) ^ The associated boundary conditions are: So(a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.* Recurrence relations for coefficients of perturbation expansions for Yukawa.j) for \j\ > m are zero. j)Smi(a. {0) X = Ha.awhere (a. J.2 Regge Trajectories where d2 Va = z^ and (cf.j l) + (a + j. Eq.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. . (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.a + + l). (a.j) may be computed from a recurrence relation which follows from the coefficients (16. where $(o.i 0) = 0. MiillerKirsten [249].9) (16.a — l) — a — l. (16.j) (16. as in the case of the Coulomb problem.16.b.5)) a = + 323 .a)${a) + {a.a+ l)$(a + 1) + (a. K.j +(a + j + l. a + 1) = a — b+ 1.a + j)Smi(a. (16.0) = 1. Sharma and H.7) is seen to be of order 1/K. so that to leading order we have VaX{0) = 0.j)$(a + j). We also observe that the ansatz (16.z) = Ha). (16. all other So(a.9) we obtain m l)$(a .5) is equivalent to a = —n. By a repeated application of the recurrence relation (16.a + j)Smi(a.
the latter can be written . [a. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. Evaluating the first few terms of the expansion (16. 1 . We first cite the final result .324 CHAPTER 16. This follows from the fact that D a $(a) = 0. (16.a\4.1] 2 . i.13) one obtains the quantity An(K) and hence with Eq. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .aj 2 (2K)A _ [a.5) the Regge trajectories I = ln(K).e.7). 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).o]i = M 0 .<&(a + n) on the right hand side of Eq.aj 2 \ [a.a]l^a) + ^2(2K)i+1 Y.[ a + l.An{K). when n = 0. (16.12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higherorder perturbation terms. [ a .j). for the coefficients M^> of the expansion Details can be found in the references cited above.12a) where [a. (16. and hence Va$(a + n) = ra$(a + n). Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. of course. .a + l ] 2 [ .n. Va+n = Va . #(a + ra) fjL$(a + n) V.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. 0 < \j\ < i. (16. oo _ .l.' — .a + j]i+l = MiSi(a. [a. i R »] = 2K[a.[a>a + J]i+i*(a + 3). (16. + (16. a . Any term /j.
14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) .1 by evaluating the first two terms.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. . we obtain the expansion for the Regge poles. (16. Ahmadzadeh. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero.2.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . (16. including an exploration of the domain of small energies E. Thus the expected appearance is that shown in Fig.^ p [ 3 M 4 ( n .^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex iplane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8).1. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n .5).e. Burke and C.1) + 3M3M02 + M22n(n + 1) (16. Such plots of numerically computed Regge trajectories for specific values of the overall coupling constant and energies varying from minus infinity to plus infinity have been given by various authors.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16. Inserting this expansion into Eq. With numerical methods one can achieve more. Tate [6]. .( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n .§ This is an important observation which indicates the equivalence of the methods.16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential. G. of course. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n . Obviously the expansion — which. P. i. However. 16.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .
a] 3 = M 2 S 2 (a.—" + In = — .1).326 CHAPTER 16. p. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig.0). through which the particle orbits during the course of the resonance. ±1) = 0. 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. Frautschi [97]. Hence Si (o. For a resonance with a long lifetime. [a.a)So(a. 16. Solution: We evaluate the first three terms of expansion (16. and the angle A#. (16. In fact. 0) and Prom Eq. (16.O) = Other terms vanish since So (a. (a.12b) we obtain [a. The conjugate variable to energy is time. However. one can expect a resonance with the lifetime determined by this imaginary part.1 Typical Regge trajectory for a strongly attractive Yukawa potential. Similarly the conjugate variable to angular momentum is angle. ai is small and A6 is large.11b) we obtain Si(a. From Eq.0) = (a." Re I 1 0 1 2 Fig.13). C.2 a = In See S. so the imaginary part ai of I — an represents its width in angular momentum. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown.a) = b . 114. . and the lifetime At of the resonance satisfies the relation TAt ~ h. a] 2 = M i Si (a. satisfies the relation ajAO ~ h.^ . . 16. just as a decay width T represents the width of the resonance in energy. where the imaginary part of I is very small. Example 16. 0).
a)Si (a. 14. 1) Hence 52(a. using the Cauchy formula (6. we have to evaluate (cf.11b).(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .1) + (a .2 a ) .1 ) + (a. Solution: Ignoring higher order terms in r.0) + 0 + 0 = ( 6 . (a.al)S0(a. a)Si (a. A" An An A" = = = (a. One sets z = 1/r so that.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".a + l)S0(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 .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .2o)Si(a. a)Si (a. Thus (with terms which are 0): Si(a.b)(a . .a)5o(a. .6 + 1 ) .2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.0) (a .1) Si(a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0. Sec.1 ) + (6 .0) Si(a. (a. i ) again from Eq.0) + 0 + 0 = ( a .A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 . (16.00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I. 1). 0) + a S i ( a .6)Si(a. (n+l)U+^pJ . 1) (a .0)+ 0 + 0 = (a1).2 Regge Trajectories 327 Analogously we evaluate S2(a.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .13) we obtain 0 (Mo . 0) + (a + 1.0) = (a .6 + l)a + A.1.n ( n + l Inserting these expressions into expansion (16.16. We obtain the quantities S i ( a . .
. Solution: For details of the solution we refer to papers of Boukema. Singh [252].* 16.328 Thus here dr K' CHAPTER 16. . I. n = 0. Boukema [34] and [35]. tSee Chapter 11 and V. (16.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. ±2HK in agreement with our expressions above. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. by expanding the right hand side in rising powers of v and evaluating the individual integrals.e.t First.l.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. *J.15). Screened Coulomb Potentials Mo .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. we can use Eq. The turning points in this case are at r = 0 and r = oo.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. Here we do not perform this procedure. however. i.2. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. (16. 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) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .5) together with the expansion (16.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). Example 16.
J.* 16.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16.16.. W.l (two additional terms are given in the literature). we can now write down the S'matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e r(/ + i^#>)e ' (16 18) ' We observe that the poles of the S'matrix are given by l + l + ^p = n. W.l)n(n + l)(n + 2) + 2M2M02 .19) +2M 3 M 0 (3n 2 + 3 n . Miiller [211]. the behaviour of the scattering phase in the domain of high energies. One can use the explicit expression (16. i. H. . n = 0. This has been done§ and one obtains . .g.19) we obtain the energy. Paralleling the case of the Coulomb potential in Chapter 11. Miiller [204].2. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\.14). the following *See e. for instance.e.0[(KZ + Mx)'6} (16. J. such as.17) and is therefore real for real values of the potential coefficients Mj. Now reversing the expansion (16.l.17). We observe also that the S'matrix is unitary as a consequence of the result (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 . which are precisely the expressions yielding the Regge trajectories of above.18) of the Smatrix now to explore further aspects.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\. § See H.
1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . Warburton [279]. p.2)(n . MiiilerKirsten and N. Large coupling expansions derived in H. 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) ^ — . . C. the transformation was introduced in the form given below by G. E. 282. Frautschi [97]. Screened Coulomb Potentials expansion.e.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n .* We recall from Chapter 11 the definition of the scattering amplitude F(9.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + 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 . VahediFaridi [212] contain as special case the expansions of the first pair of authors. Watson [280] in 1918 and later resurrected by A. p.5 The Sommerfeld—Watson Transform The basic theoretical tool for the exploration of Regge poles is a representation of the scattering amplitude given by the socalled SommerfeldWatson transform. A. J. Sommerfeld [256]. 107.^ 16.330 CHAPTER 16. Zauderer [286].l)n 2 (n + l) 2 (n + 2) 1 + • • • (16.l)n(n + l)(n + 2) .11) + 2M22M03(9n2 + 9n .< 3 M 4 M 0 ( n . i. W. Iafrate and L. 'According to S. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. B.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . J. Mendelsohn [135] and E.1) + 2M2M.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. N. k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . in the following expression with an "See in particular G.
23) f(l. (11. (11. (cos 9).2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i.22) = \F(9.k) — e 2i5[(k) ~" p—iirl (16.k) = ^T(2l + !)/(*.''' The idea is now to consider each term of the expansion (16.iJ 11 —[S(l.. .26.24) Si(k) being the phase shift.fe)P .1.kmcose).25) Setting Note the minus sign in the argument of the Legendre function.2 The integration contour C.184) to (11. \x\ <C l .. 16. For this purpose we consider the following integral taken along the contour C shown in Fig. 16.187).k) = 2zfc L " v "'' v . dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig..k)~l] = ~ k" PiSiW sin 5i(k) and S(l.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. Eq. I = n + x..TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0. z=o in which (cf.k)\2. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).k) 0ikr (16.187)) (16. The scattering amplitude possesses the partial wave expansion F(0.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. c = 1. Actually 5j lrr/2 as in Eqs. n = 0.2. ~:' sin nl (16.
28) .27) a(p) © b(q) Fig. t = (qq')2. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. a + d — • c + b. In order to see the relevance of Regge trajectories in a reaction of particles. For the momenta indicated we set (with metric +. 16. n=0 which is the usual partial wave expansion.3. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem. as indicated in Fig.26) P((+COS0) = 2 ( 2 n + l)/(n. in which case c = a and d = b. u={pq')\ (16. a + c —> b + d. The reaction therefore describes the following processes which are also described as (reaction) "channels" and in which an over line symbol stands for the appropriate antiparticle (like IT meson with positive charge and that with negative charge): s: t : u: c(p') a + b —> c + d. 16. d(q') (16. 16.fc)P„(cos0).—. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.3 Reaction channels and their respective Mandelstam variables. Thus the direct reaction may. 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'). we obtain m® = ± I dl^±^f(l.n) oo {l)1 Pt{cosO) v ' (16.3.332 CHAPTER 16.k) Zi Jc TT{L .—. be elastic scattering of a off b.—) s = (p + q)2. for instance.
u) (if any of the external particles has nonzero spin. which is an analytic function of the variables s. Mandelstam [184] and R.u are not independent. t = 2q£(i . A(s.t). one can write the amplitude as depending only on two.(<fe)^(cos 6t). the momentum k there would correspond to qs) s = ~4(ql + ml). if we had been considering the s reaction originally. the dynamics is described by the exchange of quantum numbers.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.t.t. one then has A(s.4.9s)\2. Goldberger.t) = y^Fs(s.g. S. L. .The variables s.31) i l Taking all particles to have the same mass mo. if s describes the square of the total energy in the s channel. (16.32) t = 4(q? + ml). there will be several such amplitudes which together describe all three processes). Khuri and S.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). u = .0s).c o s 0 s ) . 16. t. except for poles and cuts which characterize the reactions in the three channels. (16. Treiman [25]. M.. Ft = £ ( 2 J + l)F.§ Since the three variables s.16.t.et) (16. u. the variables t and u would describe momentum transfers in the crossed channels. § See G.t.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. For a Lorentzinvariant normalization of the initial and final states. B. Blankenbecler. e.g. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions).g.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. s = 2qf(lcos0t)(16.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. T See e. etc. F. (a) Mandelstam's hypothesis: All three processes (described by the s. Symbolically we then have the situation shown in Fig.2 ^ ( 1 + cos6 s ). N.t) = VtFt(t. the only kinematics we require in the following is given by the relations (e. A(s. In a Regge theory on the other hand. One now makes two hypotheses. N. Chew [48].u are known as Mandelstam variables. Thus. the quantities carrying these quantum numbers are the Regge trajectories.
in the domain of schannel physical values of the kinematical of the ichannel).4 Yukawa versus Regge theory. Screened Coulomb Potentials Regge oc(t) Fig.334 Yukawa CHAPTER 16.5 The Lehmann ellipse. 16. But > > wave expansion X)(2Z + l)F. the Legendre polynomial Pj(cos 9) does not possess a cut. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. Lehmann [162]. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. But for integral values of I.5 for Yukawa potentials V{r) u: lm cose dr a(/j.)e~ •fj. 16. . 16.e.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. So what can one do? 1 H.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. the partial wave expansion is not valid in the physical region of the (i. For s — oo : cos 9t = 1 + s/2q^ — oo. However.
26) but for a different choice of the contour. Moreover. i.1 / 2 .33) . the contour representation (16. Thus consider now the same integral but taken along the closed contour C shown in Fig. sin nl [16. 16. „ i. k) have to the right potentials one can show t h a t the integral along the curved porA' of the circle at infinity tends to zero. contour C / / D ^ ^ . B Iml 1 \ i All \ r\ 0 / closed .e.e. 16. and there is only a finite number N + 1 in the domain of 9W = .k) »E n=0 [2an(fc) + l]/3n(fc) p sin 7ran(k) an(k){ cos 6) 2i y _ i _ i o o = F(s.5 The SommerfeldWatson Transform 335 T h e answer is.6 The integration contour C". all poles of Ql > 0. N F(6.^ B 1 A Fig.6. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. to use the SommerfeldWatson transform. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.16.t).
J.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. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. Regge trajectories have. for plots and discussion see R. MiillerKirsten [29]. 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. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. Predazzi and T.32)) and finite. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. W. this behaviour has been confirmed in high energy hadronic reactions. Omnes and M. Eq. It follows that under these conditions. Bose and H.t)&B(t)sa°V.336 CHAPTER 16. i. p. K. at high schannel energies. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. . The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential.35) This result is valid for s — oo and t negative (cf. the amplitude can be represented as a sum over ^channel Regge poles.e. (16. as we discussed briefly in the above. Apart from more refined details. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. Foissart [223]. In particular. we have (16. for high energies in • > the schannel). 42. Regge [229]. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel.e. i. if ao(t) is the Regge pole with largest real part. Limic [177] and E.e. of course. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. 16.
The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. like a meson or a baryon. W. MiillerKirsten [209]. called "Regge cuts".g.1.16.6 Concluding Remarks 337 important parameter in string theory.** Whereas the former are related to absorptive properties of a reaction. J. the latter are related to the distinction between "' elementary and "composite particles". The way to do this is similar to calculations in Chapters 18 and 20. Related aspects are discussed in Sec. For an overview see e. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. a composite particle being one with structure. and is composed of quarks. . and also as socalled "fixed poles" . H. The above treatment of screened Coulomb potentials is incomplete. Singularities appear also in the form of cuts in the plane of complex angular momentum. 20.
This page is intentionally left blank .
Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers.Chapter 17 Periodic Potentials 17. Arscott [11]. called elliptic modulus. 25. example 14. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6.) which are functions that interpolate between the n. See F. In essence these periodic Jacobian elliptic functions. the most immediate candidate is a trigonometric form like that of the the cosine function. p. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type.0 < k2 < 1. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term).* These elliptic equations involve Jacobian elliptic functions like sn(a.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. The Schrodinger equation with this potential.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. 339 . since many analogous relations hold. K being the complete elliptic integral of the first kind. 2K or AK. depending on the parameter k. The reason for these difficulties is. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. 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. however. motivated mainly by the regularities of the crystal structure of matter. 19 and p. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. are not much harder to handle than trigonometric functions. M.g. and are themselves periodic with period e.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation.
(17.e. The first of these conditions is already tMathieu equation: R. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . . and to relate these to the weak coupling bands or regions of stability. the Lame equation possesses 2 n + l polynomial solutions.1. 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. or Sw&ve Schrodinger equation with elliptic potential. those of integral order. which one looks up in books like that of MilneThomson [196] or Tables when required. The Mathieu equation. 17. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. Lame equation: H. each with its own characteristic eigenvalue. 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. See also the discussion in Sec. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions.340 CHAPTER 17. (17.7. 8. cosine and tangent. need not be afraid of them. In fact.. J. This is a significant difference compared with the Mathieu equation. ''The reader who encounters these Jacobian elliptic functions here for the first time. which is conveniently written ^ + [\2K2sn2z]y = 0.1. c n z and d n z are handled with analogous formulas. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. For large coupling. Dingle and H.1) where h2 is a parameter and — IT < z < IT.2h2 cos 2z]y = 0. W.. i. Miiller. Both of these domains are important for a host of other considerations. B. [205]. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. 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.2 . which does not possess any polynomial solutions. the three Jacobian elliptic functions s n z . The most important formulas for our purposes here are collected in Appendix A. J.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K.4. Muller [73]. W. A main objective of this chapter is the calculation of these level splittings following the original calculations of Dingle and Muller^ with the perturbation method described in Sec. We might mention already here the much less familiar but more general Lame equation. Like the trigonometric functions sine. Periodic Potentials vide periodic potentials like trigonometric functions.
2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. these concepts are rarely explained as such there. a bounded function. since the solution there is of the form cos vAz. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. i.e. i.1 The Floquet exponent We consider the given Mathieu equation with argument z and compare this with the same equation but with z replaced by z + ir. In classical mechanics stability is treated for instance in connection with planetary motion. we see that in a plot of h2 versus A. we compare the For interesting related discussions see M. Although the mathematics literature on the subject uses the physical concepts of stability and instability. y" + \y = 0(h2). 17. Adding the negative term — 2h2 cos 2z to A. i. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. Looking at the Mathieu equation for h? —> 0. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small.e. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.17. h2 = 0 belongs to the domain of stability. .2.e. i. the case of h2 small. 17.1. as in Fig. 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 for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability.e. Vachaspati [242].^ 17.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. the semiaxis A > 0. 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. Salem and T. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.
(17.342 CHAPTER 17.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z).e. d(z + TT) cos 2z We see that there are solutions which are proportional. Setting z = 0. i. y+{z) . y~(z) = 6sin vXz. 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). — • . 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. TT. (17.B and a. ° ^ [1 + 0(h2)}. y(2vr) 2 a = —7—^ = 1.2/i2 cos 2(2 + vr)]y(z + vr) = 0.2h2 cos 2z]y(z) = 0.e. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. sin V Az or linear combinations. Periodic Potentials equations ^ p + [A . a = const. d2y(z + 7r + [A . we obtain y(vr) y(2?r) a = — — a = —^^r. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. i.
2/'(7r) = a ^ W + ^ . and by'+(?r) = —ay/Xy{n).y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. (17. y'(z + n) = a[ay'+(z) + py'_(z)}.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\. i.3) we obtain y(z + ir) = a[ay+(z) + @y(z)].y+(7r)j/_(7r)} = 0.e. we choose these with the following set of boundary conditions: y+(0) = a. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. Using Eq.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. yV(0) = 0.{ay'_{ir) + bV\y+(ir)}a Now from Eq.a2b2\ a± ~ 2a6\/A . y{z) = aty+(z) + 0y(z). y'_(Q) = bV\ with Wronskian W[y+.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3.y+(7r)y_(7r) = 0 or abVXa2 . with constants a and /3. 2/_(0) = 0. (17. This means. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa].4).aa][y'_{n) . We can write therefore. The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. the determinant of the matrix must vanish. This expression is equal to the coefficient of a2 in the quadratic equation for a. [y+(v) . \y+>y\ = y+{ir)y'{ir) . 7 y\z) = ay'+(z) + (3y'_{z). ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .17. Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions.oWX] .
(17. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . 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. i.455A2 + 1291A . 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. Prom this condition we determine later (cf. this implies y+(Tr)/a = ± 1 . (5i/ 2 + 7)/i 8 vz + — . (17. ** An important consequence of Eq.344 Setting / = 2by/\y+(Tr). 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 .7. where the constants are usually taken as unity from the beginning.8) "Observe that for integral values of v (in lowest order of h2). cr+ = 1/<T_. 8. i. 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.6) We can derive various terms of this expansion perturbatively with the method of Sec. (17. T h e result for A is the expansion A = h4 2 . _ / ± v 7 ^ V .e. (17.5) a The parameter i/ is known as Floquet exponent. <7__cr_ = 1. Eq. <T__ + <T_ = 2y + (vr) . **This point is not immediately clear from mathematics literature.1.e.. Setting cr+ = e . .1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. This calculation is demonstrated in Example 17.= — . Periodic Potentials g= 2aby/\.0 This series may be reversed t o yield the Floquet exponent. .22)) the boundary conditions of periodic solutions.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .
we must have a2 — 1 and hence e 2iv/A7r = 1} i.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. Meixner and F. 8.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. W.35A + 8 . Example 17. It follows that in these cases i? = n2[l + 0(h2)].1) 15A2 . These cases therefore have to be dealt with separately.2 how the expansions for small values of h2 may be obtained perturbatively.3). = J—^ + o(hr).17. n an integer. Schiifke [193]. VA = ±n.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. as will be done later.11) or smvz or e±ivz.e. (17.10) A = n2.„„ x n. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2. 124. p.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. For these integral cases of v we demonstrate in Example 17. (17. (17. (17.12) S e e J. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17.jl2.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . Solution: We write the eigenvalue equation A = v2. (17.5) and (17.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 .7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. . which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). (17.35A + 15A2)/l8 /. cos TT\/A 32A(A1) 2 . — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0.
2) _ __ _ (y. of course. (i/. = . We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. The recurrence relation can also be obtained by substituting the right hand side of Eq. With our perturbation formalism — as demonstrated in cases considered in Sees. ("• ^ + 2) (o) 2(2^ . . see Example 17.„ + £ h2i £ P2i(2j)yv+j (17.h2). when a = 0 or 2v + a = 0. i. (17.16) directly into the original equation in the form of Eq.e.!/) = 2A We now observe that DvVv = 0. (17.e.3. (17.346 CHAPTER 17.11). Schafke [193]. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) . v ..v + 2)yu+2]. and (v.3 and 19.2)j/„_ 2 + {v. i.2) which applies when v is nonintegral.v— 2) 2 ^ + yu+2 Vv 2 y (D = h _2(2u2) 2(2i/ + 2 ) ' Then up to 0(h2) the sum j / 0 ' +y^ is the solution provided the remaining term in Ri. (17. Proceeding in this manner we obtain the solutions ce. se and me as* oo i y = „«>) + yW + i. i. 17.(2) + . where except. (17.15) Hence a term fMy„+a on the right hand side of Eq. The right hand side of Eq. as shown.14) = = = cos(y+ 2)z + cos{y— 2)z. for instance. . Meixner and F.h2).11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. .e.v±2) = l. sev(z. v) = 0. up to 0(h2) we have A = 0.14) therefore leads to the following nextto leading order contribution (v. Next we treat terms yv+a in j / 1 ' in a similar way.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .e. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. h2(u.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. by J.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. siniy + 2)z + sin(i/ — 2)z.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. (17. h2). since j / 0 ' = yv leaves unaccounted R\. me±v(z. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17. e ± i ^ + 2 ' 2 + e±i("2)z. (17.2. '. v . v)yu + {v. W.16) *These smallh expansions are convergent. or A = 0(/i 4 ).
Meixner and F.g. 0 = h2(u.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.e.u) + h4 (". This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2.1 with A = v2 — 2h2A but with v = 1. . Following the first few arguments of Example 17. * Note... one has to deal with each integral case separately as in the following example... however. 2 ( 2 i / .R{ . „ . W. (17.17) Inserting 1 for the stepcoefficients of Eq. . + y~i) (17. fo^ + 2) . The higher order terms naturally require a little more algebra. thus determining the quantity A. (17."2). 121. J. p.. Example 17. i.17. For v = 1 we then have yi = cosz = y~\. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . the solutions are not yet normalized. Schafke [193].2 ) v(y2.v)\ . 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>.7). is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. .2—r——— (i/ + 2. v) ' ( 2 v + 2) (17.2: The eigenvalue A for u=l and h2 small Use the perturbation method of Chapter 8 to obtain the perturbation expansion of A for v = 1 and h2 small. . (17.14). Solution: We start as in Example 17.15) (but with v — 1 and a = 2) we obtain the next to leading contribution to the solution as (i) = h2 =_fe2 V {2(2v + 2)}„=1y3 8 y3' This contribution leaves uncompensated the terms amounting to R\ = ~ — 2h (A + cos2z)j/ 3 o = —[Ay3\ 4 \ 2 Hence the next contribution to the solution is 4 V 8 2(4)(2 + 4 ) / 4 V 8 48 Proceeding in this manner we obtain as the complete solution the sum y = y(0)+yW+y(2)+. See e.18) Again applying Eq.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].
n integers. cos z h cos Zz 8 h2 h4 ( cos 3z cos 5z — cos iz A 1 8 4 l 8 48 h4 (cos 3z cos 5z 1 4 I 16 48 (17. .* The associated solution is ym+ym.24c) below) . and v is to be found in ascending powers of h2. W. i. We introduce a normalization constant c and rename the normalized solution y then yc. The reverse situation is later of importance. (17.e. 120. Schafke [193]. see below!) one obtains the additional terms given by Meixner and Schafke [193]. 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]. Meixner and F. In the above we considered the case of v an integer and calculated the eigenvalue A. the case of A an integer. h2). The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. 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 . *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . h4 (17.348 From this we obtain 2h2A = h2 CHAPTER 17. Periodic Potentials h4 r ••• . 123) (there the solution is called cei(z. We consider this case again in an example. With normalization (not to 1. 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. p. m.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). J. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. Eq.
17. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .1)2/A TT2 +h8 (15A2 . o 1• • • .9b) about A = 4. J .l ) 2 (A .3 1 A + 8) ( A . S..2 Cosine Potential: Weak Coupling Solutions Example 17.1) . 128A2(A1)3 + 0(h12) in the limit e — 0 > Hence — observe the cancellation of factors (A — 4) in the term of 0(hs) h TT (ll C O S TTV = 1 H S 2 x 1 6 .• 1 + 5n h 2 s 2932 Setting in cosm. ^S.3 1 x 4 + 8) 274233 (. .4 ) + . we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i .3 5 A + 8)TT2(A4) _ 3 / / 64(A .3: The Floquet exponent i> for A=4 and h2 small Show that the expansion of the Floquet exponent v around \f\ complex expansion: v = 2 349 = 2 is given by the following iVE/h 3 \2J 7% h (108^/5 \2J 8 1185H (h 31104^X2 Solution: We proceed as follows which demonstrates explicitly how the singular factors cancel out systematically. Substituting these expressions into Eq.4)Av A2v A ~ 32A(A .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A .. (11A2 .1) (A . (17. Liang and Yunbo Zhang [189].9b) and considering the approach A — 4 gives > (A4)2 + .Q . Gubser and A. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.9b) v — 2 + <5. W. H. . Manvelyan. J.('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 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . MiillerKirsten. on the left hand side of Eq. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. fe47T2(A4) 4%/A(A . Hashimoto [124]. (17. (17.
1 Boundaries of domains of stability. pp. Arscott [11].21a) and (17. Taking in particular Eqs. in F. 28 .350 CHAPTER 17. as also conditions (17.6.29.21b). bV\ y+(0) 2/40) 07. (17. These are defined by specific boundary conditions.2. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. We now derive these boundary conditions. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. 17. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. "These conditions.2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir. = and s+w 2tW±I. solutions for integral values of v.. for instance. Fig. may also be found. We verify these for the large/i 2 solutions in Example 17.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. (17.w^. i.e. 17. M. .10) applies.W4W=UtWSta±I(.4) into account.
21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .17. /7r\j/!_(7r) y L l 7TJ . 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f. /TTW^) 2j^T±y+V2 From these equations we obtain y+{ir) in terms of functions at TT/2 by eliminating y'+(Tt) from the first and the third equations. Then y+ y 351 U . ^\y±M /vr\yL(7r) . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . which after some rearrangement can be written y+(vr) _ . o y+(fy(f)ob\/A _ = 1 + 2I&A/X .4(f) (17. .y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. we obtain y+(7r) _ i  2y_(f).2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2.
24b) 8 h2 z —— cos 3z + • • • . A *• b2n+2(h2). This is shown schematically in Fig. the case of ai with solution cei (of period 27r).1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec.24d) v 12 ' **To avoid confusion we emphasize: In Sec.2. A 2 y • ce2n=q0i. A > b2n+i(h2).2): hA 7h8 29h12 + Y l28^04. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. These may be subdivided into classes depending on whether the integral value n — 0. of v is even or odd. IT. There the second term on the right is nonzero.+ .2 we require this relation for nonintegral values of v..T + .. (17. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.• a2n(h2). (17. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. (17.352 CHAPTER 17.24c) 8 h2 sin 4z \ . (17. Mr7nA (17. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • . in this order. 27r. y > se 2n +2s 9o +i The functions so defined have respectively period 7r.— — — • • • . sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .21b) must be ± 1 . and whether the function is even or odd.1). 17. y > se2n+i=qo. Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2.21a) and (17. 17.1.3.. . was treated in detail in Example 17. ce± = cos 8 64 h4 5h8 b2 = 4 1 .' fh2 V2ceo = l . We see that there are four possibilities and hence four different types of functions.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . 19.5) that the left hand sides of Eqs. (17..22) * a2n+i(h ). . 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 .23) We can now see how the domains of stability and instability arise. y > ce2n+i=qo. (17... (17. These are defined by the following boundary conditions. 2K.
1 that the boundaries of regions of stability merge to lines. Dingle and H.17„. 17. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain.3. J. 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. . . 13824 .> (17. Thus in Fig.17. The cosine potential cos 2z depicted in Fig. the walls are not infinitely high and hence tunneling occurs from one well into another.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. of course. 17.3 17. W. .hz — 12 . B. We can see from the inverse of Fig.2 The cosine potential.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. 17. 353 . 17.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 .z3 ce2 = cos 2z . 17.2cos4.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R. In the case of the cosine potential. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator.24e) 17. Miiller [73].
1..25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. but only approximately so. Taking the remaining terms of the cosine expansion into account. We therefore wish to expand it about these points. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1.3) and is. Periodic Potentials minima at z — ±ir/2.26) 17. of course. Thus in this case the quantity on the left is not an exact odd integer. there is tunneling from one barrier to the next. where A / 8 is a remainder. (17. simply the onedimensional Schrodinger equation for the harmonic oscillator.. n 0..2. 4 (17. we therefore set 9 A A = 2h2 +2hq + —.27) .26) into the original Mathieu equation and obtain y" + 2h q + rrxr .2/i cos z V = 0.2 and 8.. Sees.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. (17.3. and we set this equal to q. as in the case of the cosine potential.2 The solutions We insert Eq. Ion (17. In the case of finite heights of the potential barriers. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .354 CHAPTER 17. 8.
(17. and substituting this into Eq.28) A" ±4hcos zA' + 2h[ q =f sin z \ A.17.30) D A): COS « = 4z + 1 iq sin 2. can be . z).3 Schematic picture of Mathieu equation eigenvalues. (17. Choosing y(z) = A(z)e2hsinz.31) We make the important observation here t h a t if one solution of the type (17.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z). 17. the linearly independent one.+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.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) .29) is known. 16/i/ (17. (17.27) we )A = 0. another solution. 2 (17.
(17.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g.30) amounting to Ri ] q = . ? .4 ] . (9l)(93). 1 + 4 ) ^ + 4 + (^. . 2 /i D^Ag = 0.9) = 0.^ 6T [(«>« + 4 ) A ? + 4 + (Q.35) But since and so a term iiAq+n on the right hand side of Eq.e._ 4 .4 ) ^ .(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.94) = (9 + l)(9 + 3).356 CHAPTER 17. 2[(92 + l) + A]. (17. (9.g + 4) . (17. Aq+4 2Jh . of course.33) (17. (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. when i = 0 — the terms not involving Aq in (17.Q)Ag + ( ? .34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. With some algebra one finds that 16^ + where (9.32) cos^'+i) [\K .9) (9. i. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. ? .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. ^ ) A + ( ? . D{q%Aq+4i = 0.36) Then ^4.9 + 4) = (9. (17.35) not yet taken care of is set equal to zero. 2 A A = («.\z To lowest order the solution A is A^> = Aq.4)A.94) \ — Aq4 (17.
(q. (17.4) (q . Then for A = A(0)+A(D+A(2) + . +4. the sum of the terms in Aq in Eqs..37) and so on — left uncompensated — must vanish.4.35) and (17. A.. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q. to satisfy Eq. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.4. h) = A(z.g + 4)(g + 4. (17. h).4) 1 1 (q. (17. (17. Thus in particular A(z. q. .q + 4)(q + 4. q. h) = A(z.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation.<J+4 Aq+& + 1 2 1 . h). ) ^ 4.40) .4 .)+ & ! = « ( . (q.. Q .g + .q + 4) (g.37) Here all the terms. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q. Thus 0 = (g.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17.17. A(z. q .4) (q . q (17. q .38) Aq8 + 1 2 Clearly we can continue in this way. .8)A. q.q 4)Aq+4 q + 4 1 q + + + + + <^(. except those involving Aq.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ . q .q 8)Aq+8 (o)' ^±V(q 1 4. ._ 8 (17.39) which is the equation from which the quantity A and hence the eigenvalue A is determined.g4) 4.
46) 2s(«. . We return to Eqs.44) where D(B) 9 = dw2 $_ !)• (17.45) {B) We recognize Dq as the differential operator of the Hermite equation." 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. 2 w2 + H* (17. The differential equation there is Eq. and discussed there.e. Before we study the eigenvalue equation in more detail. (6.41) so that the early successive contributions decrease in magnitude.i ) We encountered Hermite polynomials earlier in Chapter 6. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). (17. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. We observe that for the large values of h we are considering here.) i ( Q . i.d2B w dw2 + w(l w dB_ dw d dw 1. Note different definitions in use. where Hen{w) is a Hermite polynomial when n is an integer.43) 9 2eh .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 .358 CHAPTER 17.67). 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. 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. it is found that D(B) z+ — ]B = 0.42) (17. (17.
q + 4 ) £ g + 4 + (q.3 ) \He* ! (9+1) H. i. i..e.51) . which is an amazing and unique property of the Mathieu solutions — cf.7 T H 1 2 Z 4 <1.34).d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. The solution with contributions B^°\B^>..3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17.e.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized.47) where the coefficients are the same as before.qi)Bq_4. the expansion for the quantity A being identical with that of the previous case.e. Finally we obtain a third pair of solutions.48) This shows that this solution is valid in particular around z = ir/2.50) C = 0. Again a change in the sign of z throughout yields the associated Mathieu function y = i?exp(—2hsinz) with B[w(z)] = B[w(—z)] and domain of validity in particular around z = —ir/2.17. (17. . y — Cexp(2hsinz) and Cexp(—2/isinz). 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). with the equation C" + 4h cos z C + 2h ( q . i. 16hJ (17.e. (17. also the same coefficients in the case of functions C below.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh. (17. i.. as in Eq. q)Bq + (q. A{wz (17. 1 COS I .
17. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq.h). We note that C(z. 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/2.52) and the coefficients are again the same as before. q)Cq + (q. q + 4)C 9 + 4 + (q.7 T 1 2 Z 4 <1.A Hen . we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).q.w2)He'n .44). These domains of validity are indicated in Fig.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .4.n2(n l)Hen2 . (9+3) (2n 2 + 2n + 1) + .53) This solution is valid in particular around z = —7r/2.4)C g _ 4 .54) where the proportionality factor is complex. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. He'n(w) = wHen(w) Hen+i(w). 2' the remainder 72. (17. show that the perturbation remainder of Eq. (17. We have thus determined three pairs of solutions of the Mathieu equation. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He.360 CHAPTER 17.q. 2 J where n=(ql).4: Hermite function linearization of perturbation Using the recurrence relations (6. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. each of them associated with one and the same expansion for the eigenvalue.q. Example 17.(w2 + A)Hen.59d) of Hermite functions Hen(w). (17. ie 1 COS I . A closer look at that expansion is our next objective. .h) ocB{z.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. (17. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. (17.
j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. . (b) The allowed moves are +4.q4) (17.55) M2 = M.' 361 in 2*f*1>[i(gl)]! = (q.3).q + 4)Bq+4(w) + (q.. .17. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q. q .l)[l(q _!)].4) = (q . 17.q + A){q + A. We can write the solutions obtained previously.57) . q)Bq(w) + (q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .1i . The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.0 and —4.3. ) . r j=r. l)Ag4 + P2(q.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.l)(q .q + 4) (q + 4.56) (94. (q. (?) 1 1 + (9.. ^ 1 Hv" '^ ^' ^ ' . (q. (q. in the following compact form with coefficients Pr(q. the solution y = exp(2/i sin 2)^(2). q) = 2[(q2 + 1) + A].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). no intermediate move to or from q being allowed. 2 + ^ M 3 + (17. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4.q 4.g) 1 1 and so on. with (.i ) ( w ) 2l(.g. 2)Aq_8} + • (17. q + 4) = (g + 1)(? + 3).)Bqi(w).g4)(q4. e.
i.g + 4) ^3(9. j) +(g + 4j + 4. 0) = 1 and all other P0{q.g + 4) + 1 1 1 and so on. Miiller [73].362 Clearly CHAPTER 17. l ) .g + 4)(g + 4. j) = (g + 4j . g + 4j)P r _i (g. g + 4j)P r _!(g.g + 4) 2 1 (g.j). 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 ^ 0) = 0.j).P 2 ( g . B.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.e.l ) . P2(q 2) Pl(g. the factor j removing a duplication of factors in the denominator at the junction.j)] = 2j]jPr2(g. and in general* M2r = J]j[P2(g.g + 8)(g + 8. 2) 1 2 ' ' ^ ' ^ 1 (g. p = (<?><? + 4 ) ( g + 4 >g + 8) .60) M2r+l = = *R. x)_ (g. W. for instance.g + 4)(g + 4. J. (17._1) = Mzil. From the coefficients (17.i (g.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.g + 4 ) ( g + 4.g + 8)(g + 8.j). M2 = P 2 ( g . Dingle and H. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. Thus we have.1) + (g + 4j.j)Pr+1(g.j)^2(9.34) we deduce that Pr(q.4.3) ^3(9.g + 12) 1 2 3 (g. as jP r 2 (g. . (17. g + 4 j ) P r .g + 4)(g + 4.j).g + 4)(g + 4. j + 1) (17. j .j) = (l)rPr(q. Periodic Potentials PlM)=(M+i). and the contribution from a sequence of r moves from q + 4j back to g.58) with P 0 (g. .j).
Up to > > this point.* We observe that the expansion remains unchanged under the replacements q — —q.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. J. *E. W. h — —h. Miiller [73]. q is only known to be approximately an odd integer qo = 2n + 1. We observe that there are regions. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. In fact we demonstrate in Example 17. the solutions and eigenvalue expansions again have the same type of symmetries.4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions.17.( 3 ^ . as pointed out earlier. . [137]. Its precise deviation from qo. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. remains to be determined next. 17.2 * +3) + ^ 2 (9g4 . 17.4. Ince [136]. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. S.92q3 + 70g2 .3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . L.^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. of course.^ . as a result of tunneling between wells.284g + 57) + (17.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) .5 the following proportionalities there: B[w(z)] = aA(z). Goldstein [115].62a) f R . It will be seen in Chapter 18 that in the case of anharmonic oscillators. Dingle and H.3. where solutions overlap. [116]. B. a= (8/i)3(9x) [1(91)]! 1 .
1 4 . Example 17. 17.32).COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 . a .(17. Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.62b) *z domains of overlap Fig._ ^ l CHAPTER 17.1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA. We write and expand this function as follows: .364 and C[w(z)] =a~A(z). The first term of the expansion of A(z) is the function Aq(z) given by Eq.4 Domains of solutions and their domains of overlap.1 ) ( 7T + .Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) .7T H Aq(z) = = C O S ^ . Solution: We begin with the solution containing the function A{z). and determine the constants a and a.
A*1) given by Eq. \ql)/2(w) 2i(9D[i(.!)(. the coefficient of the dominant factor cos' 9 _ 1 ^ / ' 2 (7r/4 + z/2) in the result is seen to be J (8/i)3 ( g . we have to impose the appropriate boundary conditions. These are the conditions of Eqs. these are conditions imposed on even and odd solutions at the point z = it/2. 17.92q 3 + 70q 2 .q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . (6.22).2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed. (17.46). For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. the coefficient of the dominant factor c o s ( . s .284g + 57) + • The other relation is obtained similarly. As we saw there.61) for the connection and Eq. Since the solution with A is obtained from that with A by a change of sign of z. (8.2.g. we obtain Bq[w(z)} H. (17.36).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.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. with e.i ) 1 (g. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .63) In Example 17. (17. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq.g + 4) (q + 4. i.17. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a._i)]! w*te» 2l(91 (.e. with (8h)li''1) (3g2 .53) for the expansion. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.
Oberhettinger [181]. Periodic Potentials We can extend the solutions (17.62a) and (17.e.63) to the domain around z = ir/2 by using the proportionalities (17. He'v(0) = Hev+1{0). 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. .66) ^E. Then y± oc ^ M e ^ i n . one obtains the first of relations (17.366 CHAPTER 17. and then replacing 2n by i>.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.66) then follows from the recurrence relation (6.66). p. Thus y+ = ce and y_ = se. The second of relations (17. (17.62b). give the following expressions for polynomials with argument zero: ~19)"(^)!. i.» a \dw _ C[w = 0}c_2h a w=0 = Q (17.e.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. With the help of the duplication and inversion formulas of factorials. Magnus and F. He2n+1 (0) = 0. a ± C[w{z)}c_2hsinz^ a (17.22) and to correlate the solutions to those for small values of h2.59d).g.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. W. 80. He'u{w) = wHev{w)He„+1{w). i.
67c) 4 2 \dw)w=Q C[w = 0] [\(q .69) If the right hand side of this equation were exactly equal to zero.67a) g 106g 87 2 14 /i 2 (17.136g + 9) + • • • . .7. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative. and similarly for C and its derivative.1) 7T . the solutions would be given by q = go = 3.Qo) + 0[(q .40g3 + 18g2 . (17.67d) Considering now the second of conditions (17.» 2 6 /i the condition reduces to the equation cot{ .67b) <f ~ 82g^ . we have cot j ^ ( g ~ 1)} = ~\<(l .14) from which one obtains 1 (q + i) TV cos{f(gl)}[i(g3)]!' and the duplication formula (2Z)\\/TT = z\(z — l/2)!2 2z .39 (17.g0)3].17.82g2 .67a) and (17.( g . f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .! ) ] ! [ . Hence expanding the cotangent about these points. and using the reflection formula (8.39 2~^h? (17.67c).68) Inserting here expansions (17.87 4 2 ¥w (17.\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .106g .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.11. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .
136q + 9) + . Thus we have 2 (16/i) qo/2 T2 4. Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .40gg + 18gg .368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .72) This leads again to the result (17.9.136g0 + 9) + • Clearly the last of conditions (17. i.73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.40gg + 18gg . (17.71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 . The remaining condition (17. = _a a 6 _4h (17... .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T ..65) yields the same result except for a change in sign.70) except that now qo = 1.70) + ^ 2 (9?o .120g£ + 467g .e.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg .(17.40g3 + 18q2 . we use \(q)~\(q0) + (qq0)(^J . .528gg + 3307Q ..5. Considering now the first of conditions (17.408g0 + 1089) 2 /i +• (17.65) leads to the same result except for a change in sign.
61). The dominant contributions were also found by S. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. i. p.e. though approximate. minus sign). (17. We see here that *R. The complete.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w .e.17. Dingle and H.87) = Azpteo). results from the boundary conditions. 9o = 2n + 1. *In Sec. (17. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. (17. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. W.. 20. i. 120. J. Goldstein [116]. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.88q0 . . See also F.3.A_(g0) ^ . + ^ 2 (9<?o + 89o " 78g02 .74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . n = 0. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. quite apart from applications. Arscott [11]. B. this becomes 369 = (16/i)^+1e4h l ..74b) The above results for the cosine potential are in many respects important.. Miiller [73]. In the literature that we follow here* in all cases several more higher order terms have been given. and that for the odd Mathieu functions se qo+ i and se9o (lower.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. M.1. example 3.2. The tunneling effect.3 we call this difference 26E„n.t From Eq. 17. (17. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig.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. just as any solvable problem is important for the insights it offers in a concrete and transparent form.3. . and not the separation of these as a result of tunneling.
evident through the exponential factor exp(—4/i) in Eq. Naturally one expects the solutions there to be related to those of the periodic case considered here.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.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. _3/+(z) + ____2/_(z). y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. and one can compare the methods. The exponential factor represents. Thus we do not reproduce every step. Periodic Potentials this nonperturbative effect. y^{z) = Aq(z)e~2hsin *. 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. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). Example 17. T y'_(0) y'A±*) y{z). M±TT) .74b) is also important for various other reasons.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. 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. (17. . the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft).6: Translation of Solutions For the largeh 2 even and odd solutions (17. in effect. results from the imposition of boundary conditions. 17. yA{z) = Aq(z)e2hsin '. From Eq. The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential. . the Schrodinger equation with the cosine potential is a special case of several other more general cases.2. (17.74b). We begin with the solution containing the function A(z). 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. The splitting can also be obtained with some methods of large order of perturbation theory. As mentioned at the beginning. Then we obtain first y±(z) = yA(z)±y^(z). The explicit derivation of the level splitting (17.370 CHAPTER 17.
but also to enable a brief familiarization with the Lame equation.31). (17.4. 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+.yV+\z^ = [j/+(0)] 2 (<? . y'_ (0) = \/2(4h . etc.17.4ft) ~ 8(q . (l^ifr^i^O + Cl^ifo^j/aCO cos  . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).y]z=„ = [y+yL . Similarly we find y(z±ir) = = and by putting z = 0: V. = ±ir).4ft) sin j ^n(q . where Aq(0) = Aq(0) = V2. The Lame equation has not been a widely known equation of mathematical physics so .VA DV 371 1 V±.2hAq(n)] . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). and y+(z = ±TT) = V2(q .[A'qM + 2hAq(n)] = iV2(q . A' (z = ±ir) = (. which is easiest by comparison with the foregoing treatment of the Mathieu equation.z ) . ( . Aq(z) The derivatives can be handled with the help of Eq.%/2(g .( ± 7 r ) = c ° s  .7 r ( g ^ l ) U + ( z ) q : i s i n { .4 17.• Then y'+ (0) = 0.1 )  . y+(0)~2V2.4ft.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .7 r ( q = F l ) } j / . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) .4ft) cos i TT(<J + 1 ) 1 . 17.( z ) From this we deduce for z = 0.( 2 ) . Replacing sines and cosines by the functional expressions.) cos < 7r(q 1) [ = .c o s  ~(<1 T 1) f y .T ( ? =F 1) p + (0).3 The Ellipsoidal Potential Replacing VA.g A ? ( z = ±TT). we obtain the conditions stated at the beginning. — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .q). A ?q( Z = ±TT) = ..Ah) ~ 32ft.
L. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . and was continued by Erdelyi. Periodic Potentials far. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane. 194) remarks to parallels with the Mathieu equation:".. Liang.Q. This line of investigation.fi2A. 19.. Erdelyi [85].5 < z < 5. k) for k = 0. f E .4sn4u]y = 0.K2sn2u .75) ' T h i s means double well.1. J.0 and .5.* As Arscott [11] (p. W. as alluded to at the beginning. (17. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.0.5 The potential sn2(z.75. Tchrakian [165]). but also cubic potentials. 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 ..98. 194) remarks. *A. Arscott [11] (p. Ince [138]. Arscott [11]. however. H... the nonoccurrence of the Floquet exponent. But more recently it has been observed to arise in various contexts.* and the work of both prepared the ground for presentday investigations. MullerKirsten and D. § F . 17. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here)...0. M. . The most conspicuous difference compared with the Mathieu equation is. H. inverted double well and cosine potentials (J.0. so that Floquet's theorem cannot be immediately applied . the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940." Fig. p.372 CHAPTER 17.
17. in front of the second derivative has to be kept in mind (mo being the mass). In comparisons with the Schrodinger equation. The function sn2u is plotted in Fig.K) = qK+^±. the equation ^  + [A . J. (17.77) where q —> go = 2iV + 1. (17. we consider mainly the Lame equation. MiillerKirsten.e.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). Eq. and of the derivation of the level splitting'.5 for several values of fc.\k\ < 1.2 = n(n + l)k2. Although the results have been calculated for the ellipsoidal wave equation. W. . (17. The range of the independent variable u is 0 < u < 2K. 17.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. In order to distinguish the above equation from that with the Lame potential.1. i. . i. in the case n2 — oo.4. The first step is to write the eigenvalue A as A(q. 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. If we put fl = 0. (17.78) We follow in this brief recapitulation the description in H. Jianzu Zhang and Yunbo Zhang [206]. cmx and dnu. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.75) reduces to Lame's equation and one writes K.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.K2sx?u}y = 0. the usual factor —h2/2mo.77) into Eq. The second step is to insert (17.e.17.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. . . we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions.2k. is the elliptic modulus of the Jacobian elliptic functions snu. 2 . (17. N = 0.
80) The domain of validity of these solutions is that away from an extremum of the potential. dnu =F cnu . Thus in the third step two more pairs of solutions B. A second solution is written y = A(u) exp < / nsnudu >. /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. C replacing yl. determine their proportionality factors) in domains of overlap (their extreme regions of validity). more precisly for dnu = cnu F 1 » . one pair in terms of Hermite functions of a real variable.e. one obtains again the same expansion for A. but solutions B. C which are valid for dnu±cnu <1.e.q. k' = Vl~ k\ \dmi±cnu/ 17.77). B and C. \/8K (dmz = cn« \ ' F . the other in terms of those of an imaginary variable.79) The very useful property of these solutions is that for the same value of A (which remains unchanged under the combined replacements q —>• —q. i. _ Solving the resulting equations iteratively as before.K)=A(u. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.q.K — > A(u)=A(u + 2K). . (17. one has to derive new sets of solutions there and match these to the former (i. (17. which are respectively even in u (or snu) or odd.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). i. A are derived.Es(ii). ^ (17. A into equations in terms of the variables z(u) = ^—{k' . A.82 .e. / z J . by transforming the equations for A..374 where CHAPTER 17. . A(u. dnu± cnu K Thus one can construct solutions Ec(u). Periodic Potentials . (17.K).
86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. W. W.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. the ellipsoidal equation.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} .e. (17.75).63)} + ••• .e. (17. i. p.384n2k4(q2 + 1)} q r {(l + A. who obtained this expansion for the eigenvalues of Lame's equation (i.17. Es(2f0 = Es(0) = 0. (17. F." A = qK±(l 1 + r {(l k2)(q2 + l)^{(l + ifc2)3(5g4 + 3 V + 9) + k2)2(q2 + 3)4k2(q2 + 5)} 2WK2 Ak2{l + k2)(5qA + Uq2 + 9) . 17.85) The first three terms of this expansion were first given by Ince [138]. Erdelyi. a of _ _ B = aA. *A. J.6(5g4 . This expansion is found to be in the more general case of Eq. as well as \duJ2K 11 (17. £2 = 0).38g2 . G. Tricomi [87].4. Magnus. Miiller [205].3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a.^ ^ { ( 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£.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. Oberhettinger and F. . 64. C = a~A. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.
2 /c +256fc2g0(g2 + 5)} (17. 2K.g 0 ) ( ^ .376 CHAPTER 17. (17.Es^ 0+1 .g0 is obtained by expansion around zeros.go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .40g03 + 18g02 .Ec£ 0_1 and Es2° of periods 4K. 2K and 4K respectively. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .4k2(q2 + 2g0 + 5)} + • • • ].i Here the upper sign refers to Ec*> l or Ec£°.89) one obtains the eigenvalues from which the level splitting can be deduced. from which the difference q . and the lower to Es*>+1 or Es*5. J. Finally expanding A(g) ~ A(g0) + ( g .90) + 3. 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. Periodic Potentials These conditions define respectively functions EcX°. 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 .:J {3(1 + k2)2 (9q* .l n 2^+ i J + 0(«i) (17. .1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .j 26K2 (1 85) = A(g0) + (q . Muller [205].91) tt H. 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).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)*.136g0 + 9) n 3. go being an odd integer.87) (17.78g^ . W.88g0 .
but also more easily generalizable.87) agree with those of Ince. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite.e.93) (17. (17.Ah2 sin2 u}y = 0.74a). K = 2/i.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods.* M G .J1 O. *H. J.f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. V. L. h . that calculations with the Schrodinger equation are not only easier.86).92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues. 17. Ince [138]. in which they dubbed the classical configurations Lame instantons. ./2i and hence A^ A/ A AU I2 . A . Dunne and K. Miiller [206].90) one has in this limit: /l±±\ ~K/U = e . (17. Thus in the case of the dominant contribution of Eq. (17. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A .* Apart from the choice of notation. W. i. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq.4. (17. Rao [79].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. it = x ± ^ .2h2 cos 2x}y = 0. 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.2h2 . the conditions (17. K2 = h2 (say).17. + E. .2h2 = A. this equation becomes y" + {A .4 h ( i e M 9 0 7 2 \. A = 0. Replacing u by x ± IT/2. One can see. however.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. Hence the conditions fi = 0.
In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. Ouvry [276] and Jianzu Zhang. i.5 Concluding Remarks The potentials we considered above will reappear in later chapters. . double well. 1 1 See Z. Thus. de Veigy and S.^ The equations considered above also appear in diverse new problems of physics. inverted double well and cubic potentials.W.e. A. Liang. A generalized associated Lame potential has been considered by A.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. N. Quian [123]. J . § . W. 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. Tchrakian [165]. MiillerKirsten and J. S. Cho. for instance. This is an interesting problem. Gu and S.Q . Rana [287]. **Y. MiillerKirsten and D. Ganguly [103]. See Chapter 25. H. A host of related elliptic equations has recently been discovered and studied. W. Sukhatme [148]. M.Y. J. H. H. also because the role played by the Floquet exponent in this problem is not yet well understood Jl Another recent appearance of the Mathieu equation is in the study of the mass spectrum of a scalar field in a world with latticized and circular continuum space.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). Kan and K. Khare and U.g. Periodic Potentials 17. J. A. See e. all basic potentials.378 CHAPTER 17. D. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. Shiraishi [49].
W. Wiedemann [4] and a revised version of parts of this reference by J. D. MiillerKirsten and A. M. [19]. Achuthan. Bender and T. The recent work of R.".* The fact that a main part of their work was concerned with the calculation of the imaginary part of the eigenenergy in the nonselfadjoint case which permits tunneling. Wu [18]. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. This seemingly simple problem revealed extremely rich internal structure . In particular the investigations of Bender and Wu. Weinstein [281]. ' T h u s A.. T. H. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. J.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. Priedberg and T. f C . 379 . Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. Bender and T.Chapter 18 Anharmonic Oscillator Potentials 18. T. J. W.. Wu. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. MiillerKirsten [163]. Liang and H.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. both of which make the calculation more difficult.Q. 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.
V(z) = \\h*\z* + \\<?\z*. These contributions may be given different signs.380 CHAPTER 18. 18. and thus lead to very different physical situations. V(z) = \\h*\z2 I 2i 4 \cr\z .1 The three different types of anharmonic potentials. 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. described as the case of the double well potential. with the potential described as an inverted double well potential. (3) Complex eigenvalues with tunneling: In this case. 18. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. *~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 .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. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation.
This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. i. We begin with the latter. (18. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. that the general applicability of the method becomes evident. 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. . Thus we are mainly concerned with the double well potential and its inverted form. The eigenvalues of this case are given by Eq. The eigenvalues of this case are given by Eq. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. it will not be possible to obtain the exponentially small contributions with convergent expansions. 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. 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.86) below. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). The question is therefore: How does one calculate the eigenvalues in these cases from the differential equation? This is the question we address in this chapter. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17.175) below. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). except for a change of sign of c 2 .18. The shift of the eigenvalues is best calculated with straightforward perturbation theory. however with decay as a consequence of tunneling. Case (3) is seen to be very different from the first two cases. The result is an expansion in descending powers of h?. The level splitting will be rederived from path integrals in Chapter 26. (18. In this case our aim is to obtain the aforementioned complex eigenvalue.1 Introductory Remarks 381 normalisable wave functions. as is sometimes hoped.e. 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). since the potential decreases without limit on either side of the centre. The imaginary part will be rederived from path integrals in Chapter 26.
18. We take here h = 1 and the mass rriQ of the particle = 1/2. (18. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.2. Anharmonic Oscillator Potentials 18. If suffixes 1/2.z Fig. a point which has to be kept in mind in comparisons.2 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. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. v(z) = ^h4z2 + ^c2z\ (18. 18. h 4 2 2h1> ~1/2 2c?. .2 The inverted double well potential with (hatched) oscillator potential.382 CHAPTER 18. This implies t h a t results for THQ = 1 (a frequent convention in field theory considerations) differ from those obtained here by factors of 2 1 / / 2 .3) "1/2 harmonic oscillator h8/25c2 .2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. The potential in this case is given by V{z) = v(z).1 refer to the two cases. 18.2.1) for h4 and c 2 real and positive.1 and more specifically in Fig. (18.
The result will be that derived originally by Bender and Wu.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. n = 0. (18.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0. (18.e.2. The problem here is to obtain the solutions in various domains of the variable. i.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.d2 .5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite. /i). then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.18. 18. 4 .5) becomes . (18.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w).. _i\(w) and q = qo = 2n + l. 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. the harmonic part of the > potential dominates over the quartic contribution and Eq.. and a variable w defined by setting E= b* 1 + A A and w = hz.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr.1. (18. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. \h?\ — oo and c finite.. to match these in domains of overlap.. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. (18.1/7  Vg(w)y(w) = o(J^j. y{w) oc Di.2.
2. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.2 We are concerned with the equation d2y(z dz2 where ^ 1 .11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (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 ) .13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0.10) into (18. (18.384 CHAPTER 18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. In order to arrive at solutions we set in Eq. as we shall see.+ ^ . y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions.10) E= qh2 ^ j. This observation allows us to define the pair of solutions yA(z) = . Anharmonic Oscillator Potentials T h r e e pairs of solutions 18.2 A (18.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.4(2.9) we obtain 1 h4z2 c2z4 A L2 (18. Inserting (18. (18.9) z =0 these Thus these c2z4 1/2 + • 2 (18.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.14b) . be and A = A(q. •A~2 hrz" 2^\ c^z V2 —— + (18.14a) 1/2 VA( ) = A(z)exp (18. 1 d r A!{z)±iA{z)±{ 0. h) is obtained from the perturbation expansion of the eigenvalue. Before we return to solutions we derive a new pair which is valid in the adjoining domains. these solutions are not valid around z = 0. as encountered and explained earlier. We observe — before touching the square roots in Eq.
Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. we do not pursue their calculation.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18. (18. (18.e.13) and A(z) that of the lower of these equations.34) below) — to be derived in detail in Example 18.1 and as a verification again in connection with the solution y# — as the other solutions.20b) . (18.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials. Clearly Aq(z).20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. 385 (18. 1 (18..Aq(z) approximate the solutions of Eqs.18. / c z Aq{z)+0 1/2 .16) For large h2 we can write the Eqs. Since these higher order contributions are of little interest for our present considerations. We take the square root by setting z2h^1/2 +. /H 4 Z h +~C Z 21.19) We see that one solution follows from the other by replacing z by — z.13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0. (18.15) where A{z) is the solution of the upper of Eqs.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18. i.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).4 . 1 2 4 2 Aq(z) + 0 (18.zh2 =()*2"(18. (18.
ft = .1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A.! ) ^ . a2 = 2 1 .l)A(z) 1 = tfA"(z) A . Vq+ii = Vq + 2i. Anharmonic Oscillator Potentials i. to obtain in leading order the eigenvalues E.e.[yA(q. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. h2.• Using Eqs.. z) ± yA(q. * = — .386 CHAPTER 18. we write > V±(z) = . X » / A q + 4 i = Mg+4i .20a) and (18. zA'(z) where the expansion coefficients are given by ao = l. i.:=*£ +i(5l). h2 —> —h2). (18. .21) Example 18.. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).2 . 03 = .17) and (18. A = .1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"^ The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z). 2 .v ' = o (« . one observes that with £>. A A _ 3 . . c*i = 1 1 .2 ( * ) 2 and from this or separately A"(Z) : 1 (q .. 04 = 5 .18) we obtain A » = H^rL^z) 2 z . W i t h proper care in selecting signs of square roots we can use the solutions (18.e. „ . away from the central minimum.20b).20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. (18.^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. 128 8 16 A = i. 1 . h2. i..f . (18.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . (18. z)). (18.20a).e. procedure.
i. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . give an equation from which A is determined..(q + 3)(q + l)Aqvri(q '^ ' . r ( CJw) = ^ . 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.Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. [1(9 + 1)]! .22) are parabolic cylinder functions D i . In the next two pairs of solutions the exponential factor of the above solutions of type A is contained in the parabolic cylinder functions (which are effectively exponentials times Hermite functions). (18. of course.'2H9i) ^faDfr™) and .w — > > ±iw) or functions R B w q( ) . (18. = •>— r. when i = 0. In this way we obtain the next order contribution A^1' to A^°>.18. w = hz. +1J±iw) (observe that Eq...e.—77 [i(9l)].. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ . (18. ^f( g + i)(±H2^ + 1 > \ — 1. _1J±w) and D_\.22) is invariant under the combined substitutions q — —q. We return to Eq. (18. Rq '. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. The solutions yq(z) of the equation Vq(w)yq(w) = 0.5).23) .2 The Inverted Double Well Potential Thus a term /j. those in Aq(z). and the coefficient of terms with i = 0.
(18. As an alternative to Bq(w) in Eq.4j)yq+Aj.29) ' A . F.123. we also have Vq+^yq+tj = 0.388 CHAPTER 18.±4) S4(9.q + 4j}yq+4j.l)2/?4.q + 4j] = c2SA{q. since Vqyq = 0. (18. Magnus.27) Now.0) = = = ±(q±3)(q±7). See Example 18. Erdelyi. (18. Vq+4j = Vq + 4j. and for j ± 0 : [q. (q±2)(q±S).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 + . S2i(q.3)y g _ 4 . W. l(q2 + l). Comparison with our notation is easier if this reference is used. 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 . (18.24) The extra factors in Eq. (18.Aj)(18.q}=A + c2S4(q. G. (18.0).^ For higher even powers of w we write i w2l V<i = Y.25) where in the case i = 2: SA{q. 115 . .( < ? . Actually these factors can also be extracted from W K B solutions for large values of q. Oberhettinger and F. j=i (18. and so Vqyq+Aj(w) = Ajyq+Aj(w).3. pp.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q. Tricomi [86].28) (A + c w )yq(w) E 2 4 1  2 ^ [q.±8) 5 4 (g.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.
Thus the next order contribution to yq is y H = ie 2^ —z. (18. We can now write the solution y(w) in the form y{w)=yq{w) + Y. h ) = qh . A = ^(q2 + l)c 2 + o(J^.32) Evaluating this expansion and inserting the result for A into Eq. i.31) Proceeding in this way we obtain the solution y = yW („. .(!) («.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.—A(q + 1) .18.) + j.35 ) . 2 . and even powers of q in combination with even powers of 1/h2.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.[rp ) Y.• q. p i(9»9 + 4 J'WK7». i. (18.e.—v*+*j ( 18 .q]=0.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. so that the entire expansion is invariant under the interchanges q . This type of invariance is a property of a very large class of eigenvalue problems.1.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. Equation (18.e. (18. .3 °) For the sum y{w) = y(°)(«. n — 0. and c 2  replaced by — c 2 . (18. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. ( 18 .) ty^^w) to be a solution to that order we must also have to that order [q. .10) we obtain E(q.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. (18. h2 —> h2.
Achuthan.37) For further details concerning these coefficients.i 1 + 4j) in complete analogy to other applications of the method.38) [yB(z)]argz=7T = [?/s(2. Again we can write down a recurrence relation for the coefficients Pi{Q. their recurrence relations and the solutions of the latter we refer to the literature. q + 4j) = 0.e.g±4] ±4 ±4 [g. Wiedemann [4] .—. q) = = 1. ^VO(<L<?) ^i(?. MiillerKirsten and A.390 where for instance v J CHAPTER 18. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument. i." Since our starting equation (18. J. there is another solution y(—z). ? T 4 ] [ g T 4 . q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t.q±A) = MM!i±MM+[M±8][g±8.)]arg2=0 These solutions are suitable in the sense of decreasing asymptotic expansions in the domains They are linearly independent there as long as q is not an integer. (18. > "p. o.11) is invariant under a change of sign of z. 2 4tPi{q.36) with the boundary conditions p o(q. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. g ± 4 ] T4 ±4 and so on. We observed earlier that these are obtained by making the replacements q —> —q. w — ±iw. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. and for j ^ 0 all other P0(q. q + At] (18.argz=0 (18. we may infer that given one solution y(z). We thus have the following pair of solutions VB(Z) VB(Z) = B w . W. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. H.
like w of Eqs. with normalization since the normalization constants are also asymptotic expansions. Integrating and expanding as follows since h? is assumed to be large. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0.g. 4 . h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.4)). Such contributions arise.g. (18. 18.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. Aq(z)) does not appear in any of the higher order terms. one has to stretch each by appropriate expansion to the limit of its domain of validity.g. Thus in the transition region some become proportional. . In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.q + 4j) as in ys.h^ih.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 .40) "Variables like those we use here for expansion about the minimum of a potential (e.The solutions yc. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0.39) are with the same coefficients Pi(q. in fact.z2 . see e. (18. the solutions of type A being valid away from the minimum. Vc(Z) = [V!B(Z)}q^~q. Mauss [192].18.h^ih 391 (18. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e.2.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. In order to be able to extract the proportionality factor between two solutions. J. h2) in which odd powers of q are associated with odd powers of h?.
Comparing the solution VA{Z) of Eq. We do not require this at present. The function Di.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.0x ! [ £ ( g .l ) ] ! ( . Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=. with z in ys(z) replaced by — z.2 « . not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. (18.43) (since z — — 2 implies > argz = ±7r). y^(z) we see that in the direction of z = 0 (of course. Anharmonic Oscillator Potentials Considering the pair of solutions ju(. .42) H W2 (ql)e$W I „ .42) we obtain for the solution ys(z).7 r . (18.43) W + 1)]! w^+V A .44) In the solution y~g(z).. eh* /12c* e\z*h? (18.z). 2 ) i ' ir! i iHim 1 3 j argw < 7T.4 i . we would have to substitute correspondingly the expression (18.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.41) with the solution ys(z) of Eq.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18.392 CHAPTER 18.(«. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . 4/T 4 Prom (18.45) . 41) The cases of the solutions of types B and C require a careful look at the parabolic cylinder functions since these differ in different regions of the argument of the variable z. > n. (18. 2 ^i![i(g4il)]!(2«.w yB(z) ~ B.
e. 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.y'_(0) ^ 0. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.. The first of the conditions (18. 18.. we see that in their common domain of validity yA(z) where a = =Vc(z).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. At z = 0 the solutions of type A are invalid.1 ) e 12c 393 Kgl)]^"1) l + O h? (18. (18.49) d yc{z) Again there is no such simple relation between yA(z) 18. hence we have to .9.. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.42) into y~c(z) w e obtain Vc(z) = (h*.7.46) However.11.18. (18. We proceed similarly with the solutions yc(z). i.yc{z)pansion (18.50) will be seen to imply qo = 2n + 1 = 1.21) as even and odd about z = 0. the ratio of yA(z)iVB(z) i s n ° t a constant.41).48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18. large probability for the particle to be found thereabouts.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .50) and y+(0) ^ 0. Looking at the potential we are considering here — as depicted in Fig. Recalling the solutions y±{z) which we defined with Eq.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq. and the second qo = 3.2 — 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 h2.5. (18.2.
53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i.^{(9+1)}.54) dio q(w) . (18.l ) ] ! [ . e. we obtain i V7T24( g . Expressions for C 9 (0).(«.F ? .i ( g + 3)]l (18. IT 4 dio In fact.394 CHAPTER 18./o^Tpif (q+i) . 1 [7(93)]! n «(°) = ."*(«!) ^ ' ~ [ .u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.! ( .1)('») + ^ r + l)]! ^.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)}. ( 9 + 3)]! s i n { .i)H = e.56) D 4(. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.(0) = . [C.2.( g + 3)}.55) Solution: Prom the literature. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.g.51) and 0 = V(0) = Jim i»xM .SxW) = \ i»fl(0) . Z (18.(0) (18.fc(0) Thus we obtain the equations (18.. M.(0) = and htoi)(°) ^—'. + 1)]!' D ' j . one finds that [ 1 IT (q + 1)1! = .2: Evaluation of yB(0). y'B(0). with the reflection formula (—z)\{z — 1)! = IT/ simrz. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.5 ( 9 + !!)(*"')• i ( 1) .. Abramowitz and I.(0) + ^ ( 0 ) l z—>0 2. y c ( 0 ) . [ .(iu)].) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.58) diu [  ( g .57) [i( 9 _l)] ! 2 i(9i) B. C.( g + 3)}. A. Stegun [1].oi(. Example 18.' '.. We leave the detailed calculations to Example 18.l)(°) Thus with the help of the reflection formula cited above: D B. — C..i ) . .
49).53) in dominant order and insert the appropriate expressions for a and a from Eqs. ~ ^ 1 5(9+1) 2i(''. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(. (18.+1iH]«=o = .i .6 ) } ^ — i t a n < — (q + 3) >. we obtain 2 i(»+i)[_i(._l) From this we derive C.sin{^(g+ 1)} = J . W .+i)(°) [ _ I ( .)[i(g3)]! and ^i(.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18. _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„. Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g . 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) _^_ .56). (18.46).62) .3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs. isin{f(g + 3 . 4 r T (18.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.3)}.18.i ( « + i ) [ _ i ( . (18.1/2)!.(0) = *>i(„+i)(0> 2 . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .sin{^( 9 .
by the rotations E > einE = E. (gg 0 )^±^ y e5?. (18.9. z^ ein/2z. + 3)} = .L ^ i .. 7 . In fact the left hand side of (18. ^ .11. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. .53) becomes 8 m(I(.9o) ^ cos  .63) Proceeding similarly with the second of relations (18. we again obtain (18.^ . For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.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..e.. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.5.65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0. . equivalently.^ >^_>m e J*.1. (18..1).9. (18..63) and (18. .64) the right hand side is an exponentially small quantity.( g 0 + 3)  + • • • ~ (g .65) Expanding similarly the left hand side of Eq. 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. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions.. (18.396 CHAPTER 18._ e =*.. the energy E is real. .2. i.i . Recall > the original Schrodinger equation (18. (18. z2 > z2. we obtain cos((g + 3)) = .2) with potential (18..63) becomes (? .64) for q = g0 = 3. 7. the vanishing of the wave functions at infinity)..5. and the left hand side of (18.64) about go — 3.. . Thus we have to determine these conditions first. With a Taylor expansion about go the left hand side of (18. 18.. 18.63) vanishes for q = go = 1. (18. This is Case (1) of Fig.go) ^ (1) It follows that we obtain for the even function with q — go = 1.64) In each of Eqs. 1 1 .53).
cos 36. for fiz * ±00 we have y(z) ~ exp < ± i / dz In order to decide which solution or combination of solutions is compatible with the square integrability in the rotated (c2 reversed) case. we set z w r = \z\eie. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. or — < 6 < . 6 6 In the case of the inverted double well potential under consideration here (i. i. i. i. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied.e. in the domain — TT/2 < 36 < > 7r/2.e.66) z3 = \z\3em.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. r / c 2x 1/2^3 = exp{ ± i — —\. i.68) . + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . (18.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. the case of complex E). Now. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase.67) Rotating z by vr/2. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36.e. if TT/3 < 6 < 0. (18. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0.e. the resulting wave functions vanish at infinity and thus are square integrable.18. replacing sin 30 by " 0 + ^) 1 IT = . 2 c2>0. 1/2. y{z)~exv\i[) 7T dV'V.e. —}.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations.
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. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure. i. and there impose the boundary condition (18. Anharmonic Oscillator Potentials V(z) Fig. for z > Too. (18. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. This is the boundary condition also used by Bender and Wu [18]. We do this extension with the help of WKB solutions. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. 18. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. We have to remember that we have various branches of the solutions y(z) in different domains of z. Our even and odd solutions y±(z) (cf.e. (18. Our procedure now is to continue the even and odd solutions (18.Z\.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.3 The inverted double well potential with turning points ZQ. Eq. We therefore explain our procedure first. Looking at Fig.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).Z\.21) to + infinity and to demand that they satisfy the condition (18. we match the WKB solutions .398 CHAPTER 18.21)) were defined in terms of solutions of type A which have a wide domain of validity. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. 18.68) for c 2 > 0.
p. 6. and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\.^ T rz\ 1 + yh* . ^ 2 4 •K + —(r. Sec.18. 18. h2 (18. p.. Dingle [70].70) where z < z\.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. We begin with the exponential factors occurring in Eqs. t To the right of the turning point at z\ these solutions match on to (r.*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) .4) for E and ignoring nondominant terms) t o . 291.4 h + c z 2 1/4 < . A J24 1 ? 2 • "^ •z h \—c z ~ — q h . I [195].zi)/ \ 1/4 ywKB\ ) z — g/i* .e. ^2? 1 2qc2 /2? . 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. 291.4.B.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ .e.70).2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. Vol. z2h4/A > c2z4/2. (18. Dingle [70].=c z 4 dz 1 L2 9 + 2 4 1/2 x exp .2. The distant turning point at z\ as indicated in Fig. 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.71) We now come to the algebra of evaluating the integrals in the above solutions. 4 2 T ' i. . equations (21). B. z\ ~ . In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. f R .3 is given by (using Eq.^ 4 1/2 + icV 7T Jz + (18. Messiah. (22) or A.69) The W K B solutions have been discussed in Chapter 14. *R. i. (18.
±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c..73) .^ I 2 2 1/2 zV = F 8c2 Jz [i . (18.40)) E± = exp _/^_2 f ± In A 2c2 .i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >. (18.211/2 1 \ /j.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .^ i ^ ^ c 2 .2 \ 1/2 .2.400 i. Eq.2.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .40)). zi 2q 1 c 2 [1 . Anharmonic Oscillator Potentials E± = exp = exp . Thus the above factor yields '2c2\l/2 h4 so that (cf. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).e.69)) rzi J<z z dz Z[&Z*]W = + ^Y /2 m '"U4 1/2 2d2 + 2c2 Since we are interested in determining the proportionality of two solutions in their common domain of validity we require only the dominant 2dependence contained in this expression.\qh2 + \z2h± . (18. Eq. (18.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf. . CHAPTER 18.
20b).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) .18.79) +S_(±)exp . > » Returning to the even and odd solutions defined by Eqs.20a) and (18.76a) P P r 2/1 2 " 9 ( .(J*i) VA 0 ) = PVwKB (Z C.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.21 1/2 .77) Now in the domain 2 — oo we have > \qh2 .75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J. h2 — — h2 is maintained.\z2h4 + J Z\ nl/2 \M fZ .*l) (3y^(z) (18.71) and these into (18.*l) (18.74) Comparing these solutions now with the solutions (18.2 The Inverted Double Well Potential 401 Thus at the left end of the domain of validity of the W K B solutions we have (l. (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.78) Inserting this into the solutions (18.76b) apart from factors [ l + 0 ( l / / i 2 ) ] . we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.
/i2) = E{qQ. we see that we have to demand that 5 + ( ± ) = 0. (18. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. i / 3 ± . (18.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.82b) This is our second condition along with Eq.34)./ ? = 0. (18.^ j .6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.76a) for (3 and /3..e.81) Inserting expressions (18..2.. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .402 CHAPTER 18. i.84) Expanding about q = qo we obtain £(g.h2) + {qqo)(^pj = E(q0. E(q.3.68).. (18.85) + .^ § ( 4 g 2 + 29) + O ( ^ ) . h2) = \qh2 . this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J . (18.^(q2 + 1) . i. Eq. (18. (18.e.h2) + (qqQ)h + (18.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1. 18. . (18.5.65)..
3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. h2) of the relations KWKB (Z) =PVA(Z).18. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. S W K B ( 2 ) = PVA(z). !*WKB(Z) ~ 2 . where the superscript {r. ft6 ^ = e. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z). i.e.g ( . Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0. Example 18.l ) ] ! > + = ^ 2 .86) will be dealt with in Chapter 20.2 .h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o . Since this is not without interest (e.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.~p{q. o 2 + l)^(4.h6/6c2 E = E(q0. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. h2). above) (1.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.g.yA{z). ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .3. zo} means "to the right of the turning point ZQ" . 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.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 .36) of their Ref.ZQ) . The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.
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. cf.zo)/ \ qh2 2 ZO + z2hA 4 2 . the (approximate) solutions* 1/2 exp • ± i ( . 2 z 2 ) / 2 e \ .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 .l c y dz 4 1/2 x exp •! / —(r>zo)/ \ 2 1/4 1 2y x exp 2 1/2 L dz qh2 2H + ~z2h4 4  2 c2z4 we nave where 2 > 20. . J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. N.] { hz combined with the particular constants contained in y ^ ' ^ g .. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .g In other words.ic224 2 4 qh2 + i^fe 4 . Schwind [247].zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20. Anharmonic Oscillator Potentials where zo > z. f x Zn 4) h2 2./ ^ 2 X 1 / 2 exp(ifi. 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.O." / 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.404 CHAPTER 18.
87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(.1)! = ( 2 7 r ) 1 / 2 e . in fact. We can reexpress these relations with the help of Stirling's formula.^ 1 .z ± ) = .J ' i J . We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0..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.18.DfJ»v we see Comparing now the expressions (18.13) of Bender and Wu [19].g. in Eq. Then . (j . (18.kyil(. we shall employ basically the same procedure as above or. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.zo).3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. (18." / 4 ( ? « / 4 . 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. as in the case of the Mathieu equation.3.41) for yA(Z). (18.88) + [EV(z))y{z)=0 (18. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . But there are significant differences.89) .e.23)) appear quite naturally. h4 > 0.3 18. „ / M 1 / 2 i . i.5 "l _ o f ._ .(r. 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.yA{z) with expressions 27WKB i J'WKB ' that the factors p. 4" ) (h?z2)^q+1) for q j£ odd integers.
94) (18.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^.95a) .z±) y = 0.z Fig. . the previous equation (18. (18. (18.91) becomes T>q±(w±)y(w±) = o(^)y.90) V(i)(^)=0.91) (18. (18. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.i>5. vW(z±) = ±Qch2. (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 .406 CHAPTER 18.V(z±) l{z z±fhA + 0[(z .4 The double well potential. 18.93) w± = h±(z .z±).
(18.2..18. d2u{z dz2 + v+ 2~r u(z) = 0. we obtain another by replacing z by —z.96).95b) By comparison of Eqs. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].97b) Our basic equation.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' (18.93) for E into Eq. (18.96) dz2 + 5 « 4 +  .100) .87).3 The Double Well Potential where ^ 407 d2 •« 2 2  d8.95a) and (18. Inserting the expression (18. n = 0..97a) U(z) = (zz±)2 + 0[(zz±)3}. given one solution. i.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) = 0.1. 18.. we conclude that in the dominant approximation q± is an odd integer. we obtain d*y (18. (18. qo = 2n + l. Eq.95b) with the equation of parabolic cylinder functions u(z) = D„(z).e. (18. Thus again.3. is again seen to be invariant under a change of sign of z. (18.2 T h r e e pairs of solutions We define our first pair of solutions y(z) as solutions with the proportionality y(z) = exp \h\ f uV2{z)dz (18. (18.i 4 f W » = o..
Integration of Eq. We leave the calculation to Example 18.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. i.yA(z) are associated with one and the same expansion for A and hence E.104) Both solutions yA{z).102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18.101b). n± = V2h2.^ A"{z) + 2c V2 *A(z) (18. (18. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0.h2) . yA(q.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).101a).102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. z) may be obtained from the solution yA{q.^q(l7q2 6 4/i + 19) + (18.e. i. (18.e.106) . h2. Aq(z) = Aq(—z) = Aq(z). z) = yA{q.408 Since CHAPTER 18. ~h2.103) Looking at Eqs. h2. z). The result is given by A and E(q. h2. h2 h4 2c' with g+ = q.h2.1 6 ) .101b) To a first approximation for large h2 we can neglect the right hand sides. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —. Vq = {z2+z2)— + (qz+z). (18.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . 8hw (18. we observe that the solution y^iq.105) "8 + 4 ^ . (18.c W + 1) .4. (18. but the solution is a different one. these equations can be rewritten as + (qz+z)A{z) = . z) = yA{~q.
First.e.112) From these we obtain.108) We wish to rewrite this expression as a sum y coefficientiv4.{z2 .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . (18.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. z • . <4q+2 — z — z+ Aq+2iAq+2j (18. .3 The Double Well Potential 409 Example 18. J = 0.4: Calculation of eigenvalues along with solutions of type A Show in conjunction with the derivation of solution y^ that the leading terms of A and hence E are given by Eqs.(l + + 2  z y* .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z . (18. however.jf2i(2) We also note at this stage the derivative of the entire solution yA(z) taking into account only the dominant contribution: yA(*) =* A*2 V2 .103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. (18.4zz+q+ z\{q2 + 1)].qz+)2 . for instance by componendo z z et dividendo. (18.110) Aq+2+Aq2 9+2 A.z2 ) N t.z\) + 2z(z .. We know the first derivative of Aq from Eq. + Aq+2 + Aq Aq+2 . . Thus it is natural to explore analogous steps. 92 ZZJf (18. The first such step would be to reexpress the right hand side of Eq. (18. as a linear combination of terms Aq+2i.18. . Solution: The structure of the solution (18. . such as periodic potentials.+ 2 ^ ± l + 2^±i + .Aq 1 + 2^± 2 Aq+2 (^M^f (18. ± 1 . i. ± 2 .105) and (18.113) 2 . = Aq+2i+2jAq.101a) with A replaced by A . .qz+) [2z2 . (18.106). .103): .102).114a) . A Differentiation yields » = ^^w (18.109) We observe some properties of the function Aq{z) given by Eq.2 z\) • Aq$4 £Aq + Aq—4. we reexpress A'q in terms of functions of z multiplied by Aq. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.
(0) V2 ' 2c K + h4+Ao .3 ) A .115) and (18.108). cf. •••].4 ( g . A(°~) = Aq. In particular the dominant approximation of A is obtained by setting (q. where the lowest coefficients have been determined above as (9.e. (18.2\2 (z* . _ 4 .l ) ( g .101a) the contribution R .2A g + A g _ 4 ] . 1_4^±^+8^±±12^±^ +.4 4+* + .2Aq + Aq4] 1. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.33) and (17. Since VqAq = 0.4)A„_ 4 + (q.116) into Eq. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18. (18.aAg+S 12+ 16 (18.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18. (q.114b) . (18.4A ? _2 + 6Aq .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 . we obtain 5 K = ( ^ f ) [ ( 9 .q2)Aq2 + (q.410 and from this = Hence with Eq. (18. leaves uncompensated on the right hand side of Eq.q)Aq + (q.<? +{q.4Aq+2 + Aq+A Inserting (18.115) (z*z%) 2 i2 (z 2 . .z\Y Z2 + {~)\Aq+42Aq 1 224 Aq. q + 4)Aq+4 + (q. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.34). ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential. (18.117) 4(q + l)2Aq+2 Here the first approximation of A. (17. Anharmonic Oscillator Potentials A 1 . . QA?+4 _ 10 A + 6 .+6 + • • • ] .9=F4) = (g=Fl)(?=F3). Eqs.q + 6)A<.4]j. (q.118) 2s/2c ~^j[(?.116) [Ag4 .q + 2) = 4{q + l ) 2 .112) CHAPTER 18. (18.119) It is now clear how the calculation of higher order contributions proceeds in our standard way.q + 2)Aq+2 (18.112).l ) 2 A . q) = 0. i.
i.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9. y'+(q.yA(z) derived above are valid around z — 0.28. (g. h2 2y^ Age? h4 (18.124) . in the domains away from the minima.h2.g) Az. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq. (18.0) = y^(q.120) The sum A = A^ + A^1) then represents a solution to that order provided the sum of terms in Aq in Rq and Rq is set equal to zero.z)±yA(q. A' (0) = q/z\) y+(q. 9 + 2) Aq+2Az+ (9.2 ) (0) (18. g) 8z 4 . 9 + 4) ^+48z+ (18.(0).h2.0) = ^ . *)] (18.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. g .4 ) ( 0 ) ( g . of course. (i) _ / V2c\ [ ( g . \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.e.g ( 1 7 g 2 + 19).h2. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g . h2.0)=0.g2)(g2.{^ h2.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.105). where Rq is the sum of terms left uncompensated by AW. thus yielding the next approximation of A as given in Eq. (18.h2. g + 4)(g + 4.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.123) ^[yA(q.101a) the contribution Rq .g) 4z+ {q q)+ 2r (9. Terms jj. z)\ = 2^A^ h2 '>z)± VA(Q. y'_(q.0)=0.g) 8z+ (g. i.h2.94) 8z+ (9. h2. for i = 0. (18.e.g + 2)(g + 2.94)(g4. g .122) The solutions yA(z).
we > . (18.1 ) D (18. Since w±(—z) = —h±(z + z±). 18. Bq[w±{ ^ . (18.23) with appropriate change of parameters to those of the present case. the equation is — with differential operator T>q as defined by Eq.126b) These are solutions again around a minimum and with yB(z). (18.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)).93) and setting w± = h±(z — z±).91). 2(9+) .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.yc(z) with complex variables and Cq(w) given by Eq. but w±(z±) = 0. We see already from Eqs.3. (18. (18.2A y(w±).3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).95a) and (18. + 0(hl2). Anharmonic Oscillator Potentials Our second pair of solutions. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. and another VB(Z) ± 25^ch3wl + c2wi . yB(z).We draw attention to two additional points. This means.125b) It is clear that correspondingly we have solutions yc(z).Evaluating the exponential factor contained in VA{Z) of Eq. Moreover. Inserting (18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).99) for z — z±.yB(z) is obtained around a minimum of the potential.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). (18. we have w±(—z±) = —2h±z±.95b) that the solution there is of parabolic cylinder type.^ % [i(?l)]!24(«.412 CHAPTER 18.
97b) ~ exp . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. .147)).97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.127) Recalling that around arg w± ~ 0. (18. Thus for h very large these turning points are very close to the minimum at z+.i*l(«^ 4'"+^+ 4"'+v ( « . a= ^ . Here z± = ±h?/2c are the positions of the minima of the potential. Eqs.z±)dz . and comparing with Eq.172)) we add here comments on this approximation. Later we calculate the coordinates of turning points ZQ. (18.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z).172) obtained later. . Allowing z to approach z+ in the solution IJA(Z).« + ) 2 (9+1) e (18.• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I . we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i . (18.frW.129) . ein+z+e in+(z 22+(<?+i) .3 The Double Well Potential (cf.. 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.• • " . (18. (18.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. . 2 An+z+e 4w+ (18.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.ZI and find that these are given by z+ + 0(l/h) for finite q (cf.^/w + ± 1 . . .18.146). (18.
414 and CHAPTER 18. S. The pair of solutions of type B is defined around argz = 0. Since it is more probable to find a particle in the region of a minimum than elsewhere. de Deus [66] and W. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. R. The involvement of these WKB solutions leads to problems.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . The solutions in terms of parabolic cylinder functions have a wide range of validity. since. K. and this means at both minima. K. they assume large quantum numbers.z+)]fr+V ' (18.4 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. Furry [102] . The next aspect to be considered is that of boundary conditions. even above the turning points. g . We have to impose boundary conditions at the minima and at the origin. Bhattacharya and A.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. Rau [22]. 18. and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima.(18.7r and the solutions of type C around argz = ±7r/2. S. P. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.3. we naturally expect the wave function there to be similar to that of the harmonic oscillator. Bhattacharya [23]. but it is clear that none of the above solutions can be used at the top of the central barrier. D. J. basically. Various investigations^ therefore struggle to overcome this to a good approximation. Thus 1 E .+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . 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. H.
133) y+(z±)^0. The odd wave function then exhibits a correspondingly opposite behaviour. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. y'_(z±) = 0.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18. (18. (18.132) Fig. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable.5.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. and y+(z±) = 0. as indicated in Fig. as indicated in Fig.Zii yB{z±) a ± a yciz±) (18. 18.135) Hence the conditions (18. an even wave function can also pass through zero at these points. y(z±)=0.132). However. and y'±(z±) V'B(Z±) ± =y'c(z±) (18. We have y±(z±) = ^VA{z)±yA{z)]z^.5. 18. ^(*±)#0 (18.136) .5 Behaviour of fundamental wave functions. as indicated there.18.134) y'+{z±) ± 0. y{z±) £ 0. 18.
1 1 . .. Eqs..«*{=(..2 we can rewrite the second of (18.m Using Eq. also (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.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 . (1.136) as  M .142) (<Z9o)J(l)^°+1)(l)1/2 . (18.138) Using formulae derived in Example 18.416 CHAPTER 18. 7 . . . (18.9. (18.136) the dominant approximations we obtain (cf.5.54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula. +1)} = 4 m. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. . _ 8)} s . Thus sin {^ +1) } = ^^>^30fe^i.„*{!<.
We deduce .6 Turning points z$. We do this with the help of WKB solutions.18.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. Since the type A solutions are not valid at the minima. 18. y+(0)^0.1 " <813 1 4)  In Example 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 emphasize: We needed the solutions of type A for the definition of even and odd solutions. V(z) — • z E=V Fig.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin. y'_(0)^0.5 (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.z\.3 The Double Well Potential Thus altogether we obtain 417 ^ • .3. we must also demand this behaviour here along with a nonvanishing Wronskian. y'+(0) = 0.» ^ B ^ ' . 18. (18. Hence we have to impose at z — 0 the conditions j/_(0) = 0.
e. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) . Anharmonic Oscillator Potentials from Eq.418 CHAPTER 18. + £\h%U(z)=Q. As a consequence.e. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.144).24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. (18. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. We also note that the height of the potential at the turning points is h8 qh V(z) 20.147) Since the minima are at z± — h2 /2c with /i 2 very large.145) \qh\ + \zW Using z+ = h2 /2c.148) ("f * "The superscript (I.^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. ZQ) means "to the left of 20" . h8 25c2 (18. one finds that . to the left of ZQ where V > E. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer. In the domain 0 < z < ZQ.\qh\ + \h%U{z) (18.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. i.
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) .^ e therefore have to consider the exponential factors occurring in (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 >'^.148) and (18.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I. (18. )+^*+)2lr} rnD h\z'2)}z Q )\ = . (18.^ j O e .* 4+ . Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.^ + ) 2 + igln2(zz+).i (2q ln (18.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing. we have to match yA(z). 20 dz \qh\ + \h%U{z) .18.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .149) In order to be able to extend the even and odd solutions to z = 0.
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. 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.125a). l/2 ~h\U{z \(zz+?h\ (18.e.ZQ) . However. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.154) . since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. This is the aspect investigated by Furry [102]. of course.420 CHAPTER 18. We see therefore. t h a t the Stirling approximation is amazingly good even for small values of the argument). Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. 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 (18.152b) [\(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. the results necessarily require adjustment or normalization there in the q—dependence.126a).125a) and (18. 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 —(1.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. In a corresponding manner — i.
(18.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. However.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4. 7 of t h e matching relations y^NKB\z) i. (18.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.156b) Using again the duplication formula^ the ratio of these constants becomes 1 7 (h2+)i/2(2z+)i [faW e 2n+z+ — Aft 2 72 (18.99).159) 7 i(«i) r(9^3) ^2~ j(9l) . this result is somewhat imprecise.K?+1).143) T^^*)) 27T (18. and is shown to reproduce correctly the result (18. Comparing for z —• z+ the W K B solutions (18.103) and (18. The relation (18.104). . [\{q — 3)]! are really correct replacements of [\q}\ only for q large. linearly matched) W K B solutions. Returning to the even and odd solutions (18.e.153). \hl.155) (2z.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. as the results also seem to support.5 for the calculation of the tunneling deviation q — go by using the usual (i. we obtain in leading order (i. . multiplied by (1 + 0(\/h\))) the proportionality constants 7 .158) Applying the boundary conditions (18.e.154) with the typeA solutions (18.157) Since the factorials [^(q — 1)]!.e.18. (18.157) is used in Example 18. 27T1/2 9/4 1VA{Z). (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.n('>2o) 3/WKB(Z) =1VA{Z)I (18.
(18.. F. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.. 60 and 361.z2)(b2  z2). Byrd and M.163a) (18.iV2 ! i .149) near z = 0.05. p. p. D. G2 K [i 4.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^2 —pz I V2 dz ^?^ h+ Z 4? + ^f ^2<«2 (18.148).163b) The integral hi?) . 3G2 12V2c ^See P. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. Friedman [40]. (18.. n + uY 2c „ ^6 8V2c2 (18.164) / 2 (0) z ca [(1 + k2)E(k) . .161) we can rewrite the integral as r I2{z) = = dz^{a2 . .19. formulae 220.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.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.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. 213.k'/2.422 CHAPTER 18.
(G\ .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 . .1)]' Gfo.) = 2 \. n = 0.3 )l ! . ^ ^ . which implies small G2. 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 . 1. where the result is shown to be (with q ~ q0 = 2 n + l .. .165) in agreement with Ref. ..e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. The nontrivial expansions are derived in Example 5. /2n + l 2 ± — . n.( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .1..( ft 7. [4] misses an n—dependent power of 2 in the result.± ' 3G 2 ' T ( ^ (18.18.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. [167]. .r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . ± .1 " e " O T (18 169)  Y^V / 4 tfo.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( ..2.
.173).3 concerning the exponentials. . Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs. (i*™) We obtain the energy E(q.6 Eigenvalues and level splitting We now insert the replacement (18. we obtain for E(q0. 2 . with the help of Eq. . . .172) into Eq.1.143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.143) with qo — 2n + l.3.3. (18. 18. (18.106).174) .h2) + (qq0)^=. (18. (18. Inserting the expressions for h+ and z+ in terms of h. we see that the relation is really an identity (the preexponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton).. 18.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z..424 CHAPTER 18. h2) the split expressions 29o+1/i2(—W2 h6 (18.h2) and hence the splitting of asymptotically degenerate energy levels.^ W ^ / 4 t o .»W. 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.n = 0.175) ~E0(q0.159) — 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. we can therefore impose the boundary conditions at z = 0 by making in Eq. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. (18. (18. and obtain ( * . (18.157)..i ) ] ! e ~ s A ' .
e. i.3 The Double Well Potential where £0(<7o. i.2 = 2 .143) with (18. which evidently supplies some correction terms. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.2.n = 0.174) and (18.. 1 7 425 2 . the following expression for large q ~ 2n + 1.. (18. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 .1. T \ *n ! 1 ' (18.e.176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB. Combining Eqs..h2)E+(qQ. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency..q2 7. . i n U n / ! Thus AE(q0.157).^ 2^ h* . Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^.fr2) is given by Eq.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . K. 1 u2 c 2 (3 9 2 + l) ^ « b . for finite h2.V2^\ \n+ i . (18.h2) = E_(q0.e.106). can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. the pure WKB result of Banerjee and Bhatnagar [14] (i. 2h4' 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 / » : = \ . The result (18.7 =( e N n+1 .178) "In S.e. the level splitting.e.18.
Furry [102].180) h8 E0(qo. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. H.h2 <> — h2. (18. h4 and c2 replaced by 2E.3)).2 / 7. Eq." Of course. This. h = 1.178). and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18. (18. 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. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. We see therefore. Bhattacharya [23]. we would have obtained the result with E.2X2 V(z) = \z2tL) . these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.^ 2 + ^h2 If in addition the potential is written in the form A / . ** Comparing the present work with that of S.176) together with Eq. as is also explained by Bhattacharya [23]. the Furry factor corrected WKB result follows automatically.4/4 here. 2/z4 and 2c2 respectively (see Eq.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. however.181) "W. A>0. . that if this symmetry is taken into account from the very beginning.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. K. as we do here. and A there with c 2 / 2 here. we identity the parameter k there with ft.6) to yield an improvement of WKB results for small values of n. h2)\mo=i = .6 \9o/2 e~h&l^\ with (m 0 = l) (18. Then the splitting A ^ W B of Eq.426 CHAPTER 18. (66).. (18. (18.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.
7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator.11). (18. Liang and H.3 The Double Well Potential 427 a form frequently used in field theoretic applications. P. J. M. MiillerKirsten [186]. W. Lee [98]. and g2 is given as g2 = r]2/m3. ZinnJustin and U.h>) * 2V2^^^y/2e^3/3A. J. ZinnJustin and U. (18. Patrasciociu [110]. .c2 = A/2. 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. 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. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. Friedberg and T. M.^ *See E.182) implies AlE(l. first paper.§ 18. D. K.15). however. J. so that by comparison with Eq.W. Jentschura [293]. D.* A similar correspondence is also found in the case of the cosine potential. The comparison with Eq. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. Eqs. m 2 <• p 2 / 2 . Mansour and H. See R. Gildener and A. (18. these are presumably not of much interest in physics (for reasons explained in Chapter 20). 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. formula (4.—Q. Equation (18.181) therefore implies the correspondence rj1 <> A/2. ^To help the comparison note that in J. Jentschura [293].180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.34) and (E.^ Thus for the case of the ground state Eq. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1.3. §J. Bhatnagar [14]. (2. H.18. D. Banerjee and S. We considered above only the symmetric twominima potential. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation.88) h4 = 2/x2.
like those for anharmonic oscillators. has also been the subject of numerous numerical studies. in both symmetric and asymmetric form. mostly in connection with instantons. as he discusses. In fact. A reasonable. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. Santi and N. Perturbation theoretical aspects are "See e.[131]. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Mahapatra. we cite papers of the Uppsala group of Froman and Froman [100]. N. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. [19] — that the expansions considered above are asymptotic. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. P. Very illuminating discussions of double wells and periodic potentials. B. B.428 CHAPTER 18. Martin. Ashbaugh and Harrell [12] and Harrell [130]. in principle its Schrodinger equation is an equation akin to equations like the Mathieu or modified Mathieu equations which lie outside the range of hypergeometric types. The double well potential. 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. . though not exclusively numerical. for instance. As references in this direction. 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. can be found in an article by Coleman [54]. There is no reason to view the anharmonic oscillator differently. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. Pradhan [182]. Simon and Wightman [178]. 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.g. as some relevant references with their view we cite papers of Harrell and Simon [129]. 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].
Gutschick. T h e r e f o r e it is convenient t o deal w i t h this here. 2 a. We obtain the expansions from Ref.^ + • (18.18.G2q + . We have 4 ~k' 2u(l . X +2 + ' ^ (18. fc ~ 1 . [122] as 13 1. q ~ 2n + 1.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18.2G^2q + 4G2q . Thus k' and hence k' = v ^ ( l .186) (18.\G\ + • and 1k2 =2Gv/2q4. Wave functions of symmetric and asymmetric double well potentials have been considered in the following reference in which it is demonstrated that actual physical tunneling takes place only into those states which have significant overlap with the false vacuum eigenfunction: Nieto. (18./4 E{k) = 1 + • l n In (18. Saxena.183) S o l u t i o n : We have k2=1^. Srivastava and Varma [24] and Bhatacharya and Rau [23]. which is assumed to be small.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 .f + • • •) 2u\ .u)1'2 = V2u{ 1 . k! (18.189) =2u 2u2 (18.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. Datta.. Example 18. 1+ u : Gy/2q.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. Bender.187) We now reexpress various quantities in terms of u.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . + (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 (18.192) 'F 12 16 . (18.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 . Biswas. Cooper and Strottman [221].
430 and CHAPTER 18.3 u .u){4 + 3u(l .1)(1 .197) . (18. (18.4.3) 1 + In ^ .a = (l«)(jl 2u/ 2 ) . 1 3 « 2 = . ) J 4 (18.u) + 2} + i u ( u . .1)(1 .u) 2 {8 .3 u ( l .13)} + ~u3(u .13)}.^ L ) H[(i _ U ){4 + 3u(l .18u 2 + 39M} + 12u 3 (u .6w2 + 13u} + 12u 3 (« .u 3 ( u .195) we obtain E(k) 1 + ln uK(k) .u ( l .3) .13)} + ±u3(u .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .« 2 + 3u] In [ ..3 « .e.u){24 .u ) (^){« + 5^ 1 )} + 5.u){24 .u)} u{u(l .u ( l .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .— ) . (18.195) With (18.6 in the denominator and in the last step pick out the lowest order terms in u (i.3u(6u .3) . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.« ) ^ { 2 4 .1)(1 . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + . 8 + 9u 2 .3u(6« .u){24 . up to and including u2): u(u .193) 2u(l .u)} + — u ( u .3u(6u .{ M 2 ( 1 .3).u 13 \ 1 ln(± k> fc'2 +  (18.«) + 2} + V [ .3) = = ~ .{l u ( l u) / W 2 + 2} .196) Now consider the last line here without the factor 2.3u(6u . 8 + 6 u 2 . 8 .2{«(1 .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .u)(u .u) 2 {24 .194) ln '^J + 2 In i/2u )+H l + 2u(l .u) + 2}] 1 H 1 o u(u .1)(1 .u)} .13)} + 12u 3 (u .194) and (18.3 .u) + 2u} + i u 2 [ 2 + (1 .3 y .
155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).u H I .3).6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. [(1 . Inserting here from the above expansions the contributions up to and including those of order u2. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / . / to + 3GH~ \ 8 .198) 9 «. V^LJ 4 2 16 ^ .196) and (18.«){4 + 3u(l . we have.. (18. . Eqs.173) t o be evaluated. l2 = ^ ( l + uf/2[E(k)uK(k)}.uK(fe) ~ 1 + In I — = Mu2(3u.2 (18. 2 . (18. where w±(z) — h±(z — z±).2 /o 2 431 (18. i.196).200) Remembering that u = Gy/2q.3 The Double Well Potential Now consider the bracket [. (18.128) and (18.4u .] in (18.2{w(l u) + 2}] = u(3u2 . ' 8G 2 q = 4 2 q ln f 26/2 \ o .128) and (18. i.3) ..it)} .zo) HWKB normalized _ ( h* y^eM.^HfHGH Example 18.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.e. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .4u ..18.148) (for particle mass 1/2 and in dominant order) is given by H. From (18.e. (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. + iu^Zu2)^~u+ — +0(u3) l .197) we now obtain 4 1 • E(fc) .199) We now return t o the expression on the left of Eq. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .195) with (18.2 l_ln(1 JL\ V {2.
143). zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.*o)/ ^WWKBOO (18.** oo / dwD\. The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf.204) .146) and (18. Oberhettinger [181]. and obtained as Eq.e. (18. i. oo .432 CHAPTER 18. We start from Eq. Magnus and F.147). y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji.e. 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 . Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.l)]![i(0 . Eq. W.hA/2 = h^/4 this implies .35)).3)]!23<«D [5(91)]' >A. 93.158).g. 1 7(l. are given by Eqs.f hi\ VWKB. i.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+. normalized = We see that the normalization constant is independent of a quantum number.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.Aw) = s/2i 2("!) V (41) and [(9 . (18. by using the periodic WKB solutions below the turning points. Solution: The turning points at zo and z\ on either side of the minimum at 24. 82. ° \Edzy/\EV(z) V(z)V4 VWKB. Example 18. Then ((r. pp. (26. (18./ .zo). 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 [ . (18.
i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. i. (18..132) and (18. i..2.g.e.207) (18. z{) we expect tt (18.J^Wj +(18.211) 7 z l .210) Choosing the point z to be z+.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 . i. 1/2 .Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = . from ZQ. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. 7T\ + ^ 4 7r\ dz . Sec.( .2.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.6...209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.e. . Then at any point z € (zo. JV = 1.133) 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. of course.3.212) tt See e. 6.r + ^=co S [r. Messiah [195].. this implies sin cos sin cos Thus e. from z\. approach the minimum also from the right.g. \h\U{z) 1/2 . We could.e..iK h\U{z) + / 4 (18.18. A.205) We now apply the boundary conditions (18.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .^r + + .
215) which results from the factor of 2 in front of the sine in the WKB formula (18. Banerjee and S._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ .149).213) The integral on the left can be approximated by 1. M. 9. K. (18. + l ) j } ^ 1_ 27' (18. We now see that Eqs. The prefactor 2\/2h2/n is.215) we can expand the left hand sides about these points and thus obtain . (18. .27 9 .3.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . (18. .5.2 .146)). Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.210) assume a form as in our perturbation theory. D.215).434 CHAPTER 18.209) and (18.214) {«+"i} 0 for q = go = 3. .X++4 7T We rewrite the quantization condition in the present context and in view of the symmetry of the potential in the immediate vicinity of z+ as r J z0 1/2 dz qh [\ +\h+u^ \ 1/2 •K (18. Bhattacharya [23]. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. . ' /^i .205). 7r"y in agreement with Eq. (18.148) and (18. K. The formula thus agrees with that in the literature. Lifshitz [156]. P. where u> is the oscillator frequency of the wells.. 1 1 . and L. (18. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).215) we can insert for 7 / 7 the expression given by Eqs. S. the level splitting is given by AE(q0. 2IO/K.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. In Eq. they become c o t { ( .143). 7 . (18.216) Inserting this into Eq. i.. (18. Bhatnagar [14]. in fact (with our use of qo).90 — =F— for q0 = 1.e.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. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. W "Cf. 5. Landau and E. and cot< (q + 1) '!}» for q = 50 = 1.Zi / r.
MiillerKirsten [9]. Feshbach [198].* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. also in P. Case [45]. are not as troublesome as one might expect. 435 . Giittinger and H. f K . Nonetheless it was thought that a certain formal analogy could be seen if the field theory is supplied with Euclidean spacetime concepts at the expense of sacrificing the interpretation of the interaction in terms of particle exchange. n > 2. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. W. W. Morse and H.1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state. there is no need to have only field 'Generally a potential more singular than the centrifugal term in the (3 + l)dimensional Schrodinger equation is described as singular. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. However. the investigation of singular potentials in nonrelativistic quantum theory was motivated by a desire to obtain a better understanding of the (then presumed) singular nature of the nonrenormalizable weak quantum field theory interaction. M. 'For a review from this perspective with numerous references see H. M. in an early investigation Case^ showed that potentials of the form r~~n. Aly. H.* However. J.Chapter 19 Singular Potentials 19. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful.
Singular Potentials theories in mind. 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. MiillerKirsten. N. Manvelyan. that with the hyperbolic cosine replacing the trigonometric cosine). Park. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. MiillerKirsten and Jianzu Zhang [226]. G. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. H.436 CHAPTER 19. In fact.2 19.2. D referring to Dirichlet boundary conditions) in 10dimensional string theory. Callan and J. 19. Hashimoto [124]. J. Recently. J. W. D. some also permitting further research. § 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). but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. Singular potentials arise in various other contexts. K. Liang and Yunbo Zhang [189]. R. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane.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 particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. J. H. S. S. (19. Tamaryan. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. Gubser and A. for instance.2) § C . .Q. Maldacena [41]. W.e. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. In the concluding section we refer to additional related literature and applications. It is therefore natural that we study this case here in detail. S.
It may be noticed that since I is an integer. Liang and Yunbo Zhang [189].( I + 0.28) below).Q.mo = 1/2 and h = c = 1. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . R. H. . This is rarely possible and therefore this case deserves particular attention.4) first for small values of h2 and then for large values of h2. r = 7 e z . 8. We observe that with the replacements p2 z^iz.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. MiillerKirsten. (19. (19. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. and as we have in mind in Sec.3a) In the case of the attractive potential (as in the string theory context referred to above. using the method of Dingle and Miiller of Sec. ig h \j g (19. Eq.3b) 7 = % j.7 that we employed also in previous chapters. J. J. In the following we study the solutions of Eq.19. (19. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively. \[\ in the correspondence is nonintegral. W.3 below) g2 is negative and hence h? is real for E > 0. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. 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. It is clear that we draw analogies to our earlier considerations of the periodic potential. The following substitutions are advantageous: y = r1/2cf>. This has advantages compared with higher dimensional cases. (19.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. and so =4 5 and n ez = r = ~ = e^4fir. 19.g. h2 = ikg.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. h = ei^y/k^. Manvelyan.
(19.Nv(w).2.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. VahediFaridi [10]. H. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively. J.e.438 19. (19. (19.6) terms amounting to 1 „ 2 d2Zu A V (19.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0.11) We follow here largely H. W. i. (19. dwA w dw { wA ) (lg. The zerothorder approximation = Zv[w) (19.7. MullerKirsten and N. and H„ (w). where £>„ : = —_ + i _ + J l .2 A / 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.1. H. w 2 .I . Singular Potentials Small h2 solutions in t e r m s of B e s s e l functions We now develop a perturbative procedure" for solving the associated Mathieu equation in the domain around h2 = 0. 8. —Zu — (Z„_i + Zu+i)..7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. Proceeding along the lines of the perturbation method of Sec. (19. — Zv+i). (19.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .5) so that Eq. where these are the Bessel function of the first kind.i l .9) leaves uncompensated on the right hand side of Eq. First we make the additional substitution w = 2/icosh z.2 CHAPTER 19. Aly.Hv (w). Eq. . The solutions Zv(w) are written Jv(w).^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2].
10) in terms of functions G„ defined by Gv+a = —nZ v+a .u (19. (19. where We observe that DUZV = Q. v)Gv + (y._2 + {v.15) (19. . v + 2)Gv+2].6). can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.12) (19. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.2 ) * ( z / .i/) = 2A.u + 4fZu+A}. I / ± 2 ) = 1.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.14) (19. v . The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.20) 2)*ZV_2 + {y. v .17) Thus a term fiGu+a on the right hand side of Eq. v + 2)*Z„ +2 (19.2)*Zi.10) is now particularly simple: i?i°) = h2[(v.v + 2)*{u + 2. it is more convenient to use these relations in order to rewrite Eq.13) (19.2.. z .i/4)*Z 1 / _ 4 +(u. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). .2 ) * i ^ 2 + (!/. (19.13). Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.19.u + 2)*Z„ +2 ].13) therefore lead to the first order contribution jfU = h2[{v. where the starred coefficients are defined by (19.!/+ 2)*4°J 2 ]. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ .16) ( I / . The terms (19. and this means in (19. (19. = §.18) Now </>(°) = Zv left uncompensated Rv . However.21) +(i/. wz The expression (19. of course.10) as a linear combination of various Zv. when a or 2v + a = 0. (i/. v2)*{y2. v + 2)*(v + 2.2)G„_ 2 + (y.
24) P2i2(2j + 2){v + 2j + 2. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j.4 + 58. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . (19. and p 0 ( 2 j ^ 0) = 0. subject to the boundary conditions P2i(2j) = 0 for  j  > i.2 + 2 9 ) ^ 64(^ 2 .v) + hA[{v.17)). i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. P2*o(0) = 0.1169)/i 12 .v)] (19.2.9) 16 + + ^ j " liy '2° This expansion is seen t o b e familiar from t h e theory of periodic Mathieu functions where (Z + l / 2 ) 2 represents the eigenvalue. we obtain 9 h4 = a 2(al) 3 2 (13a .25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv .g.25)/i 8 32(al)3(a4) . (19.v) + {v.l)(z/ 2 .. .4)(»/2 .2)(v + 2j .23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . i/ + 2j). Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .v + 2)*(i/ + 2.j^0 (19. p4(±2) = (i/. . (I/. v ( 4 5 a . (19. j=i.I/±4)*. v + 2j) (19.26) iy — 2. . . Adding these terms and setting the coefficient of Gv equal t o zero. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .440 CHAPTER 19. etc. we obtain t h e expansion . Evaluating t h e first few terms. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. i/)] + • • • . Ri.v2)*{v2. i/ ± 2)* (i/ ± 2. .I/±2)*(I/±2.i/±2)*. . we obtain 0 = h2{v.455a + 1291a . v) +(i/. p 0 (0) = 1. v + 2)*{v + 2. in Eq.22) where p2(±2) p4(±4) = = (i/.v + 2y{v + 2. v — 2)*(v — 2. . i/ ± 2)*.
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 . ±2. £>„ = — . .29) sinh vz or e±uz.cosh 2z)4>. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. The solutions 4> of the modified Mathieu equation are now completely determined.7) into Eq.v + 2)<t>v+2]. (19.2. we can rewrite the latter as j ~ .4). v)(f>v + {v. Thus v is real for small values of  h21. (19.34) . 2 .32) (19.19. v . Thus to O(0) in h we have (jr ' = 4>v — cosh vz. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. Substituting (19.31) It follows that Dv+2n4>v+2n = 0. and is therefore real for both cases g2 > 0 and g2 < 0. (19.h2[(j)u+2 + 4>v_2] h2 [{u. 19.v2(j> = 2/i 2 (A .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.30) £>„&.cosh 2z)4>v = = 2h2A(/>„ . ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. (19. (19.^ + e±(v~2)z. so that J2 (19.i . We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .2)<t>„2 + (v.An(u + n).33) = = cosh(^ + 2)z + cosh(^ — 2)z.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. = 0.
h) = <(>„ +Y.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.h). For rigorous convergence and validity discussions of any of these solutions — our's here differ only in the method of derivation with the perturbation method of Sec. we prefer to adhere to one equation with different solutions.37) where p M (±2) = (I/.2. 19.4 Notation and properties of solutions We now introduce standard notation as in established literature. Arscott [11]. > These solutions correspond in the periodic case respectively to the solutions and me^.h2i *=1 J2 j=i. (19. F. Schafke [193]. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. 8. The solutions in terms of cylindrical functions are written *J. Singular Potentials (I/. W. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. M. Meixner and F. The use of the symbols (v.38) (f> with (\>v = exp(vz) — Meu(z. > (f> with </v = sinh vz — Seu(z. CHAPTER 19.I/) = 2 A . .J7t0 p2i(2j)<f>u+2j. 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.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions.442 where (I/. v ± 2) etc.7 — we refer to Meixner and Schafke [193].I/±2)*.J/±2) = . • (19.h).1 .h). etc. In order to avoid confusion arising from the use of different equations. In particular the following notation is used for the solutions obtained above which we characterize here by their first terms: <f> with (j>v — cosh vz —• Ceu(z. (19.
h) = exp(vz) + Y.e.41) i=l We see immediately that Meu(z + niri. Schafke [193]. . h) = exp(inwn)MP(z. W. (19. (19.40) does not.g.42).h).h). Mi.h). Magnus and F.43) and hence similarly M^\z + inir. h).42) The solutions Meu(z. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. this normalized solution possesses contributions exp(i/z) in higher order terms. 'See e.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). {2h cosh z) • MJ?\z. as Mev{z. Meixner and F. p. as we remarked earlier. 443 (19. (19.h)±Sey{z. ' (z. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. 16.45) T Below we frequently write the normalized solution as in J. Me„(z. 17. W.40) Cev{z.h).(2hcosh z).2 The Potential 1/r4 — Case of Small h? similarly: <fi 4> <f> (p with with with with <j>v = J v (2/icosh z) . h). For the relation of Hankel functions used below.h2) oo = Y.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19.19. whereas the — as yet — unnormalized solution (19. (19.44) implies. <f)v = N„(2hcosh z) MJ?\z. h) = exp{vniri)Me„(z. i. h) are therefore proportional to each other as a comparison of Eqs.MP(z.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. h). i. which. Also from the expansion of Jv{z) in rising powers of z. — oo cv2r{h2)e^+2r>. h) = av{h)M^ (z. see this reference p.44) (19. (19.e. As emphasized in Chapter 8.h) and oo i M^\z. 4>v = Hv (2h cosh 2) * M^\z. h). 4>v = Hi. Oberhettinger [181]. (19.
h2) = ] T cv2r(h2)e{y+2r)z.e. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions.h) . H™(z) = e^HfXz).47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z.43). h) . the first relation is obtained as follows: 2M<£1 = M^l+M^l.48) The series expansions of the associated Mathieu functions M„ (z. p. Mi 3 . 2M™ = eiuvMi3) + e~ivlrM^\ .e.h) Using further properties like (19.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. i. r=—oo (19. . ^ . p. 130) to converge uniformly for all finite complex values of z. MJ>>(0. These points are important in our derivation of the S'matrix below. For later essential requirements (i.exp(TM"r)MW(.g.h2): oo Mev(z. h) = M™(z. oiv{h) = (19.444 CHAPTER 19.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. h). The functions Me±v{z. s (19. As mentioned earlier.49) Inserting this expansion into the modified Mathieu equation (19. (19. we have the following for a change from v to — v.8. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. Meixner and Schafke [193].50) E.h) on the other hand can be shown (cf. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.h) can be shown (cf. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). Meixner and Schafke [193]. Singular Potentials where clearly MeJQ. i.46) H^z) = e^H^(z).e.
^ p 2 2 ^ or c 2r+2 2r2 1 (19.T22r[\ A .51) we have 2r (19.54) 2 C2r 4 ^ h. eg = c~v = 1. Schafke [193]. see J.* For 'Actually. .[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. Thus now c 2r . Meixner and F.(/^.19. (39).^ + 2r + 2 ) 2 ] . Alternatively taking the inverse of Eq. \) /. 122.2 [ A .2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . . 117. p. + 2 r ) 2 ] . (19.53) /i2[A(^ + 2 r ) ] .^ C 2r+2 (19. h~2[\ ./ i 2 C ^ . + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. 2r (19>51) 2r For ease of reading we give the steps in rewriting this.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 .(v + 2r)2} / z . \ . Eq.52) in agreement with Meixner and Schafke [193]. W. p.
+ i).[v^. H. . 128(i/+ l) 2 (i/ + 2 ) ( ! / .^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10)._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(.55 ' Example 19. M.+4)!ll_. Then di 2 1 t" + 2(^1) + • • • . Here we insert for A the expansion (19. MiillerKirsten and N.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. . (19. and also R. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .( " + 2)2] ~ h2 h. Aly.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq. MiillerKirsten. W. Liang and Yunbo Zhang [189].52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. W. t For this purpose *We do this in the manner of the original derivation of R.+0^"> < 19 . Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19.C + l) 3 (i/ . + 0(/. J .l ) 19.Q .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. Manvelyan.1 for the evaluation of a typical case): hb + 0(hw)..27) and truncate the continued fraction after the second step. H.2. J. Spector [257]. H.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy.55). Solution: We put r = 1 in Eq. (19.. but follow our earlier reference H. J. VahediFaridi [10] (this paper contains several misprints which we correct here).446 CHAPTER 19. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .5 D e r i v a t i o n of t h e S .
. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. (19.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo.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 . and then the continuation of this to infinity.19. 22.3a) and (19. we have W r = ip + ut — 7r/4' .57) ^ 1 . this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves).h) 1/2 w) (19. In the case of the repulsive singular potential here. Oberhettinger [181].j7L0 P2i(2j)exp i[v + 2j + 2 2 . Then y reg = = r rV2M®(z. . I / + 2 2 l\TT +E ^ i\ E j=i. Magnus and F. (19. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. p. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const.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. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.59) which tends to zero with r. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. and considering the propagation of this wavefront.56) where by Eqs. W .
The series M^ (z.3a) and (19. Singular Potentials so that when t — oo : r — 0. h) through the entire range of !&{z).h) = exp(Mr^)il4 4 ) (z. This means that the origin of coordinates acts • > as a sink.A14A.3b) ln • 7T Q"*4' r fg ° Vk (19. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo.A1"—2 z '+"2 A~~ n or rA .61) In fact. p. Magnus and F.62) Also with Eq. p.h) can be shown (cf.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i.V + § 9 2 > 4hh* ./i). ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz). Meixner and Schafke [193]. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. .(2j)<gH • (19. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.j¥=o Ki(2j)exp(Tmj) (19. (19. 2h and the square of this expression implies 2 2 ~ ' A. Now from Eqs. (19.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. 17./l ~ 2 ~ 2z + " 4 > 0. '(z.j^Q p2.448 CHAPTER 19. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.60) Using the above asymptotic expressions for the Hankel functions. AV . 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. We require therefore > > the continuation of Mi. 178) to be convergent for  cosh z\ > 1.#. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. Oberhettinger [181]. 4%r g >0 4r^r ' rg See W. '2"4 .
Originally we chose the sign of m / 4 as in Eqs. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. (r — r+)(r + r+)(r — r_)(r + r_) > 0. Thus we choose . is satisfied only for r < r_ < r__ or r_ < r + < r.e. These solutions converge uniformly for all finite complex values of z (cf.1.3b) and (19. r 7r for 0 < r < ro. 19.19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 . p.49). 449 This condition. .r2_) > 0 with r£ = rg(2 ± >/3). A suitable set is the pair of fundamental solutions Me±v defined by (19. which has to be bridged by using another set of solutions. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. and in order to maintain this symmetry. (19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig.62). we have to assign different signs to the imaginary part of z in the two distant regions. i. 19. Since the real part of z changes sign at r = ro. (19. This interchange in the > solution My interchanges y reg with y+. the domain r_ < r < r+ — as illustrated in Fig.63) Thus there is a gap between the two regions of validity — i. (19.3a). Meixner and Schafke [193]. 130).rl)(r2 .e. i Im z z = In r/r0+i7t/4 .1 The domains of solutions.38) or (19.ijt/4 Rez .
65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z. h) = Me^u(—z. p. Thus. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. 131). h) (as one can check or look up in Meixner and Schafke [193]. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. (3 = a . 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. i. . Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}.[ r l/2 M. Singular Potentials Then starting from the region r ~ 0. (19.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+.B^0.l/2Me_i/(T.66) a = (3 .67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_.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_. (19.e. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.^)] dr z=in/4 dr (19. at this junction we require also the imaginary part of z to change sign.450 CHAPTER 19. as just explained.z=—in/4. i.(2)]z=w/4 + 0Me. to r^2[AM^ + BM^\ A. At r = ro the real part of z vanishes and then switches from negative to positive. and we determine these coefficients from this equality and the additional equality obtained from the derivatives. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.v{z)\ .v(z)\ —m/4) + ^[r^Me. Hence we match the right hand sides of Eqs.
19. we use the notation W\f.68) From Eqs. for instance. and the second of Eqs.65) and (19. (19.70) by MeU and subtracting . the nonderivative parts (from the first term on the right of Eq. (19. (19. (3. (19.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. (19. (19. (19. (19. ^ ] = ^ + ^ ^ .3b) we infer that (z = l n r + const.69) Thus if we replace in the derivative relations of Eqs.3a) and (19.71) We now determine the coefficients a.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.64) the relations M&Hz) jz[M^\z)] and from Eqs. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . Then. (19. with the replacements of Eq.68) the derivatives with respect to r by this relation. A and B. Thus we obtain from Eqs. multiplying the first of Eqs. (19. (19.g]:=f(z)*&g(z)WZ) dz dz Thus. = aMev(z) + 0Mev(z).70) by Me'_u (the prime meaning derivative).70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. dr BM^(r)]. (19.2 The Potential 1/r4 — Case of Small h2 451 This solution can be continued into the domain below r = r+ by matching to the right hand side of Eq. and we are left with relations expressible only in terms of z.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.65).69)) cancel out in view of the nonderivative relations. (19.
(19.u. (19.u}W[Me. (19. M^4)] + W[MJ>3).4] = . p.3] = [1.76) .M^] W[Mev.Me^} (19. = IT <*„..MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.(0.73) We now use Eq. Me.77) av I 2ism VK \ ctv av ^Thus. 7T (19.j] given in Meixner and Schafke [193] (p.46)) W[Mev..57) and (19. .74) Moreover. (19.g.M^} W[Me„. W[MJ?\ M^] W[Mev.^ 47 9? MJ>4)] MeV] [3. the leading terms of the respective cylindrical functions. and the second is obtained similarly: W[M?\Mev\ W[Mev.e. 170/171). B = J 2 L _ e .Mev]' P l ' From Eqs. Meixner and Schafke [193].. W[M^\Mlf)]W[Meu. TV W[Mev. we obtain the first of the following two equations.^ ^ a „ Q _ . 7T W[Mev.M^] J™aV.e.Mev] = = = av. (19. (19.58) and considering large values of \z\.46). and Wronskians W[M^\MIJ)] = [i.48) (or cf.Mevy W[M?\Mev] W[Mev. i. Singular Potentials the second resulting equation from the first resulting equation.A] = W[Mi3\z).47) we obtain (av defined by Eq. we use the following circuit relation which can be derived like Eq. h)/M^\o. i. . (19.71) we obtain in a similar way A B = = W\MP\ Me. av{h) = Me.75) With these relations and Eq. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.\ \ ( av 2iwKav ^ v .M^]= IT e~iv*av. h). we obtain immediately W[Z. (19. or obtainable by substituting e. 7T [1.] W[Mev.452 CHAPTER 19. using Eqs. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.4] = .* « * r r z H .
85a) .79) where the second expression follows from Eq./ ( . (19.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). (19. (19. (19.83) The Smatrix element is therefore given by ifc*"* .(l)le^}Pi(cos 9).h) = Me_u(z. Then f' .80) 2^fc (1981) where \7r/ i t sin VK /(*. eii/7r/2 (19.19. f c r . Z ) e ^ ] .f c .h) + BM^\z.(ite™)" 1 { 8 J ' f(k.g. (19. in Chapter 16).46) together with the relation Mev(z.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq./) = (l)1/2R2^2i^)ei^)U^. 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 .h). as we recall from earlier considerations (e.78) we obtain Ay reg « _ .l) ' We note already here that with the substitution R = ei7r7._ [ / ( f c . (19.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#). B=tfD'"". Z ) e .
19. (19.56) to (19. (19.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.58)) in the limit r — 0. We can rewrite Eq. (19.2 The repulsive potential and the various waves. 9?. 19. Fig.2. V(r)=g2/r Fig.Smatrix in the form s i n 7T7 Si = CHAPTER 19. and recalling that yreg is proportional to a function Mc> {z.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. (19. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u).86) . and with the left hand side following from Eqs.56). we can write this (multiplied > by 2ismvTT.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. h) (see Eqs. R 2i sin TXV. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19.58)) 1/2 2r e~ 2hir cosh z. we can define respectively as amplitudes of the incident wave.454 we can rewrite the . cf. R — — = 2i sin ivy. Thus in terms of the (3) variable z.
h) is (cf. (19. Jv+2r.e. J. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. 180). unity minus reflection probability = transmission probability. (19. The expansion is: 4 r ( / i 2 ) M S ( z . The latter is what happens in the S'wave case of the attractive potential.e. i. (p.2COS2VTT = (R.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . (19. H. W.28)) and R = ey real. however not here.h) The function Mo (0. Eq. i.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. M^l)(0.Q.79).84) of the ^matrix. / i ) = Y. S. Liang and Yunbo Zhang [189]. (^)l4lu{h2)Jir(hez)J±u+i+r(hez). unitarity is preserved.±)2 + (2sinj/7r) 2 . A derivation is beyond the scope of our objectives here.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz).** 19. (105) to (108). This implies basically the evaluation of the expression R of Eq. Manvelyan. l=—oo 1 1 lite*"* .In Meixner and Schafke [193] (p.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. Gubser and A.89) . 178). (19.e™*/R\2 ' ' .88) where the factor Me„(0. (19. this expansion is given as 1 oo MJ> Hz. if R becomes a pure phase factor and eil/7r a real exponential. but in the 10dimensional string theory context. i. h) serves the use of the same coefficients as in the other expansions.( i J e ^ ) " 1 ! 2 = R2 + ^ .e. Meixner and Schafke [193] therefore developed an expansion in terms of products of Bessel functions which is more useful in practice owing to its rapid convergence. Eqs. By inserting in Eq. " S e e S.. i.6 E v a l u a t i o n of t h e S . therefore we cite it from Meixner and Schafke [193].2. and then exploit it for the evaluation of the quantity R. Hashimoto [124] and R.19. (19.e. J.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .m a t r i x Our next task is to evaluate the expression (19. MullerKirsten. in which v is complex.
. h2 is complex but v2 is real (cf. Singular Potentials 4r (h )M$(0.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. we obtain M&(0. setting r = 0 in Eq. (19.g. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . Here in the case of the repulsive potential. As a matter of introduction we recall the definition of the coefficients with Eq. (19./>) = Co> 2 ) £ 7 " = — OO j ± > . /=—oo (19. h) in many different ways.6 2 ) m 6 (^_i)3(z. (19. e.3 9 ^ .49). by allocating different values to r.49) from which we obtain for z = 0: Me„(0. u»^. (19.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h). The formula (19.456 so that in particular for z = 0 oo CHAPTER 19.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. (19./i) with the help of Eq. (19. r = 0 and 2. /n ^ .g.94) .h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. Eq.e. 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.28)). i. one obtains *.90) and choose r = 0 (one can choose e.2_4)(l/2_9)iv2y) + <AA. ^ (19.91) Inserting here the coefficients given in (19.90). r = 2 as a check).55). Next we evaluate M^1}(0.
95) These expansions imply for the quantity R: I/! 2v ri1 i .4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .12i/2 + 64u .2^ 2 ± 59^ . + # "+" ( I / . 4(z/ T H ^ 3 .!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy..2 _ ^ 4 ( ^ 2 _ 4)2 (19.1) 2u(4u + 15i/ .1 ) 2 ( I / + 1 ) ( V 2 .111) / / i 5 4 3 1+ 4i/ 2 + +' (z./I) h u .. In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential. _2_(h\2 .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).] 2 \v i + {v2 ..IV 2 2 2(v =F 3i/ .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.4)(i/ .32i/ . From expansion (19.l) 5 (19.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .19.93) and the coefficient expansions (19. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu . one obtains M£!(O.7) F " (i/±l)2(^^l)(i/24)V2. We observe that this approximation involves v and h4 and hence is real. .96) With this explicit expansion we can evaluate the S'matrix.9) V2 1 fh ±v 1+ 2 + • (19.
+ ( i + s w (E 4 . 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3.104) We now consider the case of a — 0 implying that g of Eq.7 CHAPTER 19. and hence allow for the general case of complex parameters v (the Floquet exponent in the case of the periodic Mathieu equation) although here in the case of small \h2\ we know from Eq.^ i 4 2+^ # 1 / •. (19.28) that v is real. and sin 2TT/3 = ft/. (19.102) (19. This is the case of real . Eq. where n is an integer and a and (3 are real and of 0(h4) We have with Eq.eiun/R2)(eiu*n ~ (19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia. (19.99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19.2.28)). In this case R — R* and so 1/i?2 ~ Ofji4). (e iun e~iv*^/R*2)' (19.101) where / is complex and g is real.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.101) is > zero.84): Si ST 1 (11/R2)(11/R*2) . (19. Then cos 2TT/? = 1 . (19.98) (cf.R*2 ^ * #2 \ 11 tf2 R* (19. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).458 19.9 / .103) Then SiSt (1+5) i .
W.3b)) h2 is real for E = k2 > 0. 1 29/ ^ (1 .106) and evidently violates the unitarity of S. H. Liang and Yunbo Zhang [189] (there Eq. (19. J. ( I l l ) ) in agreement with a result in S.577.l ) 2 2 \) +0{h«) (19.e. With t h e help of Eq. Manvelyan. At = 2i[l + 0(h2)}.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.SiSf » 4 sin7rz/ R (19.107b) = Ah2[l + 0(h4)}.19.102) we obtain Sis. MullerKirsten. "This is the result effectively contained in R. We can also evaluate now the amplitudes (19.3. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 . (19.107c) Here h2 is actually Vh4.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. . Here in t h e case of the attractive potential (as remarked after Eq.105) sin7r/3 (19.61n2 . f W .* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . J. 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.61n/i} + ~9~ 0(h8) (19. (19.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].97) this can be written A 4TT2 [v\{v . Oberhettinger [181]. Hence with Eq. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 . Magnus and F.1)\} 2 h \2 iv 1 + Au {y .1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. (19. p. Gubser and A. S.6C . = 1 .Q.^ ) ~ 2 2 459 1+ 9 / • & ( ! .28). i. Hashimoto [124].
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 = 0.108) For the large values of h2 that we wish to consider here. J. N. We are again concerned with the associated Mathieu equation given by Eq. . 2 2 a = [l + 2 (19. S. we again make use of the replacement (17.460 CHAPTER 19.H.$ q(l.26) and hence set a1 = 2hl + 2hq + A{q. 2. This is therefore a highly interesting case which found its complete solution only recently. J.3 The function q(l. i. MiillerKirsten and Jianzu Zhang [226].3. Park. Singular Potentials 19. W.e.4 Fig. Aly.h) (19. Tamaryan. K. completely different solutions.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z .1.h) 10.3. H. where the part of interest here is based on H. 19. (19. VahediFaridi [8] and G. H.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5matrix obtained above by using the asymptotic expansions for large values of h2. Wannier [278]. W. Miiller and N. H.3 19.109) ^We follow here mainly D. h) for large h and upwards I = 0.
(19.63). See also Eq. (17. with an essential mathematical step.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. We begin.3. 19. J4(0)=0. .e.111) we have W[y+.21b). which is even about z = 0. and a is its normalization constant. y'_(0) = 1.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case.e. which therefore cancels out or can be taken to be 1.§ a W[y+.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.110) a where y+(7r) is the solution evaluated at z = TV. The replacement (19. Therefore the second term on the right is nonzero. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.21b). i.y} where VF[y+. y± oc A{z)e2h sm * ± A(z)e2n sm z .l + 2y+(l. y(0) = 0. (17. (17. (19.yJ\ — 1 and cos TTV . We also found the boundary condition (17.19. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.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. (19. 19.3. The behaviour of q as a function of h for some values of / is shown in Fig. to which we have to return in this subsection.113) 'We are considering nonintegral values of v. With the boundary conditions y+(0) = l. we defined the Floquet exponent v and observed with Eq.h*y_(l.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. however. i.h*y (19.
(19. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h.112) from evaluation of Eqs.114) at z = 7r/2.32). .43) we see that z = 7r/2 implies W(TT/2) = 0. B.118a) "We emphasize again: For integral values of v the normalization constants are different. J. which is \/2 at z = 0.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. Miiller [73].462 CHAPTER 19.A(z)e2h sin 2 sin z sin z (19. Dingle and H. i. K [ (z)}2hsinz (19.117) a a Referring back to Eq.e. in the second expression 2h is obtained in addition from differentiation of the exponential factor). (17. 1 1 Or see R. 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 . Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . Hence we require (obtained from Eq. This paper contains the normalization constants of large/i 2 periodic Mathieu functions. (19. (19.114) a a Our first step is therefore the determination of the normalization constants No. Dingle and H. and hence with Eqs. W. (19.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq.46) together with the expansion of the Hermite function from Tables of Special Functions" ) B[w = 0] 7T 4 [lfol)]![l(g+l)]! 2 /i 5 82q2 . See R. We now obtain the expressions needed in Eq. W.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).39 2 14 /i 2 + (19. J. B. (17. (17. Miiller [73].
p.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193]. (19. (19.112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .! ) ] ! [ . . (19.1Q11QM cos irv + 1 ~ •==— e .! ) ] ! [ .62a) and (17.117) are known from the matching of solutions in Eqs. 210.3).87 2 14 /i 2 [ i ( g . (19. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.2g + 3) 27/i ( 3 g 2 . Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.19.\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. Tamaryan.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (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 .[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 .118b) The factors a.119a) This formula — derived and rediscovered in Park.62b) (see also the explicit calculation in Example 17. (17. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .i ( g +3)]! 1 24/i {2hl>2). a in Eq.h .
109) into the equation (19. With the solutions as they stand.109) remaining unchanged. h.h. „ 1 A „ —A cosh z——H . of course. z) = ip(—q. —h. (19. z) can then be written as . —z). h.h) =?==.464 CHAPTER 19.h. the expression (19.3.h.122) Correspondingly the various solutions ip(q.4.e.122) that given one solution ip(q.z)exp[±2hismh z\. h(q.h)iP(q.z) = A(q. 17.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19. we can obtain the linearly independent one either as ip(—q.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. —z).3.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q..124) Since this function differs from a solution ip by a factor k(q. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . (19. .z) = Aq(z)exp[±2hiswh z] ^~°° exp ±z/te ( ')eT^/4> V cosh z e x = Aq(z) exp[±2hi sinh z] ** ~~°° ^ h e ^ V cosh z e ^q/\ (19.h).108) and set ip{q. z) are TP(q. V— 2in (19. .z).h. Vcosh z \ 1 .z) i/>(q. h. —h.2.120) to (19.2 we insert the expression (19. for h — oo). each with a specific asymptotic behaviour.h.123) We again make the important observation by looking at Eqs. In the following we require solutions H e ^ ( z ) . it is still a solution but not with the symmetry property ip(q. We define these solutions in terms of the function Ke(q. h.3 Following the procedure of Sec. z) = ip(q.z). h.? s m h z) Vcosh z (19. dA 1.3. z). Then straightforward integration yields > inq 4 „. . Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.120) The resulting equation for the function A(q. tp(q.z) or as tp(q. h.h. h. i = 1.
. 2j<.3 The Potential 1/r4 — Case of Large h2 465 Instead.45). achieved with the help of the Floquet solutions Me±v(z. i. we h a v e 77 = 0—> 6 = 0. It follows t h a t S e t t i n g s i n h 77 = t a n 0 .<5±1) (19 125)  (the expression on the far right in leading order for h2 large).h2). / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . >• * F $ 0 * e".126) For large values of the argument 2/icosh 2./h). Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. av{h2) = ^(0J ^ Mi i . 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 . we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr.h2) = OvMPfrh2). (z./i 2 ) M (19. We observed that these satisfy the relation (19. (19. (44) of W a n n i e r . as we saw earlier. This is. in performing this cycle of replacements the function picks up a factor. i. the Bessel functions contained in the expansion of the associated Mathieu function M{.19.fi2) can be reexpressed in terms of Hankel functions.e. : 7r/2.127) sin z T /'1 M TT\ — <>{ 4 . (19. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. . w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . Me„(z.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]. (0.e. In order to be able to obtain the Smatrix. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.^ s i n h 7?)]g. 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 . 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].e. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. i.125) is t h u s seen t o b e i d e n t i c a l w i t h t h e p r o p o r t i o n ality factor in E q . [ \ . * ** = > rojKe( ".
h) = e(z) exp ihez .128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs. (19.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.q.q.h) He (z. h. (19. (19. In this way we obtain in the domain .129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H./i). 3te»0.q.466 CHAPTER 19.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T . KeW(z. z): Re(2\z.130a) Re^(z.q.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.vr ./i) Be^(z. 3te»0.h) (3) = = = = Ke(q.q. (2) (19.h).h.h). HeW(z. (19.4 T h e c o n n e c t i o n formulas We now return to Eq.130b).g.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.g. (19.130a).z).q.q. exp[—ikr — in/4\ kr Re{2)(z.3. 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) Be^(z. He (z.q.130b) r_0 rll2exp\g/r (ig) ' 19.eW(z.
q. h) — sin 7T7 He^(z./i).* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. h. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .130b)).li 2 ) Oil. h) = Ke(q. these are exponentially increasing there. z) are proportional there.g. Substituting the Eqs.134) Considering now the first of relations (19.134) in the region where the function He^(z. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.q.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. h. q.131) where the second relation was obtained by changing the sign of v in the first.132) into Eqs. (19. (19. q. and He(*> exponentially decreasing. (19.129)) He (2) (z. t With Eq. — sin 777 H e ( 3 ) ( 2 .125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. z) and He (3) (z.q.h). q. h) = Ke(q./i) — sin 7T7 Re^(z. and thus can be matched in this region.g. h) = sin 7r(7 + v) He (1) (z. we see that the solutions (compare with Eqs. (19.132) These relations are now valid over the entire range of z.g.19. (19./i). (19.(z.h) sin irv = sin7r(7 + ^)He (3) (2.q. Eqs.h2) Me„(z./i 2 ) = Me_j.135) sin irv 'This means.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. we obtain in the domain tfcz <C 0 the relations: Me. / i ) + sin7r(7^)He(4)(2.h2) a.i ^ / 2 H e ( l ) (z> ?) (19. (19. Changing the sign of z.(z.130a).h) is asymptotically small (cf. .133) In a similar way one obtains the connection formulas — sin nv HeW(z.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . (19. g .
133) we can now deduce the 5matrix Si — e2t51.85b) obtained earlier. (19. (19.5 D e r i v a t i o n of t h e ^ .119a) or (19. The quantity 7 is now to be determined from Eq. which in our case is the solution He^4^. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7.m a t r i x From Eq. of course.(19.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. (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. i. Singular Potentials The factor on the left. q. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . (19. and therefore remarks after his Eq.1 .l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19.119b) as „4/i 1 + COS TXV ~ rrl ( g .139) ''Equation (19.468 CHAPTER 19.119b). (19. where 8i is the phase shift. Wannier [278]. Squaring Eq.135). is only the dominant term for large h2} 19. sin TXJ.135).119a) and (19.137) We see that this expression agrees with that of Eq. (60) in the work of G.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr .e'i7ri 1. . H. (19.( .133)) — sin TXV He^4' (z. Wannier did not have the relation (19. (19.l ) (l„—ikr e From this we can deduce that Si (19.Sin TXV. Thus here the S'matrix is defined by (using Eq.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19. The latter is defined by the following larger behaviour of the solution chosen at r = 0.135) is the equation corresponding to Eq.e. cos TXV + sin TXV.
(19. we can approximate the equation for q ~ h (i.139) could actually be neglected) implying  cos m>\ = cosh nvi. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. as a comparison with Eq.19.141) COS TTU irvj — i s i n WR s i n h TTUJ.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19.28) shows.8h COS From Eq. (19. the real part of v must be an integer.e.4 The absorptivity of the S wave (attractive case). .140) From this we obtain the absorptivity A(l.e. This is different from the behaviour for small values of \h2\. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^.137) together with (19.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. imaginary part uj (so that 1 on the left hand side of Eq. h) of the Zth partial wave.§ Since the right hand side grows exponentially with increasing h the Floquet exponent must have a large 1=0 1 large h approximation small h approximation Fig. (19. since cos TVU is large. i. T h i s is r e a l for UR a n integer. 19. i. 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.e. A{l. for the attractive potential. 9/2 (91) with the approximation q ~ h obtainable from Eq.138) we obtain. (19.109).
19. Pao and J. Y. in further contexts beyond those already mentioned. Pope and T. Lii. (19. D. Esposito [88]. N.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. Challifour and R." as well as other aspects. Lii and J. "G. Fubini and G. S .** 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. and G.P. Tiktopoulos and S.107b) and (19. Partial or other aspects of the weak coupling (i. see also M.k) in the approach to unity are too small to become evident here. such as phase shifts. Treiman [271]. This is sketched schematically in Fig. Rosendorf [225].142) we can see the behaviour of the S'wave absorptivity as a function of h. W Some of these sources can be helpful in further investigations. 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. A. Lew [128].470 CHAPTER 19. L. J. VazquezPoritz [61]. H. Cvetic. (19. Dombey and R. Apparently this is one of the very rare cases which permits such complete treatment. C.e. D. Furlan [20]. The diminishing fluctuations of A(l. In essence. N.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . N. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi .4 . Eden [46]. S. . Jones [78]. Bertocchi. Masson [191].139) 2Tr(16h)q /32\ f e Thus with Eqs. F. YuanBen [285]. Shabad [248]. A. Paliov and S. E. Limic [177]. Apart from the papers already referred to. Handelsman. A. H. **A. it seems it is not possible to guess from the above the result for the absorptivity of a singular potential with an arbitrary negative integral power.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. B. H. and. I r a n [61] and M. 11 J. **R. one may expect corresponding results for other singular power potentials. 19. in fact.4 (there are tiny fluctuations in the rapid approach to 1). Cvetic. small h2) case have been considered by some other authors.
1 Introductory Remarks The subject of the large order behaviour of perturbation expansions — meaning the study of the late terms of the asymptotic expansion of some function with a view to extracting information about the exact properties of the function — received wide publicity with the publication of the anharmonic oscillator studies of Bender and Wu. the subject is much older and had been explored earlier in great detail in particular by Dingle^ with the subsequent investigation of the behaviour of the late terms of asymptotic expansions of the eigenvalues of the Mathieu equation and others. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. e.* In fact. B. 471 . the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. f R . Wu [18]. *R. and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. Dingle [71]. the asymptotic solutions in different domains all have the same coefficients. We shall also see explicitly. J. We concentrate in this chapter on the large order behaviour of asymptotic *C. 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. T. B. Bender and T. This is indeed a remarkable connection. W. Muller [74]. also convergent expansions are known and a lot of literature exists on the equation. M. [19]. Dingle and H.Chapter 20 Large Order Behaviour of Perturbation Expansions 20.* However.g.
2 The function ur = r\/pr vs. Presumably this has never been done so far. 5. 20.2.7.8. r for p = 4. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. 20.1 The function ur = pr jr\ vs. This approach leads . Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. 10 12 14 16 18 20 Fig.6.6.1 and cases in Sec. 19.2. 8.2 and below.472 CHAPTER 20. r for p = 4. 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. Fig. Since the method is also applicable to convergent expansions. as evident from Example 17. 8.
this term represents the size of the error. in the power expansion of the exponential function).1 Introductory Remarks 473 automatically to asymptotic series. 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.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. 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.J2 2n h E (20.2) or. The behaviour of such terms is illustrated in Figs. The formal perturbation expansion of a quantity E(g ) . the function E(g2) is given only approximately by Yln=o En92n.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. The behaviour of the rth term (r large) of the convergent expansion is generally of the form of a power divided by a factorial (as. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. even if each En is known.20. and the integral E(g2) is called the Borel sum. i. and the expansion is said to be Borel summable.1 and 20. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH).2.3) n=0 Thus for any given value of g . for instance. For an authoritative discussion of these aspects we refer to Dingle [70]. i.e. equivalently for g = l/h. (20. T V lim h2N 2 n EQ?) . as we discussed earlier in Chapter 8. In some cases the answer is affirmative. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. 20.e. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. The traditional method of using an asymptotic expansion is to truncate the series at the least term. It can be seen from these. .
474 CHAPTER 20. (and E0 = 0). El . Thus if CO F(z) = aiS(z) + J2 anzn. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]).1)? Asymptotic expansions originate from integrating a series (i.4). .n _ En+\ n! En+1 ^ (!)"«!. . In this ideal case the integral of Eq. (20. n=0 (20.5) then with the integral representation of the gamma function F(z) = (z — 1)!.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. In general this is not so easy. In any case. In fact. . . The function F = 1/(1 + z) is an analytic function which can be continued beyond this circle to the entire positive zaxis as required by Eq. n=0 (20. (20. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V .4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.6)). . . (20. Large Order Behaviour of Perturbation Expansions Eq.e.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. (20. 1 we obtain oo F(z)£(*)" = — . Q. 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. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: .
. or can we determine Ar independently? Fig. ? When ?s E ^ 0./ \p\dr\.. 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. M. . Popov and V.1 Introductory Remarks 475 except for possible subtractions. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. Weinberg [228]. N = l. e. The potential (20. S. $s E can be calculated by the WKB procedure.. and E represents an eigenenergy.2.9) The question is therefore: How can we determine 9 E(h).g.e.2..20. we have a quantum mechanical problem where probability can leak away. 20.N = l. (20.3 and g < 0.. if the potential is V(r) = +grN. i. for < ^ 0.10) is considered in detail in this work. (20.. This is the case if the real potential has a hump of some kind. 20. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).10) r as shown in Fig.11) V..3 The potential V(r) gr»... § where 7(ff) = To exp  . (20.
Popov and Weinberg. Popov and V. 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 .76) (see also the comment on the comparison with the Kepler problem before Eq. the cosine potential being t h e most completely investigated will also be seen in the following. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. . as we expect on the basis of Eq. (14.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. An interesting application is provided by Example 20. (11. the BreitWigner formula (10. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. However. Weinberg [83].1.1: Application to the Yukawa Potential Evaluate 9 £ = —7/2 for the Yukawa potential V(r) = g— r / \ r and insert the result into Eq. S.73) in the W K B or semiclassical approximation.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. 8. of course. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. (20. 20. (20.11) is. Here n is the principal quantum number of the Coulomb problem (see Eq.476 CHAPTER 20. Example 20. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. Equation (20. (11. L. the perturbation method of Sec. M.9) for Ar.^ The logarithmic factor in Er is a novel feature of this case. as was (and still is) not the case for t h e equation 1 V .134)).11). Solution: The solution can be found in a paper of Eletsky. 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. Eletsky.113)). Thus in the following we apply these methods to our typical examples. V.
This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. Dingle and H. (20.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.j + 1). . (17. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.j) are known.13) represents effectively a partial difference equation. 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. (cf.l ) +2[(q + Aj)2 + l + A]Pl.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .Cq.58) and (17.55)) X = E(q. i. B. J. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const. (17.3 ) P i _ 1 ( g .14) Thus these coefficients can be obtained once the coefficients Pi(q. Eqs.2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential. j .2 A = Mx + ~M2 + 7^WM3 + •• • .34)) that the coefficients Pi(q.l(q. Miiller [74].M2i+1 for large i.60).13) In the special case j = i the boundary conditions stated after Eq. or equivalently of the quantity A.J) = (g + 4 j .j) a n d M2i can be deduced.20.Bq. obey the recurrence relation jPi(q. (17.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.l ) ( g + 4 j .12) (20.26) and (17. (20. (17.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf.h) = 2h2 + 2hq+j. The recurrence relation (20.j) of functions ipq+4j. Eqs. h). W. 3)? . M2i = 2j2 J[Pi(q. Eqs. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. are given by i . (17.e. % = Aq.
and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .15b) The zth term in the expansion of the eigenvalues A of the Schrodinger equation with cosine potential (Mathieu equation) is therefore given by ith term i\ (8h) il 1 2g2 + 3 2i )[{[\{q 7.*. (20.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.478 CHAPTER 20.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods. 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. see also below) .16a) (20.91 l)]![i(g3)]!} 2 { [ .l (20. Hence exact Borel summation is not possible. Using this relation and the duplication formula lw('\W±^.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula. also written lCj.18. .•91 {[l(? + i)]![(? + 3)]!}2 (20.
2 . in the calculation of the level splitting as a result of tunneling. . . t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. Previously.e.4 The cosine potential shifted by TT/2.20.* However. n = 0 . this difference was found to be real. i. J. we follow here a formulation in the context of our earlier considerations of the Mathieu equation. The potential we consider here then has the shape shown in Fig.5 \ \ / \A Fig.4. but with different boundary conditions. Stone and J. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. . f H .3 Cosine Potential: Complex Eigenvalues 479 20. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. l . we emphasize a few points *M. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. which means that the potential is there approximated by an infinitely high harmonic oscillator well with levels enumerated by the quantum number qo or (equivalently) n. 20. Reeve [262]. MiillerKirsten [200].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.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. 20.3 20.3. the difference q — q$. remains unchanged. . W. At the risk of repeating what we explained earlier. We retain the previous boundary condition at one minimum. . when this is g0 = 2n + l . A means to achieve this was found by Stone and Reeve.2 solutions obtained earlier. \ 1 \ \ \ \ \ 0.
480 CHAPTER 20. 20. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). Setting w = 2h1/2z. Here we choose t h e second way.T K ^ a D_i_{q+1){±iw). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . 20.21) <u> + ( i « . (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. Now we change this situation at the minimum on the right. 20. The physical wave functions there are therefore the solutions which are real and exponentially decreasing.4.4. Thus one chooses not the solution with real argument at the minimum.2h2 cos 2z]ip{z) = 0. as we found these for this potential.e. and allows its imaginary part to tend to infinity.4 (around z — 0) to the one on the right (around z ~ ir). a boundary condition which at the face looks somewhat abstract. we shift the argument by n/2. 8 the equation can be approximated by d2ip(w) (20. but achieves exactly what we want. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. which means replacing z by z±ir/2. We can distort the potential there or alternatively impose a different behaviour there on the wave function. but t h a t with complex argument there.Ah2z2 • • • }^{z) = 0. Previously this situation was the same at the neighbouring minimum. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . (20. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. . and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. Effectively we consider the tunneling from the left well in Fig. i. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them.20a) As stated. X + 2h2 = 2hq+—.
22). 17. however. that at z = 7r.e. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. . . we naturally constructed the even and odd solutions with these (cf.e. (20. Eq. Thus. .> 0 for z > ±ioo. such that this deviation becomes imaginary. 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. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. whereas here we start off with a solution even or odd about a minimum. whereas here we construct these from oscillatortype solutions. A derivation is given subsequently. A (which we can loosely dub solutions of WKBtype).63). by a different method. 3 .23) . 2.24) Considering t m O i n Eq. Here — again as before — we take the finite height of the barrier to the right of z = 0 into account by taking q only approximately equal to go = 2n + 1. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. Since the large/i2 solutions valid away from the minima were the solutions of types A. . 20. 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. i. (20. .20.3 Cosine Potential: Complex Eigenvalues 481 As explained above. we see that ips ~ cos \j2qhz for z ~ 0. 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. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g.4. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. We consider here explicitly only the case of the ground state around z = 0.
Setting v = 0 in T(—v) = (—v— 1)! gives infinity. Oberhettinger [181].30) 'See e.•w) + 2ir T{u) <"+!)% Dvi(iw). or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. The arguments of the functions appearing in Eq.28) Extracting Du{—w) we have A. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.g. §See e. (20.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing. Magnus and F.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. tp(v).27) differ by an angle IT. p. 17. (20.h(2h1/2 J £2=0 (20. Oberhettinger [181]. However.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{. we know from the solutions of types B and C in Sec.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/.g.3. this can be handled in a few lines.27) requires the differentiation of this expression with respect to the index v. . p. From Tables of Special Functions'. 2. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq.29) Equation (20. (20. Thus the only nonvanishing contribution comes from differentiation of Y[—v). (20.482 CHAPTER 20.25) qo) E2 0: (20. W.w) = e T Dv{w) 2TT n») AvViDv^iw). We see this as follows.(. 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. W. and then evaluation at u = 0. 92.24) are linearly independent around z = 0 for q ^ qo. Magnus and F.
pp. 92. .29). 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. and "Prom W..32) which we can use here since all other problems have been resolved: d D {w) dv v (20. r(i/) We now return to Eq. which are valid around z ~ 7r/2.20. Thus calculations for qo ^ 1 are not given by M. dropping from the derivative part of Eq. (20. .34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. 5 7 7 . Oberhettinger [181]. including the derivative of DE2ufl(2h1/2z)).+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. .31). . We now go to the WKBtype solutions. For u — 0 one obtains the result (20. Oberhettinger [181].e. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. (20. (20. those of type A. W.//n and ij>{y) = C .4.29) v=0 (in + In w)D0(w) + v2ire (m + lnu.27) exponentially decreasing contributions. Magnus and F.g.33) = IV2~TT.)e" w /4 2 l ^Dx(iw) (20. i. Differentiation of this equation and subsequently setting v — 0 yields. p. Magnus and F.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .31) dT(u) r*(v) dv 1 i/>(v) ^ o l. i/^0 483 T(v) It follows that d dv [T(u)\ (20.e. "See e. 3 .3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1. Stone and J. This result is » not easily generalized to <jo > 1. Reeve [262]. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e.
34). y COS c\ J (20.37) Comparing Eqs.cos z — 2/icos z ipA^ Tcos  and i n A —. cos z/2 ~ (20.41) 2TT .e. Eqs. Then V'WKB becomes (with cos z ~ — 1 + z' /2.35) We construct with constants c\. where we can then match these to the solution ips.484 CHAPTER 20. Thus the WKBlike solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^Tsinf 2/i hz2 _z_ 2 (20.36) and consider this at the other end of its domain of validity.29) and (17. (17. i.e.39) Now. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .e. To this end we set z' = z — ix. by there replacing z by z + ir/2. (20. (20. (20.34) and (20. 2 c2e2hehz UJ2 . Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.36) cos  . i.40). (20. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. smf (20.. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). We now return to the WKBlike solution (20.38) in its limiting domain towards 2 = 0. (20.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h.— 721 . we obtain the required solutions from the earlier ones. x .sm I We now want to match this solution to that of Eq. Thus in the present case and with q ~ qo = 1 we obtain „2ft. Recalling that we have shifted the potential by 7r/2. near z = n the boundary condition (20.36) V'WKB cie~2hehz' 9h  .40) From Magnus and Oberhettinger [181] (p. We observe that in the matching domain (i. that in approaching z = ir.
38) can be written TAWKB (20.2/. We derive the general result in the next subsection using our own method. 1 2 4 1 2 8h ^^2 2 (20. ^'mliw <2a47) For qo = 1 we recover the result (20. i.46) Repeating these calculations for the general case (i.43) we obtain.46).39) and (20. (20.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. J Hence Eq. (20.45) Comparing this result with tps of Eq.43) Comparing now Eqs. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial. (20. . (20. (20.1/V)_le^_ Z With this result Eq. The general result stated by Stone and Reeve [262] is therefore presumably guessed.e.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.20.e. and hence (adding and subtracting equal terms) Di(i2hV2z') = ^[D^h^z') + D_1{i2h1/2z')} Z + hD_1(i2hl''1z') = \V2^D0(2h^2z') Z D_l{i2hll2z1)] D^(i2hl/2z')} + hDiWh^z1) + [D^i(i2hl/2z') Z ^D0(2hl/2z') Z . (20.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(.34) we obtain E2 = i%\ = ±ih22Qe~8h.^.44) = DQ{2h / z) ± (2h7r) / e. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . with c\ = exp(—2h) from above.
. W.e. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there.20a). Considering the original equation. (20. i. H.*.()#0. We have to impose two sets of boundary conditions. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .3.5.46) we consider a half period of the cosine potential as illustrated in Fig. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. and expanding cos 2z around the point z = — TT/2. i.E curre it \ j. 20. V(z) . Fig.e. i.5. (20. MiillerKirsten and A.2 CHAPTER 20. J. •+(=)* V(f) 0.486 20.5 The cosine potential from — TT/2 to n/2. Eq. At the minimum to the right of the barrier shown in Fig. 20. Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e 7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P.z. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. Achuthan. (20. •_(:)*<>.e. at z = TT/2.) + 7T V> = o.Ah2 ( z + . Wiedemann [4]. this equation may be approximated there as d2ip ~dz^ + E + 2h2 . 20. the result is formula (334) there.48) These conditions provide the required quantum number qo = 2n + 1.
e.21) into the Mathieu equation we obtain. (20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.50a) V'AfeM) = where (the second expression is needed for later comparison) Aq(z) cos§(9i)(iz + lyr) + W) [ t a n (±z + W)]«/ 2 [sin(i* + ±TT) C O S ( ^ + ±vr)] V' ' sinfr+Vaz (20.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 . (20.49) 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.z)=tpA(q. three pairs of solutions of types named A.r/2 ~ e i(2hq) V 2 . (20.3 Cosine Potential: Complex Eigenvalues 487 (using Eq.21)). For the continuation we use again the WKB formalism. (20.h.z). for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. i. we demand that ^(*)L<.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. Proceeding as in our previous cases and substituting the expression (20. B and C. Recapitulating these from Chapter 17.20.h. 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. as shown earlier.51b) 2i(«1)[i(9l)]'. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.
48) and proceeding as in Chapter 17. ^ W H .^ i . we obtain 1>±W = \ a a (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. (20. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. \vw). (20. Here the upper sign refers to the even states and the lower to the odd states.52b) The first solution is valid asymptotically around z = —vr/2.56) Applying the boundary conditions (20.488 CHAPTER 20.e. (20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) .57) where go = 1) 3.55) Extending these solutions to the domain around the minimum at z = 7r/2.52a) = Cq[w(z)}.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}.=F2 2\1/2{28h2y/4e~Ah l + O (20. We return to this condition after derivation of the . and the second around z = 7r/2.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. 5. we obtain our first transcendental equation.qo . These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2. the relation q . i. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.
Ah2 cos2 z ip = 0. D. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.[ [Ah cos z . (20.2hq] / dz) We know that the solutions I/JA{Z). .Ah2 cos2 z ~ 0.61) [Ah2 cos2 z2hq]1/2dz. F. p. Thereafter we extend the latter across the turning point down to periodic WKB solutions. (20. formula 280.67).3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum.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 . as we shall see with the result of Eq. I^A(Z) overlap parts of the domains of validity of these solutions. Our procedure now is to match these to exponential WKB solutions.49) and we can apply our boundary condition.58) Ignoring the correction A / 8 . Jz+ ' T In principle the integral could be evaluated exactly — see P. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. Byrd and M. we see that the turning points at z± are given by 2hq . The solutions of type A are valid to the right as well as to the left of the maximum of the barrier.01. Some nontrivial manipulations are again required to uncover the generation of important factorials. and we obtain the proportionality by appropriate expansion. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .20.e. These then provide complex exponentials like (20. i. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . Friedman [40]. e x p ( . 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.
490 CHAPTER 20. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(.sin2 z = l .cos2 ( z+^) = (z+ . 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.63) We handle the logarithmic terms in Eq.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20. (20. We have for the following expression contained in (20. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20. using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).64) .59) 2h COS Z+  ~ 0.62): 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. we write cos2 z = l . (20.) +0 z+ IT Inserting this into the integral I.62) as follows.» _ i y v . Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z.
. (20. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan ./ * [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. (4^2)i/4 [ i( g _3)] ! V/ l (20.63) and (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.50a) and (20.60) by (4h2 cos2 z)1!4).3 Cosine Potential: Complex Eigenvalues Inserting the relations (20.62).^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . (20.20.66) The WKB solution therefore becomes (approximating the denominator of (20. .68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated.50b).64) into Eq.+ qj V2 4 A. on using in the second step Eqs. We ignore this here.
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. (20.55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). i 27r(8^/ 4 2/t /4 ' r(±) . Large Order Behaviour of Perturbation Expansions again using the Stirling formula.492 CHAPTER 20.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .z)}]. S1H (2/ KZ ) / (z f 2 (2^).U ^K^.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20. io(8/*)g/V^i6ft2)V4[±(g3)]! . we obtain T _ ( ± ) = 0.1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T ..67) and (20.50)). we obtain l(8/i)g/ 4 e.l^l !T ^^* T2 7 r ( ^^ /. (14.49). We now return to the even and odd wave functions defined by Eq.T 7 77 r )V2 /4(2 .69) l(8M9/4e2^ \2 [ i ( g . i.49). (14. where _ 1 + [±) *) + £ (20.+) + B^) HAh^/\ir.i ( 2 / i g ) 1 / 2 ( z + . Eqs.™) . 1 2 1 ±2 . Thus with IPA(Z) and TPA(Z) taken from Eqs.?/ 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.KW4^8 4 . (20.2 [ ? ) Imposing the boundary condition (20.
57) by the left hand side of Eq.3.75) .h)+2h(qq0) (16h)i°+le'Sh E(qQ. 20.21) expanded about q = qo. (20.74) This result agrees with that of Stone and Reeve [262].57) and (20.72) is our second transcendental equation. In both cases we wrote (cf.26). the result (20. we obtain E(q. in fact.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .72) We observe that with the square root as on the right hand side both sides are complex.47).qo) El Ah' (20.h) + i l + O '4{[^ol)]!}2 E(q0. for q an integer both sides are imaginary.h) = = E(q0. (20. Equation (20. (20. (17. Ex = 2h2 + 2hq0 + A 1 (<? .73) Inserting this expression into A = E(q.h) + (qq0)( BE — 90 E{qQ.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.e.72).25)) E = A = Ei + E2. We now have the two equations which have to be satisfied.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. Combining these equations by replacing (28/i2)<?/4 in Eq. Eqs.3 ) I = }(»« !7rl/22I(. Eqs. h) of Eq. (20.3 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. who derived it for qo = 1 and guessed the form for general values of qo.20. (20.h)+i%Ego. i. (20. (20.
cf. A = .494 CHAPTER 20. The same formula can be derived in the case of the double well potential.4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. the discontinuity across its cut from zero to infinity. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo.2 below.+  .* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. Achuthan. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. Setting ng AEqo:=2iSfEqo. Eq. i. (17. 25Eqo=8h[ 2\1/2 . Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions.W [£(901)]! < (20.74b). (20. f E . We had A= or _2ft2 + 2ft. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. Bogomol'nyi and V. by others. . W.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. 20. its tunneling properties and the large order behaviour.) {I6h)™/2e4h 1 + . B. A. J. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.I g  ± l ( 2 „.80) *P.9) the coefficients Ai of its perturbation expansion. (20.e. Wiedemann [4] and Example 20. we can recalculate with the help of Eq. H. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. 7 8 ) i=l V .3. MiillerKirsten and A.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. Fateyev [26].
For convenience we write Eq.19) for the large values of i under consideration here. a = J + \<1> (2084) .l ) ( < ? + 4 j . the integration is seen to give (20. f H . i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. and the method can be applied in many other cases. Dingle [74]. But as Dingle remarks. *R.l ) +2[(q + 4j)2 + l + A}Pi_1(q.81) We observe that this result agrees with the result of Eq.20.4 Cosine Potential: A Different Calculation A further very instructive method of obtaining the large order behaviour of a perturbation expansion was found by Dingle* in application to the case of the cosine potential or Mathieu equation. (20.3 ) P i _ 1 ( c ? . B.14). (20.83) One now sets ftH = ( 2 % P * ( 9 . W. j . The integral is recognized to be of the type of the integral representation of the gamma function. (20. J).j) + {q + 4j + l)(q + 4j + 3)Pi1(q. Using again n = (go . MullerKirsten [200]. 20. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results.13) out again: jPiiQj) = (g + 4 j . (20. so that an exact Borel summation is not possible.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. J.j) satisfy the recurrence relation (20..j + 1). there is nothing special about this case. (20.74).t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . We also observe that these terms do not alternate in sign.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.1).
(q + 4j) 2 ~ 4j(4j + 2g).a R . a. Whittaker functions WKjfl. 4j(4j + 2g + 4) + / .h) +p(a + l.87) Taking terms of 0(h~l+1) of this equation.1) + 2 P i _i(a) + Pi_i(a + 1). (20.z)WaA/2(z) Setting V. 0 4j(4j + 2g) pi_i(a . 1  1 / 2 (4 (20. we regain Eq. (20. (20.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1. z2 (20. e. the recurrence relation (20.l.83) can be written j . A. Taking dominant terms.4. Abramowitz and I.^(Z). which are solutions of the differential equation known as Whittaker's equation.90) See M.1) + 2 (a — 1) a .h) = ^2pi(a) with p.496 CHAPTER 20. d2WKh dz2 + 1 4 K z zM2n WKtll = 0.l/2l Wa.86) one can convince oneself that Eq.h). (20. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.(a) oc .K + ^ ] L + K^)WK.31. p(a. (20.ll{z)= L . Stegun [1].g. we obtain (2a . Pi(a) i.e. Now.h) 2+ 8h a p(a.1/2(z) (a1)! (20. nx pi_i(a + l).e. i.89) Replacing here K by a and \x by 1/2. formula 13.91) + Wa+lil/2(z) = (1 . .27h pi(a) a = 4j(4j + 2 g .85). (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .85) Considering now the generating function of the perturbation coefficients. g < 4 j .4 ) ^ . i. (20.e.85) implies p(a .
Magnus and F .(« . (20. 89.84)) {8h)q/2( Ah for j < 0. Dingle discovered the solutions^ p(a. .92) with Eq.. (20.i/2(z). v .86) have to be chosen such that p(a.j — —j.\)V . a (20.e.94) Hence to insure that p(a. (20.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. g(h)(20. h —> —h. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. such that p(a.(« 1) 2 ]. In the present case this means the replacements q — —q.(« .h) = \~\Pi(&) with pi oc h l. Oberhettinger [181]. we have to remove a factor (recalling a = j + (7/2 of Eq.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y.93) (a1)! ' where f(h).88) is known.h) ~/(fe) ' (a1)! .88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential. h) is expressed in descending powers of z = 8/i.g(h) are functions of h which in view of Eq. h) is expressed in descending powers of h. > (20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).87).92) Comparing Eq. (20. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f . > > ^See W. ^Observe that if one solution of Whittaker's equation (20.i/2(z) + Va+lA/2(z) = 497 2+ Va.20.4 Cosine Potential: Another Method of Calculation we obtain Vai. p. i.
(20. h\ oc (8h)q/2e~4h ( _ ..1. B.8 / t ) for j > 0.94) here we have a series in descending powers of h. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2. n = 0. This expression can be obtained from the recurrence relation (20.95a) Inserting (20.498 CHAPTER 20. Comparing > > > this result now with Eq. ^ .94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. (20.. we obtain the result of Dingle: (20. _ i s. Dingle and H.95b) Then for j > 0 we obtain from Eq.).l ) ] ! 11 R. J. (20. Muller [74].84) we conclude that the original coefficients Piiflij) a r e f° r large i given by the following expression — apart from an as yet undetermined proportionality factor cq — We determine cq by imposing as boundary condition the expression for Pi(q.+ i f or 3 < 0 (20.j _ g / 2  1 / 2 ( . g — —q.96) and a similar expression for j < 0 with /i — —/i. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. W. j — —j.83) and is given by 1 1 . i). 2.. Correspondingly we set for j > 0 / 1 A pla = j + q.1/2(8^) ( . . x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .
(20.98) 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.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.e.+i?_l]!}2 (20.0) = 1 as desired (although not really enforcible for large values of i).l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20. formula 0.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. M.3)]!2 2 '+^.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. (20. i.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . . Ryzhik [122]. 5. l{Q. B. W. E* J'=I i+M 2 [2i + 9 . (20.14). of Eq.97d) Then the coefficient M<n of the eigenvalue expansion.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20. p. Dingle and H. J. Gradshteyn and I.100) **R. Thus ° 9 (i + \q) 2«(i+Igl)! p . is M2l = 2j2j[Pi(<l.20. Miiller [74].97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. S. note the misprint there: (r — s) ought to be (r — s)\. formula (30).157(4).l ] ! {[. A special case of the formula can be found in I.l ] ! (20.97c) We assume that in a leading approximation we can cancel the factors (i + \q).{j + \<l).j) (i + lg)[2i + i ( g .4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q.e.
i.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.]\.i ) ] ! [ ^ +  ( g .1.. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20.e. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. i.. since \{q — l) + \{q — 3) = \q.81) for q — qo = 2 n + l .e.101) This formula allows us to cancel all factorials [i.o i • . i. n = 0. .17).. (18.5 20..l. [^ + i ( g .86). (18. 20. the result of Bender and Wu [18]. i.3 ) ] ! _ 1 Hence M or.102) Thus the result is M m 1 2«Fi + g .105 ) the latter because we see from the eigenvalue expansion (18.4)) £ = ^ = ^+ w and y= ^ > ( 20 . Eq.e. i.2. i\ M i large 1 .34) that £ is an expansion in ascending powers of h 6 /c 2 . The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity. (18. [19].' 1 i • (20.l l ! with the large i behaviour in agreement with the result (20.Am i ..5.e.86) and set (cf. We refer back to Eq.500 CHAPTER 20. With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator. Eq.
(20. in terms of K where go = 2ftT + 1. Once this has been obtained one can insert the imaginary part into the Borel transform and obtain the behaviour of the late terms in the eigenvalue expansion. (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.20. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . We leave this to the following Example.107) we obtain (i)r r 71" JO yirZ£{y')dy'.l ) r .e. The result establishes the series unequivocally as an asymptotic series.e.110) This is the result in agreement with the large order formula (4. the discontinuity across its cut.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. in fact as one with Borel summability in view of the factor ( .4) of Bender and Wu [19].5. i.(20. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20. 20.106) With £= E r=0 Mr (20. . (18.108) Proceeding in this way. and hence inserting here the imaginary part from Eq.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.109) or.86). and then the relation (20. i. This can then be compared with the level splitting we obtained for this potential.77) can be verified.
The relation (20. Its usefulness has also been demonstrated there. 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. M. R. Large Order Behaviour of Perturbation Expansions Example 20. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. of International Workshop on Perturbation Theory at Large Order (Florida. Silverstone [52].77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. Grecchi. G. It was later explicitly established for the level splittings and imaginary parts of arbitrary states in the cases of the cosine potential and the double well potential. Cizek. Graffi. Bender and T. J.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. Damburg. t p . Grecchi. Wiedemann [4]. Harrell.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . J. K. Gram.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. . R. Harrell. Cizek. Propin and H. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). Silverstone [62]. J. 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. Wu [18]. J. MiillerKirsten [164]. Achuthan. S. W. S. Paldus and H. in t h a t when the quantity on one side is known. H. E. Achuthan. M.502 CHAPTER 20. J. J. Eqs. 1982) [139]. MiillerKirsten and A. K. J. Liang and H. J. [19] can be traced back from articles in the Proc. MiillerKirsten and A. *R. (357) to (371).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. W. J. Harris. M. J . Propin. E. J. Damburg. V. V. Paldus. T. Wiedemann [4]. as in later chapters. .77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. H. S.Q .^ 20. it can be used to obtain the other. A relation similar t o (20. t The relation (20. " P . *Some of the literature in this field following the work of C. R. Nakai. J. Solution: For the solution we refer to the literature. W.
Feynman [94]. Coulomb problem) we refer to existing literature.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. A brief introduction is also contained in R. Hibbs [95]. Schulman [245]. This formalism is particularly useful for evaluating scattering or transition amplitudes. R. The formalism is not so useful for boundstate problems (in this case the initial wave function is.e. Felsager [91]. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. P. P. *R.Chapter 21 The P a t h Integral Formalism 21. For the application of path integrals to the hydrogen atom (i. which explains also difficulties. Probably the most readable introduction. S. P. unknown). Feynman and A. A further wellknown reference is the book of L. since in such cases the probability of this system choosing a path far away may be considered to be small. Kleinert [151]. however. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. 'See in particular H. Feynman [93]. in general. Chapters 2 and 5. can be found in the book of B. However.* 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. 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. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. We illustrate the method here by rederiving the Rutherford scattering formula with this method. The path integral is. 503 . 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. However. A standard reference is the book of R.
2) where K is a Green's function. (21. Obviously K(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). x. be ip{x.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.tf.ti).tf\x. *In Dirac's notation we write this equation later (xf.ti\tp).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.t).x).3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.tf\ip) . we can compute the action S1.x.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).504 CHAPTER 21. */ = U + (n + l)e. U.1) Let the wave function i/> at time t = ti.e—»0 Sn. we wish to know* ip(xf. ti) = 5(xf . the initial time.e.ti)ip(x. (21.5) Sn = X / fc=l mo (zfe . i. =J dx(xf. The Feynman Path Integral Formalism 21. ti).e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. (21.tf) = / dxK(xf.ti){x. (21. 2mo dx2 (21. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.
1 A path in (x. The usefulness of the method therefore depends on the fact that some paths are more probable than others. as will be shown to be possible below. 21. of which one is illustrated in Fig.tf).28)): f dr\ exp ia a) ' (21. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation. 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.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. 21. the Fresnel integral (13.t). Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. Here and in the following we need one or the other of the following integrals (cf. The multiple integral involves individual integrals like f dxk exp Xk)2 + {xk Zfci) 2 } with Xk contained in the step up to this point and in the step away from it.21.ti) to its final position at (xf.6a) .1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig.
ie f me\ 2a\a~) 1/2 (21. 2 + 2a. dxneiSn/n.5)): G(xf.8) Note t h a t in spite of the n integrations.Zfci) 2 } I +{xk+i x*.506 CHAPTER 21. dx exp z—{a. We then integrate and finally let /x go to zero.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT..xk_i) dx exp i — { 2 a . One therefore removes the troublesome factor by defining (Sn being the sum in Eq.ti) '•= f g m ) (N(€\]n+i dxL. n > 0. the power n + 1 in the denominator has t o be seen in conjunction with the n + 1 terms in the sum of Sn in "Thus.6b) —c ia o drjr] e x p—V € (21.s f c _i) + (xk+i .7) Thus we expect t h a t — in view of the multiplicative y/e here in front of the phase factor — in the limit e — 0 > oo />oo /•oo / dxi oo / J—oo dx2.(x fc+ i . (21.0 ^7 (x f c + i . one has j ^ dr1eifa+i^ /£ = .mQf Xk L ~2e^ ^ . for example. with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.. (21. tf > U.xk+x oo —oo oo .xk+i) exp .2 + (x + x f c + 1 . The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.Xi. J—oo dxneiS/h — • 0.37)) dxk exp d(xk .tf.t ~Xk+l> + xk+i 2 ) +{xk .x f c _ i ) 5 \m0 J (21..z f c _i) 2 } } .m0 .6c) In the evaluation of these integrals we replace a by a + ifi.
B. These integrations are indicated by the horizontal arrows in Fig.26)) that — ignoring the potential — the multiple integral (21. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).ti) For tf > ti we now write* G(xf.tf. (21.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. which all vary from —oo to oo. § See e. (21. p.ti) for tf > U.Xi.e.t In the following we demonstrate that the function constructed in this way is. (21. B. however prefixed with a new factor.2 The Path Integral and Green's Functions 507 Eq.5). (21. However.21. it may be noted that with a different philosophy concerning the summation over paths.Xi. we arrive at a path integral without the measure factor N(e) above.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. The consideration of the denumerable number of possible paths connecting the particle's initial position with its final position may be handled in a number of different ways. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. (21.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims. 183. 183.7) for n — 1) to give a time interval.g. Felsager [91]. (21. as in our discretization in Eq.g.tf. nor the limit of integration of some variable Xi.U) = / JXi=x(ti) = K(xf. See the square root in front of the exponential in Eq. in fact. i.e.tf. It is neither a variable of integration in T){x} —> Ylndxn. We shall see later (with Eq.Xi. i. and will therefore not be attempted here. 21. Felsager [91]. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.24). p. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. .5). and so on. the Green's function K of Eq. G(xf. (21. Different constructions of these families of paths usually lead to different measures.2). summation over all possible paths.1. 11 For discussions see e.
U) imo 2 G(xn+i W)Idr.xi.74)).ti).11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.3 in the evaluation of I a specific path integral (see Eqs. the other with that of the action of the free particle SQ. Consider a time t = tf + e for position Xf = xn+2.1 and allowing n to go to infinity. we obtain the same as before except that there is no factor N(e).Xi. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. we have G(xn+2. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals. 21. 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. Expanding this in a Taylor series around x. We have G(xn+2. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{.xi. (23. we have See.ti. however. Each of these linear pieces is completely characterized by its endpoints. I We employ this method later in Example 23. Then the measure drops out.188. .tf + e.508 CHAPTER 21. Then we have on the right the factor G{x+rj).Xi. so that the piecewise linear paths fill the entire space of paths. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. one with the full action S.ti) = 1 j^p: f°° cte n +iexp } • fm0.Then we can construct a recurrence relation as follows. (21.tf + e.72) to (23. It is for these reasons that — unless required — the measure is frequently ignored. the discussion in B. pp.tf. 187 . Felsager [91]. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.exp 2e n ieV(xn+2) G(xn+2 + V.
tf.tf.tf. It follows that ip(xf.Xi. tf.xi. we obtain: G(x. U) .tf. tf + e.tf.— p G ( j . U) + .xi.ie 1 d2 2nV0dx^G{X.tf''Xi.Xi. Xi.2 The Path Integral and Green's Functions so that G(x.tf + e.Xi.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.U) +••• \ 509 G(x.(21.6a).Xi.6b) and (21.21.ti) = j^rr dnexp I ^rj1 . .tf + e.ieV(x) + ri< —G(x.ti) N(e) or G(x.xi.16) .tf.G(x.15) We see that G(x.Xi.xi.ti)\ .Xi.t Thus + 0(e2 (21.xi.xi.ti) = G(x.tf.ti) 2 —l€ — ^2 + V(x)G(x.13) We integrate with the help of the integrals (21. (21.U).ti) = 1 d2 + V(x) G(x. U) = e^G + 0(e 2 ) dtf 1 d2G(x. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.ti) +V(x)G(x.6c).U + 0(e ).Xi. tf.tf + e.x.tf. tf.ti)ip(x. ti) satisfies the timedependent Schrodinger equation.ti) +\if(^G{x.ti) .Xi.xi.tf. d and this means d i—G(x.ti) for tf > ti (21.tf. Retaining only terms up to those linear in e.tf)= dxG(xf.Xi.tf.14) = G(x.tf. 2mo dx2 (21.xi.
Xi.tf. Eq. now called So. denoted by KQ.U.16). X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. K0 = 9(tf .tf.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.ti) = = 9{tfti)—G + G5(tfti) .10) is indeed to be identified with the timedependent Green's function.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.510 CHAPTER 21.U).tf.17) Differentiating K with respect to tf we obtain —K(xf. as may be verified by inserting this into Eq. 21.tf.15) is a wave function (cf.U) = G(xf.ti) = 5(xj — x).U) = / T){x} eiS\ [fdt \mx2. In a problem where V is small it is convenient to calculate K with a perturbation method.3. (21.U)6(tf . (21. Eq. (21.U)G0.1)). Hence our original function K is the same as G.x.8)) Go(xf.20) . as this is also called. 21. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation.x. and the Green's function KQ\ S0= Here (cf.x. (21. The result. We have therefore established that the Feynman integral (21. is the Fourier transform of the nonrelativistic propagator. The function G which satisfies the timedependent Schrodinger equation (21. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. (21. or K(xf.x.
oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.t) K0(x.23) for Go and obtain Go(xf.Xi.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function. (21.25) T h e quantity Ko(x.3 The Green's Function for Potential and hence Go{xf.ti)9(t). We obtained the same expression previously — see Eq.Xi.37) — by explicit solution of the differential equation which the Green's function satisfies.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.tf.tf. exp 2 (21.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.z / c .22) • \ n/2 .ti)= iX (o./ . 2t Go(xf. . f c . oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.tf.ti) = = G0(x.24) ( s / .ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a. (7.21) fe=l and t = (n + l)e = tf — U and h = 1.21. In Example 21.x 0 )" n + 1 (in agreement with Eq.Xi.. n + i .ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.ti) Ko(xf. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.i ) 2 . But t = tf — ti = (n + l)e.Xi.23) We now insert our expression for In into Eq.Xi. Then Go(xf. we have Go(xf.Xi. (21.8). (21..tf.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.tf. exp where Xj = xo and Xf = xn+\ and there k = 1).tf. (21.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.1 we show t h a t oo /.
10) in a product form. the latter being a Green's function with interaction like that for the harmonic oscillator.6a).t0) (21.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .f_^y e. (21. (21.)+(1"+2.x „ + i ) 2 = ( x n + 2 Xi + Xi .5) is carried along.512 CHAPTER 21.22).25) we can rewrite the Feynman path integral or kernel (21.tn+l•xn.x „ + i ) 2 = (x„ + 2 .. ^ (xi .2y(xn+2 . (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .1: Evaluation of a path integral Verify Eq.xo)2 + (x2 .x ^ 2 + y2 .22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. In our proof of induction we assume the result is correct for n.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._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.54)..26) with n integrations and n + l factors KQ. 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.e. (21. (21. Solution: The result can be obtained by cumbersome integration.tn)••• K0(xi. For n = 1 (see below) h This result can be verified like Eq. If the potential V(x) of Eq. we have instead of 5*0 and KQ expressions S and K.)!} J — OO . ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l. (7. as / JXi f r D{x{t)}eiS°lh . or else by induction. Then oo 2 / (21.xi)2 = 21 xi ^2o(„. i.7) by direct integration with ^ .ti. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x0. Example 21. Eq.
iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .. ) H — (xn+2Xi) n+2 n+l n+2 y2 . A(n + 2) 21. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . and + 2 2 l n + 2 ( x n + 2 . (21.29) Hence with KQ from Eq.3.i) H —3/ . n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .n+2a. The second integral is an integral representation of the step function. (21.27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq.Q .21.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. The first.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.a .27) we have (with a = t/2m0) .28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.Xi) + ( x n + 2 Xi)2.e.x2rriQ 1/2 toij_. (n+l)A n n j(a.2.25): Ko{xJ) = = (S) e{t)(^X'2 y J . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. a dp exp [ipx + iap2] = exp — i(21. n + 2 .2 y ( a . (21. This quantity is also described as the nonrelativistic free particle propagator.i j ) H n +— z 1 .6a). We pass over to momentum space with the help of the following two integrals. (21. it .3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj).2y(xn+2 .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.i) . Using Eq. 21. i. e > 0 .
The Feynman Path Integral Formalism Im t \t>0 Rex Fig.*' where 2mo — r. an expansion in rising powers of the coupling constant (contained in the potential). (21.Q (21. The generalization to three space dimensions (required below in Eq. Replacing 9(t) by its integral representation (21. the second order two factors of V and so on. ( 3 b) ° The expression E+ V — it 2m.e.514 CHAPTER 21.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. Thus in first .28) this becomes K0(x. The first order correction contains one factor of the potential V.51)) is fairly selfevident.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. This is the type of expansion most people describe as perturbation theory.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion.2 Contours of integration. i. 21. 21.
^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.tf. \2meJ n+l J.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/. .i. .^){iV(a. (21.Xi. Eqs._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 .32) Since — we recall this for convenience G(xf.33) xG0(xi..xfc_ OO J— OO J—OO L J. x (n+l)/2 \ lim .t/. dxn exp i ^ —.ti. (21..)} (21. li f f 515 order (cf.4 Including V to First Order: Perturbation Expansion h=l.20)) and with elS = exp i S i o / hi dtV(x.(xfc .21.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x. } e °° f l f •/ oo /"oo F ( a .t) JSo dtV(x.i V e / D { a . .+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.tf.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2. dxr. . (21.5) a n d (21..U).t)] n+l ..t)\.Xi. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.^.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.xi.i.^ J_00 V(xk.t) the first order correction is (G\ ~ G — Go) Gi(xf.
It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.x2.t) =li dtV(x. Jti Consider the term of second order. (21.t) (21. in writing down the expansion exp i dtV(x.x. dx2Ko(xf.ti).34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). i. Xi.t2. (21.36) {if •j. OO /"OO / oo dt / J—oo dxKo(xf.I dtV(x.tf.xi. t) Ju Thus the term containing n times V contains a factor 1/n!. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.37) Jti The last two contributions do not describe the time evolution from ti to tf.35) We also observe here that the problem of timeordering arises only in perturbation theory.xi. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.tf.t2)V(x2.t) ! Jti r*f dt'V(x.t)K0(x. i. t2. The factors of "V" originate from the series expansion of the exponential exp ftf . (21. This is cancelled by n! different time orderings.t2) xKo(x2.e.ti) t2)K0(x2. / 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'.t2)V(x2.e.t)V(x.516 CHAPTER 21.tf. .U) xV(xi.i / dtV(x.xi. t') + • • • .x2. = i dt2 / dx2Ki(xf.ti)K0(x.t.ti). They must therefore be deleted.
U) (21.tf.t) is V(p.Xi.ti). of course. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj.3 Feynman rules and representation of the Green's function. The Fourier transform of V(x.ti) (21. or in momentum space as mostly in field theory. x"^ X x"2.Xi.t)Ko(x. Such diagrams in the context of field theory are called Feynman diagrams.21.t) ) K K K0 'V X K o Fig.t.Xi.3 along with the diagramatic representation of Eq. For various reasons — such as illustration. E) which is given by (in onetime plus onespace dimensions) V(x.ti) i dt = K0(xf.t)V(x. These may be formulated in configuration space as below. .E).ti). (21.t)K0(x. Here.39).t)V(x. (21.x.39) dxK(xf.x.t)= f dpdEeipxiEtV(p.tf. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. tf. called Feynman rules.t2 K o (x 2'l2 .Xi. 21.xi. The diagrams represent mathematical quantities and hence are designed on the basis of rules.tf.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.tf.40) The rules for the present case are given in Fig. we are concerned with quantum mechanics.i ] P I dt J dxKn(xf.Xi.ti) = K0(xf.tf.38) .t. x r l i 'V(x. 21.
i. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions. for the allowed values of k: kx = 2imx/L. and dn = dnxdnydnz. mo V since ip is normalized to one particle in volume V. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. We recall for convenience from **Note that from the wave function we obtain. The plane wave function normalized to one over a box of volume V = L3 is. There the wave functions ipi. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.ti)il)i(xi.t) = .518 CHAPTER 21. T = duration of time for which potential is switched on. * k 1 • . .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. to one particle in all of space.e.ipf have to be properly normalized.Xi. The Feynman Path Integral Formalism 21. (21. ^(x. dn = (L/27r) 3 dk.tj)K{xf. with p = Kk. here with E — k 2 /2mo.k = 2wn/L. (3) The transition probability per second is = \A\2 .5 Rederivation of the Rutherford Formula As an application of the path integral method we now consider scattering of a particle off a Coulomb potential.ti).tf.= l. i.7 = e i k ' x _ i : l i with f i/}*i/jdx. 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). (21.41) We have to pass over to three space dimensions.e. (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.
# = ^an(i)VnM) n = a^(t) + \a^(t) + . P(*) = j r (2145) (cf.t) = e~^lT\ r where r = Ixl.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential. V(x. «= — . Eq. and with these the following perturbation ansatz: ih^* with an(t) = HV.. The transition probability per unit time is .k. (10. in which T is a large but finite time.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.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.e. H = H0 + \H'. u = ^\H'kfp(k). (21.— .= t7Ei4 1} wi 2 k It is frequently assumed that the only time dependence of H'(t) is that it is switsched on at t = 0 and switched off at time t > 0.4).117) and Example 10.44) = u(Ek. Then H'ki can be taken out of the integrand in Eq.21. 47T 1 3 7 (21.. i. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'.44) and f_oo can be replaced by fQ. Proceeding in this manner one obtains Fermi's "golden rule". (21. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.47) .
tf. t. (21. dt dxip*f(xf. t)K0(x. The threedimensional generalization is (taking tf » T and ti <C —T. ti) = co5(t — U)5(x — Xj) etc. We can relate this amplitude to the amplitude ark of Eq.t) = ^ J*ie*<*'*><<*'*>. The lowest order approximation of A. In our previous onedimensional considerations we had with Eq. Xj. we have to replace the onedimensional quantities dxf etc.* / . also called Born approximation. (21.49) i.44) (here in the scattering problem with initial and final energies equal).52) .tf)K0(xf. i. t.x. x.yLi. T a large but finite time. (x*.44) by contracting the propagators to instantaneous point interactions.t)V(x. (21.39) OO /"00 / dt / OO J —OO dxK0(xf. (21.ti).tf)ipi(xf. by threedimensional dxf and so on. Thus with the ansatz Ko(x.tf.t.f.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.Xi.ti). (21.tf)ip*(xf.50) Here x — xj — Xi and t = tf — ti.520 CHAPTER 21.i x (i)V(x..t) (21. (21. (21. £. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.t)K0(x. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.e.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P . (21.tf)Ki(x.51) Inserting this into Eq.tf.tf).f.ti)'ipi(xi.48) where from Eq.>/>.tf.xuti) = ^ J°° d p e i p x _ i ^ * .tfx. Ai = dxf / dx.).30a) K0(xf. This now has the form of the amplitude (21..^ / . we obtain A1 = clS(ti tf) / dxf(i)V(xf.x. so that 0(t) can be dropped) K0(xf.
5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. this becomes A(p.k. k2 ipx + i—t + ikx — i 1 2mo 2mo (21.59) . £).fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.k.i ( k 2 .T) = JdtJdx(^\V(x. 2 „ x v~ m o (21.k. (21. Then we require the following integral.).p ) .58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand. At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 . Then integrating over q and q' we obtain . (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 .~ .T) = x exp L z/ q/2 k2 521 f dx.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 .56) (21.k. which is evaluated in Example 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.54) Now inserting the expression for the potential V(x.21. (21.p 2.57) we have A(p. .iaT2/8: L With 0 0 • .r) = = y .t)exp A(p. P2 .T) = ^JdXJdt%*t2/T2exp i ( k .
„. This expression assumes that the integration over time t is normalized to 1. e (21.\ 2?7lo / 1 (21. In our case this has not been done. otot contains the factor \A(p.51 — 2 /w2 y ~ —. dE = PjV = ^ E ± .T)2/T. (21. e 8{k) = lim 7=*—^ (21. k2 . 66) o . { I i _ F T l } ." + M ] p) ' "•"• ^2 . mo m (21 . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k .64) Inserting the result now into our expression for the total cross section. / / P 2 _ . write down the probability per unit time which is ( 8 \1'2 _a2 16^ . For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . .e.63) We establish this result for T — 1/e in Example 21. T ) ~ .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. Using Eq.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.P ) 2 + M2]2 4 VvrT2.522 CHAPTER 21. Using the result (21. <rtot.59). k .„ 2. y 2 [ ( k .< ^ £ \M VTTT2.62) We want to consider the limit T — oo. we obtain .59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .a ( — ) ^ . where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).3.60) Now. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21.63) we have: Probability per unit time = 77777. ( 21 .T2f 8 V/ 2 .^ . i. (21.k.65) But E=*. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.
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. 2mo / dft. 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 .* 9 ' d r e . so t h a t (21.cos 9) = 4k 2 sin 2  .4.70) . Then (x = r) foo „2jj. This result may also be obtained from lyukawa of Eq. (21.68) Inserting (21.65) with the cutoff M > 0.^ .67) T h e delta function here implies energy conservation. (21.21.j . /Yukawa = ) Solution: We choose the 2axis along the vector q.3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a*o 2K Joo for a > 0 and hence the following representation eikx = lim — .67) and (21. and inserting the expression thus obtained into the final result of Example 10.59) by replacing there q 2 by ( k — p ) 2 .p ) 2 = 2 k 2 ( l . —Mr 4w = . (21.^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.M £ M * = / m0kdn = m0^2m0E (k .69) This is the wellknown Rutherford formula.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .e . Example 21.e. giving the potential a coupling constant. setting M2 = 0. esfn (21. i. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.e.= e " f c a ^ o 2v^fa /4a = lim £^o —. p 2 = k 2 .68) into Eq.^ .2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e ..j .
(yx) = 6(y . Since dxe" 1 1 1 = « / . We have x> = I J ^ l p X p l x ) . In configuration space the state of the particle is written x). In the present case. ' i a = lim e —0 pk 2 /e2 e^TT 21.. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /.q). Va we obtain g(k) = Je~k V a = 0.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . The two representations are related to one another as we saw in Chapter 4 by the following expressions or Fourier integrals: p> =  d x  x ) ( x  p ) .524 The function required is CHAPTER 21.71) 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 .* .x) (21.72) co From (21. Differentiation of (21. Direct integration /4a . We consider again nonrelativistic quantum mechanics.71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. a.70) and (21.75) .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. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. '<*> = . (21.oo J / )eikx. (21.^0 a^C 2^/TTO.71) with respect to k yields the same as °° 2 . however. 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). a simpler method suffices. (21.
. 159 .x.t> =iH\x.77) (21.i) = K(x'. . i.t)} constitute a "moving reference frame" in the sense that ip{x. (21.0) = e""x).We now introduce the Dirac bra and ket formalism. i.82): The general Schrodinger equation has the state vector to the right. (21.We define x.^ l x ) . i. (with ft = 1) d/dt\ip) = —iH\tp). at initial time ti to its final position xy at time tf. i.i) = e iH4 x. H.e. We consider first an infinitesimal transition.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). x.e.76) These relations imply the completeness relations 1 = J dxx><x.79) Feynman's principle is expressed in terms of the propagator or transition function (x'.t'x. we must have d/dt{x. i. t) = (x. so thattt d — x. 4) is to satisfy the Schrodinger equation. See also L. (21. pp.£) = ( x V ^ ' .t). so that \if>t=o)s = H>)H. Ryder [241]. Hence for the infinitesimal transition (x'.t) = e* H t x). Then the states {x.t). Thus (qPp) = p(qp) = p(27r)3<5(q .t'. The time development is given by the Hamilton operator H.t\.t). x' — x infinitesimal. M^ip(x.80) (21. Similarly we have a position operator X with Xx) = xx).t). d/dt\x. (21. (px) = e~ipx. as had to be shown. The equation 1 = J ^p)(p. t\ipo)s. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.t) = —iHip(x.e.21.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. t' — t. which describes the propagation of the particle from its initial position x.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc ..81) Comments on signs in Eq. Then ( X  ^ .t\ = iH{x.i'x.82) (21. (21.e.160. ) H = (x</>o)s.If '/'(x.e.t) = +iH\x.83) (21.p).
i. 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.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).88) Thus P has the operator representation P= /dxx)V(x.* .V 2 < 5 ( x .x') + V{x)5(x' .P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«. 2mo Since (21.x = V(p .89) Using this and previous relations we obtain for the Hamiltonian (21.91) .85) the amplitude or expectation value or position space representation (x'#x) = . (21.90) (P'IVIP) = y<&. (2187) (21.p'). "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.i p '.526 CHAPTER 21.85) The momentum operator P satisfies the relation P = /(2^3PP)(P' (2L86) so that it projects out the momentum q of a momentum state q).e. Pq) = J 7^3Pp)(pq) = J dppp>*(p . (21.x y(x)e i p .y<&(p'ix')(x'ivix)<Xp) = f dxe..x).
j.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 '.92) We now return to the expression (21.fx. We now consider the transition of the particle from Xj. But e'ipx + 0[{t .92).p') + V(p .82)).e.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 .t) (cf. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.84).p') • • • j W < 5 ( p ' .6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .94) In the above we have chosen the signs such that p = —iV and H = i— ot (21. i. the signs are reversed when these operators are applied to the ket vector \x.21. (x'.p) + i{t . 527 (21.95) when applied to the bra vector (x. ti to x j .tj+i\x. for the amplitude of an infinitesimal transition.i'x. tf subdivided into infinitesimal transitions (xj+i. (21.tj). (21.t') \^(27rf5(p 2mo +V(pp' where we used (21.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21. t). EQ — Hi . We have <x'. i.p').
x i ) ( t i . .+ V(x f c ). d x n .i/xj.94) and thus obtain a phase space expression.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) ..tj) = / ••• / d x „ _ i d x n _ 2 .99) .tfei) fe=i .tf\Xi. t n  x n _ i .tixo.97) H hH(pn. x i ..x i ) H h pn • ( x n . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. d x i . T h e integration over the n—1 intermediate positions.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . d x i ( x n . zmo T h e n t h e integration over all p ^ can be performed by using Eq.528 CHAPTER 21.X n _i)] h) (21.. We now insert for each of the infinitesimal intermediate transitions the amplitude (21. . (21.ti) rx/=x(i/) d x i dx2 .. . t n _ i ) t„2><xi.tf\Xi. .. . x n which the particle passes at times tQ. . (Xf.27) in which we make the replacements x —> eifc. i. .xk (21.. x f c ) = ^ .e.tn respectively. d x „ _ i . = 1 X X<jPfc We now replace H(pfc.a — —e/2mo and divide b o t h > sides by 2TT.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.i { f l " ( p i .dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi . then corresponds to the sum over all paths. i.xic) by fc — Xfej H(pk.ti.t 0 ) + H(p2.2.x 0 ) + p 2 • (x 2 . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.. ..t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . x exp [ .xn)(tn in_i)]. dx.e.98) tk — *fcl # ( p f c .
6 Path Integrals in Dirac's Notation 529 Forming products of expressions of this type we see t h a t this relation can be generalized to the threedimensional case. (21. (21. (21.8) as N(e) so t h a t the integral does not vanish for e — 0. G.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk . Smolyanov.98).101) Then Eq.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.100) over V{p}= L J ^ rwooll V (27T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O.e. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. / {xf.e.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. Tokarev and A.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21. G.] ex H ^I (.21. A.. (21. i.H(x. to / m0 \ 3 / 2 las. > Thus the phase space representation of the pathintegral* (21.103) is another form of the pathintegral for which the integration with the help of Eq. . i.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.tf\xi.p)]\. Truman [254].
t'}.e. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. pk) = L(xk.ti) where nl D{x(£)}exp '. t to x. .104) where for the particularly simple Lagrangian L under consideration here m0±l .tf\xi.8). £ J ) with no operator in between. £ /  X J . Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation. this agrees with the corresponding discrepancy in the onedimensional case of Eq.8).x) .x) (21.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. (21. Garrod [106]. (21.V(xk) = m0±l .105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. x.tf\xi.530 CHAPTER 21. We have now obtained the path integral representation of the matrix element ( X / .iJ(x fc . 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'. pfc) = Pfc • xfc . p . t. nl (xf.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). (21. I hi dtL(x. Thus finally we have Feynman's principle: (Xf.H(xfe. operator (21. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. i. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C. xfc).
(px).76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ).0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.t"\x5){x5\x'./ ^ I < « .*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.t"\x\x'. . x fdxG  ^ P <p"x3)<x3e^V) x (21.109) . .t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".i p '• x .107). With the wave functions (21.5 l^ t 'lp / > e .* * f dp = j (2^ p) ^ d .77) and thus have 531 *!*') = *M = * 7 ^ I P > « . .t'}. Thus <xV'xx'.21.v ' x ' and thus {x". so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.* * . (21.6 Path Integrals in Dirac's Notation For the verification we use Eq. (21.
Inserting here the integral representation (21.iit!' 1') 2mo + vtf) + <5(x"x'). (21.90) we can rewrite this expression as (x". We now consider the transition of the expectation value of operator x from Xj.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .ti = to as before.0 dp x„ / (27T): 1 . . (21.t') (xV'xx'.. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.• dx1 (x n . £ n _ixx n _ 2 . Thus with x^ = xn. tn2) • • • (xi. (21.£'> = x"[«5(x" . We now proceed from there along parallel lines.t n _i) x ( x n _ i .tj+i\x\xj.532 CHAPTER 21.t"\x\x'.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') . ] .tf subdivided into n transitions (xj+i. x / / e .£j) = / ••• / dx n _idx n _2 .*/<x"x5)(x5tfx'> + . (21.i t " ( x "  #  x 5 ) ( x V ) +. This is the expression corresponding to that of Eq. ( x " .x ' ) (2^ i^+vM + D «p(x"x') (21.i{t" .t') dp .£/xxj.94) for the case with no operator in between.111) 1 .£ n xx n _i. we obtain (here and in the following the limit n —> oo at the end being understood) (x/.i ( t " .£j to Xf.f ) i f e i p .t'){x"\H\x') = X + •••} (21.74) of the delta function.x') . we obtain with (21.x') .85) (xVxx'.113) Inserting for each of the infinitesimal intermediate transitions the amplitude . tixx 0 .110) Recalling Eq.i(t" . .tj) at equal time intervals e. t0).tf — tn and x$ = XQ.112) where x" is a cnumber. = x"6{x" .
{x".t') .21. (21. .. we obtain* (xf..2 .« • > We reconsider therefore the transition amplitude (x".t"\x'.2 .xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. .t') = (x" + 5x".tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.114).. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.7 Canonical Quantization from Path Integrals (21. (21.t') = and con 5x"Vx„{x".t') (21. x (21...115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .x 1 )(t 1 Z & ) +H(p2.89) is consistent with the conjugate expression dL (2L116 * . dyi l xi x 2 .x 0 ) + p 2 • (x 2 . the string of factors xi X2 . dxnn _i Xi X2 .y (27F (27F •" (2^(2^)3 exp[.x2)(t2 ~ h) + • • • + H(p ••• + p „ .iyxxj.x .t"\x'. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .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. (21.t') sider the variation S(x". tj \ xt. x n _ i x n reduces to the single factor x^.( x „ .t"\x'.114) x exp [i{pi • (xi . .tf\x. U).105).t"\x'.\Xi.10) or (21.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq.x n _ i ) ] .3 " n d x f c .. Xn _i X„ xi dx.t"\x'.{^( P i.112).. tj  xj1 x».. For simplicity we consider the case of one spatial dimension. n—>oo *•»• 21. We demonstrate first that the operator representation of P given by Eq.ti) 533 rxf=x(tf) = / • • • / dXidx. tj) = xf (x/.and hence may be put in front of the integrals to yield (xf.
" M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. " (t). dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. (21.t').118) — = (21 5x"(x". The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.tt).t') =  6— / 1 V°{x}t .122a) = 17) 5x"Vx/. Thus in this case of classical mechanics dL_ d fdV.t'): 1 fx" 6(x".t"\x'. and we demand the validity of the equation of motion. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21.i/(x)/fi ) j ^ ) dtL(x.x). Then. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. For the purpose of enabling some formulations below.t i_(dL_ x't' h\dx J t„ (21. x) Jt Jt dL.(x".118) JI(x)/h . dx dt \ dx _ (21. t".l&cdi.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. <9L' dt —dx + — 5 x ox ox f . Hence w « = (t). with N = N(t" . . jt/ 37 ( £T. dL r /•*" .534 CHAPTER 21. (21. we separate the metric factors from T){x} in Eq. (21.t"\x'.10) and set D{x} = D°{x}/JV(i/ .
125c) Corresponding to our relation (21. 535 (21. / Jxi V°{x} / Jx' V°{x}.t'). S.1>) = ^{x". (21.115) in three space dimensions. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.125b) and the measure property (3) N~\A + A') = A^. . (21.126) R. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. x' (21.t')(x'.21. Streater and A. 6 (21..125a) (2) the completeness relation holds. I(x) = / JA dtL(x. F.122b) By comparison with Eq.1 (A)iV.7 Canonical Quantization from Path Integrals or JL(x»y\x. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti).t"\p(t")\x'.Xf = x(tf).124) where with A = tf — ti.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization.tf\xi. ^2\x'. § The evolution of the system is expressed by the Feynman functional integral (xf. (21.1 (A / ). Wightman [264]. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21.x).t'\ = l.
tf\x(ti)\xi.tf\T(xix2)\xi.t').t"\X'. Then the relation follows with the help of the properties (1).tf\xi\x'.tf\T(Xl • • • xn)\xi.ti) = Xi(xf.. (21. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.. we have the relation (xf.128) where tf is later than ti and t\. (21.ti) = x' 'Y^{xf. Thus if T denotes such ordering..127) We can now write down an expression for a timeordered product of operators. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.t\x'.t>).129) = ^ y With Eq.t'){x\t'\x2\xi.tf\xi. We have (cf. (21.t"\x'.t') = ih(x.ti) f ' V0{x}xlX2eu^lh. Eq. (21. (21. (2) and (3) above. (21. .t"\x"\x'.t\x'.127)) (x". We consider the case n = 2.131) = ih(x".133) = ih{x". tn lie correspondingly in consecutive order between U and tf. (21..ti).t') + x"^J^{x".536 CHAPTER 21.t') \ i ox" i ox"J or in commutator form [P.130) ~i5x7jX"{x".t'). Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.ti) = —!— p V0{x}xx • • • xneu^h.t"\x'.t')=x"(x".t') (21. Thus (xf.t"\x'.t'). (21.122b) we can deduce the nontrivial canonical commutation relation arising here.x](x.t"\x'. We can rewrite this expression as the commutator relation (ITU*" x"^](x".t"\x'.114) (xf.132) The generalization to higher spatial dimensions proceeds along parallel lines.
but asymptotic. An essential singularity is really an ambiguity rather than an infinity.e. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. Watson [283]. T.g. z = a + i/3 with a — oo or f3 —> oo or both.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. i. if a function is expanded in the neighbourhood of an essential singularity* or if an expansion of the integrand of an integral representation is used (i. E. Such expansions are strictly speaking. it has some finite. Whittaker and G. W = e x p ( l / z ) . In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. for sufficiently large r.e. in the sense of rigorous convergence tests. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. divergent. p. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. see e.e. nonzero radius of convergence in the domain of the coupling parameter. Such an expansion parameter is usually known to be or assumed to be small in some sense. Expansions of this type arise.e.g. Consider e. N. In physical applications such cases are rare. such as the fine structure constant of quantum electrodynamics. integrated over) beyond its radius of convergence. in the context of quantum mechanics are not convergent. the series converges. so that successive contributions to the first approximation do not invalidate this approximation. Besides. although the first few terms indicate a convergent behaviour. It was therefore natural to search for alternative methods of expansion. as we have seen in Chapter 8.Chapter 22 Classical Field Configurations 22. The vast majority of perturbation expansions discussed e. If the expansion is mathematically well defined.g. 102. i. > 537 .g. i.
and in general this change is accompanied by certain constraints. A method which achieves this is in particular Feynman's path integral procedure. In the following we consider various typical examples.) l/h2. 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. and has led to insights which previously seemed unimaginable.g. higher order corrections are of orders ±.h2. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. A further challenging task was the development of methods of quantization of theories which incorporate classical. i. 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.. such > a series begins with a contribution proportional to (e. and the corresponding expansion is called the "loop expansion". The classical.h°. cnumber. such as soliton theory.h. 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.e.538 CHAPTER 22. In searching for other means of expansion. The fundamental extension of the method of canonical quantization (i. although this terminology is not precise.e. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". One therefore faces the problem of quantizing a system which is subject to constraints.e. These methods are often described as methods of collective coordinates. i. cnumber first approximations. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. Many of these aspects are interesting by themselves ...h3. However. 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).. the latter's topological properties if any. This type of expansion is decsribed as "semiclassicar. There. On the contrary. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). The study of expansions of this type has turned out to be extremely fruitful. 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.
Thus references to field theory will be exceptional and hopefully do not irritate the reader. 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. monopoles. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity.1) ' T h e reader who wants a highly advanced overview of instantons. d^] = d^d^ . to keep this case in mind. or even a solution to a modified field equation (e.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. may like to consult the article of D.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.g.2 The Constant Classical Field 539 and in any case deserve detailed study.V(4>). 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). This means theories with field densities which are therefore infinitedimensional. vortices and kinks.22. . The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". Tong [272]. However. modified by going to imaginary time). a static (time independent but spacedependent) solution of the classical equations of motion. We write the complex scalar field <f)(x) in four spacetime dimensions The spacetime Lagrangian density of the specific theory we consider here is given by C[4>. We shall not be concerned so much with the case of a constant first approximation. 22. V{</>) = m§0V + A0(<^)2. We begin therefore with a brief recapitulation of the simple Higgs model which exemplifies this case and exhibits the phenomenon known as "spontaneous symmetry violation" or the Goldstone phenomenon. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). (22. it is important for a better understanding of the considerations which follow.
1 Different potentials for different signs of m§. Classical Field Configurations where d^ — (—d/d(ct). 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). 22. .S£/<90(x. p. B.540 CHAPTER 22. £)) is bounded from below we must have Ao > 0. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage.3) See e. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0.e. of the equation of motion? For these we must have all derivatives of <j) zero. V[<t>] m 2>o o Fig. i. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. m l + \04>*4> )$ = <). and hence here [d^ + m20}(l) = Xo(f (22. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. V ) . 22.1.2) We ask: Are there classical (i. The EulerLagrange equation is d„ dC d(d. Felsager [91].e.£).g. 433. § (22. cnumber) solutions <j)(x) = 4>c = const. as illustrated in Fig.
the only such solution is the case we wish to consider.e. w A iP 3ml 5* "XT" (22. TTIQ > 0. and so ft[<M=m6:(^* 1 + Ao(0*0)2. the equation has this invariance. aq> For (b = const.5) .</>2)plane. H = —C = V.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig.22.2 The Constant Classical Field For Ao > 0. We can convince ourselves that <fic is the field associated with a state of minimized energy. with 0 = 0. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected.2 The spontaneously chosen phase. The Hamiltonian density H is defined by the Legendre transform H[<f>. (22. 22.e. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. 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. TT = —J. i. mg < 0. not the solution x(t)). But for Ao > 0.ir]=7r<fiC. 22.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. we have 541 0. Fig. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion.
^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . V?) (22. i.3) obtained above.3).g. (22. This is the condition (22.e.542 CHAPTER 22.7b) In Eq. the one for (3 = 0. Suppose we choose one such configuration. and we wish to investigate the behaviour of the field cf> in its neighbourhood. we set cf>(x) = <f)c + ri(x).e. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi.e. i.3ml = 2ml = ml + TA 0 A 2 = 0. >° for m 0 < °> . Then we can reach some point in the neighbourhood of 4>c{(3 = 0) by travelling from 4>c partly along the direction of minimum configurations. along the trough of V (which we can call a longitudinal direction).2) and separating real and imaginary parts. &H/d(fi = 0) if [m20 + ^>]</> = 0. e.6) Inserting this into Eq.7b) the constant on the left and the coefficient of A on the right of Eq.Then m\ m\ = ml + A0A2 = m2.7a) vanish on account of Eq. and partly by climbing up the parabolic wall on either side. in a transverse direction. (22.2.tp2. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (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). . 22.7a) (22. i. (22. Classical Field Configurations the density Ti is minimized (i. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. as indicated in Fig.e. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 . Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. (22.
22. tangential to the classical path. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. r'.e. i. the existence of individual perturbation contributions). 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 £. 22. We can see the problem here by looking at Eq.r')S(t . This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction.e. 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. Goldstone's theorem applies to fully relativistic field theories. whereas ipi(x) is y ' " " " " " ^ . Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. The elimination of the . t') = d(r .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. There the wave function with an associated vanishing eigenvalue is called a "zero mode". of course.t'). In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. though not identical phenomenon.3 The spontaneously chosen phase seen from above. t. ^ = h eip / \ ^2 l . then a massless boson exists.7b). (3=0 Fig. Thus the Green's function is similar to that in electrodynamics. 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. which. 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. (22.
Classical Field Configurations other components results from the vanishing mass and the constraint called the gaugefixing condition which can also be looked at as the condition which removes from the Green's function G the divergent contribution.544 CHAPTER 22. i. see e. Lohe [179]. MullerKirsten [291]. (22. however. we shall need only at a later stage. W. . spinless field $(x. and we choose the metric ?7oo = + 1 . instead we shall consider static and timevarying solutions in important models of one spatial dimension. 22. Later we wish to develop a perturbation series about a classical configuration of $. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. Rebbi and G.t In particular the sextic potential can be considered along parallel lines. equates to zero the coefficient of this wouldbedivergence.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. F. Soliani [237]. 7711 = — 1 ( 1 6 Minkowski manifold). *S. C.£) in one spatial dimension. The Lagrangian density is taken to be C[^. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. N. To insure that the energy (see below) is positive definite we require V[$] > 0.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). 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.e. Behera and A.* Potentials of higher polynomial order in $ than $ 4 can also be considered. Zimmerschied and H. Khare [17]. J. A. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. t Potentials of this type have been discussed by M.g. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc.* We consider the theory (later: quantum theory) of a real. We make one more assumption concerning V which.d^} = ~d^d^V[^) = ^2^(^j V[$].
but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. ^ ^ * + V'[*]=0. 1]. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0.e. The first such condition to be imposed is that the energy of the solution of interest must be finite.13a) We write such fields <j>(x). .e. In a quantum theory the field $ is an operator defined on a space of states.«{x) = V'(<p).12) where the prime denotes differentiation with respect to $ . such that J > — — — = 0.22.l}. i. 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).12) we see that they satisfy the Newtonianlike equation <j. i.13b) We are not interested in just any solution of this equation.g] = \ v m For instance.t) has an explicit timedependence it is a Heisenbergpicture operator. The other condition which we impose (and study in detail) is that of stability. the socalled ^theory = ^V[g$.9) with quartic potential is defined by ^V\g$\ = ^V\g*.§ Unstable classical configurations are also important and will be considered in later chapters. (22.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. at least in the present case. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. The field $(x. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static. (22.l]. (22. (22. (22.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. — m$ 1 — cos Im 1 I ^ $ (22. Since $(x.11) (22. From Eq.
14) where the overdot implies differentiation with respect to time t.c = <£z + .T^ = 0.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. (22. 22.4 The wellknown soliton of $4theory.T = <I>.e. (22. <9$ i.e.2\dx 2 ' 1 fd& + V[$].7r) where IT is the momentum conjugate to $ defined by dC ? r = . Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($. (22.17) Since we have to integrate over all space. . the classical solution will therefore have to be such that this integral is finite.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. d. HI tanh m(x>k) 9 Fig. dC • Too = r?oo£ + =r$ = C + vr$ = H.546 CHAPTER 22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. We observe here already that the integrand of the integral in Eq. n = & . (22. i.15) The zerozero component is equivalent to the Hamiltonian density H.
11).18) is depicted in Fig. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. Again XQ is an integration constant. the $*theory.3 Soliton Theories in One Spatial Dimension 547 As we shall see below. 22. in the case of the first.5 The soliton of sineGordon theory. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". — XQ). In fact. O±m(xxo) (22.5. . (22.22. it is not difficult to find classical cnumber solutions 4>c to Eq. (22.777. 4m . i. m ( x . 22. > . (22.18) Here XQ is an integration constant. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . < = — tan [e >  ° ] 9 2 rrm/g — Fig.1 .13b) for the specific potentials (22. See J.21) This sineGordon configuration is depicted in Fig.x n ).20) (22. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .4. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq.10) and (22. Rubinstein [240]. 22.e. "The name sineGordon is derived from the analogous KleinGordon equation or theory.
With we obtain g(/>(a.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.o) =m7r.18) and (22.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .13b) which we can write d \W? = !"(•».* ( m' m 2 r)] ^2" .1 "»JtKxo) 9 1 .548 CHAPTER 22.20) are obtained from Eq.e. Classical Field Configurations The solutions (22.9 2 4>{x) = i. <^>(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 . In the case of the sineGordon theory defined by the potential (22. and hence 4/71 9<t>{x. or \(<P'? = V(4>)+ const.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). Here the constant is zero for V(<p) and <j>' zero at x = ± o o . 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. 4m In tan tan 5 l/e±m(*xo)tan^£^)U. Inserting the potential (22. (22. . ± 1[tanh.
19). instead of doing this now.17).22) and that its second variational derivative be positive semidefinite.22. (22. for instance.17) implies dx Jdx dx n£)' + ™ ( ' • # \ dx J S(f>(y) \dx and with partial integration we obtain f dx d fd<f>\ dx\dx J  5V(</>) 5(x y) + 5(j)(x) Tx5{x~y) . (22. . This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. The relation (22. (22. SE((f>) 5cb(y) 0. Then we have to demand that the functional derivative of the energy E(<p) be zero. Jackiw [141]. this does not require the action to be minimized. the curves rising monotonically from negative to positive values are known as kinks. A state of stability is therefore associated with minimized action and/or energy.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks".4 Stability of Classical Configurations The concept of stability can be complex. the discussion of Bogomol'nyi equations) in a simpler way. in the onedimensional Ising model. however.23) See for instance R. (22. Suppose now that we demand stability in the sense of minimized energy. we obtain these expressions later (cf. However. 22. The expressions (22. The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as. i. 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. the others as antikinks.e.21) for the energies can also be obtained from Eq.
e.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration.e. i. This is. in fact ip0(x) const.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. Classical Field Configurations Since y ^ ±00.550 CHAPTER 22. = 0.26) Comparing this equation with Eq. the function 4>c(x + a). a very general phenomenon which can be established as follows. of course.13b) with classical solution 4>c. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). (22. i.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 / 0 is not the same.13b).28) . The original Lagrangian was. The Newtonlike equation (22. written down in Lorentzinvariant form. its eigenvalues w\ > 0. Thus the vanishing of the first variational derivative yields again Eq. if we apply the generator of spatial translations d/dx to the classical equation. dx (22. the integrated contribution vanishes. (22. In fact. The eigenfunction 4>'c{x) is for this reason called a "zero mode".e. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. dx2 + V"{4>) (22. namely (p'c. with eigenvalue w2. in particular the invariance under spatial translations. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. Since it arises from the violation of translational invariance by 4>c. Thus for every translational shift "a" we obtain a new solution </>c in much the same way as we obtained new solutions by a change of phase in the Higgs model which we discussed before. (22. Eq. if 4>c{x) is a solution. it is also called a "translational mode".A. and the EulerLagrange equation retains this property. (22.13b) retains the invariance although the solution (f>c does not. we see that the latter has one eigenfunction. in fact. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. (22.25). i. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0.
We see therefore that the occurrence of these zero modes is a very general phenomenon. it is an expression like (22. 0 = S'[UC] = S'[gUc}.20) are such configurations in two particular models. and we have seen that Eqs. this integral has to be finite.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. We return to our considerations of stability. we are interested in classical cnumber field configurations of finite energy. / oo Thus for E(4>c) to be finite.17) we obtain oo dx[V(4>c) + const. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. (22.e. = 0. Then the vanishing of the first functional variation implies that 5S = 0.18) and (22.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. lim [V(<f>c) + const.]. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero. (22. (22. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). in fact. i. and that S[UC] is finite (or equivalently the energy).27). We can choose the potential V such that the constant is zero.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. In the onedimensional case under discussion we can rewrite Eq. (22. (22. i.e.33) x—>±oo .32) = V'(</>).31) Inserting this into Eq. As stated above. Thus TUC is a zero mode.] = 0 or x—>±oo lim V{(/)c) = 0 for const. (22.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. (22.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const.
.6. . This number is defined like a charge in field theory by the spatial integral of the time component of a current. ± l . ± 2 . the classical field (bc must approach one minimum for x — . Classical Field Configurations V Fig.e. in view of the antisymmetry of eM^. . Thus the classical energy or vacuum configuration is not unique. n = 0 .34) jfc" = e^dv(l since d^k^ = 0. In the sineGordon theory > • defined by Eq.552 CHAPTER 22. (22. 22. In the case of the 3>4theory we can define a conserved current k^ by (22. (22.c o and another for x — oo.6 The minima of V in $4theory at x — ±oo. oo J—oo y The topological quantum number q is defined by 0. . V{4>) = 0 at different values of </>). » In the <fr4theory defined by Eq. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory.35) and it is seen that in this case q can be ± 1 or 0.11) we must have correspondingly for x — ±oo: » 771 9 2vrn. (22.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. 22. The solutions can be characterized by an integral number called the "topological (quantum) numbcf^ "topological charge" or "winding numbed (for an illustration see Example 22.1). The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. . the potentials have degenerate minima (i. For a configuration to have finite energy.
We can see this as follows. continuous values. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. First. and we call this "topological stability. . Tfl \ (22.7T (22.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. termed topological. ± l ± 2 .cos(27rnA)].37) for the onedimensional case here. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition.A2)2 + 0 except for A2 = 1. zero is not a meaningful topological number. In the case of the sineGordon theory the minima of the potential (22. Again we see that E(\<f>c) is finite only for A = ± 1 .7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). . Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. one may ask how topology comes into the picture. A property is. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . and exp(ix(0))<pc reduces to 4>C when x(#) = 0.11) are given by ( so that for x — ±oo: > — J. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. . For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22.22. n = 0 . The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . 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.38) 4 V(\<t>c) >m ~2[l . 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . . in general.
43) **See e.g.13b) or ^ ' ( * ) ] 2 = V(</>). (For a broader view we consider briefly in Sec.41) dxy/2V{$)4>'{x). Eq.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). Classical Field Configurations involves the difference An = rioo — n. if oo / dxy/2V{<l>)<f>'{x). W. The inequality is saturated. (22.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. Thus with one space dimension higher there. 00 (22. its integral is nonzero only on account of nontrivial boundary conditions. (22.^ = N. Eqs. 22. We observe (a) the topological current k^ of Eq. the factor A = exp[ix(#)] varies over a complete circle).e. the energy is minimized. pp. J.40) (22.428.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. i. 22. (22.31) to (22. (c) the time component k° is a spatial divergence. .39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. 425 . where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. (d) the topological quantum number iV arises in a completely classical context. 00 (22.42) Thus the energy has a lower bound given by the right hand side of this inequality.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion).9 the foregoing arguments in a higher dimensional context. MiillerKirsten [215].554 CHAPTER 22.17)) f 00 > 0. H. (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.40)) this occurs when 4>'(x) T V2V(4>) = 0. 555 (22. B. (22. y/V(ft) > 0) oo =2nnm/g r<p=2imm/ g / dxy/2V{<f>)</>'(x) oo /•rj)=2Trm/g = / J<j>=0 d</>y/2V(<f>) = = n Jd>=0 rt=Mmlg n / J<p=o d<f>y/2V(<l>) 2 / ^\ — # W 2 1cos— .21).n > 1. In the sineGordon theory we can consider.44) This first order equation which saturates the inequality (22. . (22.39). In order to obtain a better understanding of how topology comes into the picture we recall that the sineGordon theory has an interesting analogue in classical mechanics.^ Clearly it solves the second order equation or (22. n E .40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.22. Bogomol'nyi [27]. (22. for instance. 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. Eq. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g.5 Bogomol'nyi Equations and Bounds We observe that (cf. We discuss this in the following Example.19). Here (V(c/)) > 0.
Solution: Consider a string along the xaxis as described. Initially the pendulums are suspended freely in the gravitational field. xn+\ — xn = a) 00 a —> dx. ka —> K.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis. so that oo > u. rotate) only in the (y.t) <j>(xn.g.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. We can now write down an expression for the energy of such a system of (say) n pendulums.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . z)plane and only such that their elastic strings remain taut.556 CHAPTER 22.t)\2 +mgr{\ .ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. Fig. Chapter XI. .7. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22.e. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.7 The pendulum analogy. Goldstein [114].cos</>(x„. Assuming that the pendulums can move (i.e. rotate) only in the (y. t)). total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) . Classical Field Configurations Example 22. H.46) **See e. with mass m and connected with short strings to neighbouring bobs. Construct an expression for the energy of such a classical system and its Lagrangian. all being identical. 22. At equidistant points xn = na along this string we attach strings with pendulum bobs.t) 4>{xn.t) measured e.g.2 V dt + k[(f>(xn+i. 22. } n= — oo roc —> / L J . from the yaxis as shown in Fig. Assume that the pendulums can move (i. 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. 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.
Eq. (22. If some w\ were negative. We can understand what stability means in the present case. (22.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt).t). (22. The equation is called stability equation or equation of small fluctuations. wk would be imaginary and so the factor exp(—iwkt) in rj(x.50) This equation is identical with Eq.t): {w With the ansatz r)(x. (22.47) into (22. and so would invalidate the procedure which assumes that rj(x.22.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. the topological quantum number. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i.49) exp(iwkt)ipk(x).25) obtained earlier. t) is the fluctuation (field). It is instructive to obtain this equation by yet another method. Hence we set ${x.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.7 linking the pendulum bobs). which represents an expansion in terms of normal modes. of the continuous curve in Fig.49) would imply an exponential growth in the future t > 0 or in the past t < 0.13a) and (22. 22. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc.t) = <l>c(x) + r}(x.47) where n(x.e.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. Inserting (22. is therefore also called "winding numbed. ' Equating coefficients of exp(—iw k t).13b) we have for small fluctuations rj(x. (22.t) = ° (22'48) (22. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). t) of Eq. 22. (22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x.t) is a small .25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. This number.
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£^)*.50) has the general form of a timeindependent onedimensional Schrodinger equation. (22. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . V"(<b) = 2m2 + 6g2(f>2. The equation of small fluctuations (22.1. In fact. i. Inserting (j)c of (22. Classical Field Configurations fluctuation. cos46» = 2cos2(2(9) . we shall see later that the fluctuations n(x.52) tan 6 := e±m^xo). approaches a constant value which is nonzero in both the <54theory and the sineGordon theory.558 CHAPTER 22. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction. In fact. and need 9 = tan" 1 j e±m^Xo) 1. .2 m 2 + 6m 2 tanh 2 m(x .x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22.20).18).A \ Inserting <f)c of Eq. Thus for stability w\ must not be negative. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. This means that the stability is not universal but only local in a certain sense. the zero mode. i. If that is necessary. For x — ±oo the quan> tity acting as "potential".e. V"((/)c(x)). we obtain V"(<t>c) .e.
53) = m(x — (22. (22. which satisfy the boundary conditions V>(o) = o.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . (22. ^+(0) d 0.tan26> l + tan 2 i9 / 2 (cosh 2 m(x — XQ) 1) — cosh m(x — XQ) cosh m{x — XQ) 1cosh m{x — XQ) 2m 2 cosh m(x — XQ) Thus V'fa) = m2 xo)) d2 .22. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.55) The differential operator is even in z.* For such a potential the spectrum can be b o t h discrete and continuous.e. We can therefore construct solutions of definite parity.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.2 (22. A  *(* + !) cosh 2 z tp(z) = 0.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. We can investigate the spectrum as follows. .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. Those of the other case are complex and periodic at infinity. — XQ)] cosh[±m(x — XQ)]: 1 . ip+. i.ijj^ respectively. Consider the equation fd2 dz2 . even and odd solutions. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.
I) .Z ) + ^(A + Z2) = 0. i. .58) = (1 + sinh2 z)~l/2x(z). A ^ An = .( 2 n + l . Since for £ — oo: * the function ^>+ is normalizable provided Z > n.( 2 n .59) Setting (22. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . (22.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22.60) terminate after a finite number of terms.e.e.l2 = . 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.61) Discrete eigenvalues are obtained when the series (22.Z)2.e.4 n ( n .560 CHAPTER 22. A * An = . i. (22. + l2) = 0. 2n < Z. when in the even case n ( n .e.Z ) 2 . and in the odd case n+\)U+\l)+\(\ i.57) Then the equation for x{z) 1S (22.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n. i.
63) we see that the continuum starts at A= 2 2^ = 0.54) by setting X = k2 Then 1 .65a) _dz2 cosh 2 z l=2. N = 0.2Z = . We can see the continuous spectrum of Eq.10) and (22.e. i. the continuum is w > Am2. for N = 0.is normalizable provided 2(n + .Ny.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0. We can now derive the spectra of the stability equation for the potentials (22.u2)—^V 2 dio and ip(z) = elkzV{to).2.1: w2 = 0. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .N<1. m i. (22.62) where even (odd) iV are associated with even (odd) eigenfunctions.11)... d w = tanhz.\=Kl2 From (22.51) into (22. °2 i.j < I .50) we obtain with z = m(x — xo) tp(z) = 0. (22.. (a) ^theory: the equation Inserting (22.22.62) we see that the equation possesses discrete eigenvalues w —2 . w2 = Am2.( 2 . From (22.1. (22.. (22. 2 .64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz. 3m 2 .e. m N<2.e. (22.6 The Small Fluctuation Equation 561 and the function ip.
562 CHAPTER 22. %(x) Fig. for N = 0: to2 = 0 and the continuum starts at w2 = m2.. .e. In the $ 4 theory we saw that (cf.66) We see that this is a nonvanishing function for any finite value of x. Thus either spectrum contains the expected eigenvalue zero. . m d . i. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22.50) we obtain with z = m(x — XQ) the equation (22.53) into (22.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. This property can. 22. of course. also be deduced from d(j)c/dx. The zero mode is depicted in Fig.. the zero modes. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. . 22.27)) . so that its derivative is nowhere zero (except at ±oo). and — as expected — this eigenfunction is even.=1A=^_. (22. . Thus 4>'c(x) has no node. . Classical Field Configurations (b) SineGordon theory: Inserting (22.8 The zero mode as typical ground state. d .65b) dz2 cosh^J. We know that 4>c is a monotonic function. N<1. Eq. It is now particularly interesting to look at their wave functions.8 and we see it has the shape of a typical ground state wave function.
1 o rrr tanh x — .e. .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ). (22. Finally we consider the normalization of the zero modes.z 0 )]oo = 8 .70) . (22.69) The quantity MQ here is called the mass of the kink (it is the mass of the soliton solution only in the classical approximation. . . i. d .6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to .e.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. (22.= E(<f>c). 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 .t a n h x . without quantum corrections). — XQ) OC Hence the conclusion is similar to that in the previous case.22. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. .68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i.oc (22.3 . Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.
41) for the energy of the classical configuration but now with respect to a ddimensional position space.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. ^For basic aspects of scale transformations see e. . The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. ddx {V4>f + v{4>) (22. to underline the fundamentally different nature of topological and nontopological finite energy configurations. Affleck [5] who uses in his Eq.e. it is important to understand their physical implications.72) with T{$) = I' ddx]{V(j))2 > 0. P(</>) = f ddxV{(f)) > 0. i.4) <j>(x) —• a<f>(ax).g. P. for if such configurations exist in a realistic theory with three spatial dimensions. H. (22.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . One can have different types of scale transformations — see e. H. J As a first example we consider the expression (22.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. Derrick [68].564 CHAPTER 22. I.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. (2. *See e.2d. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. we consider E(<P) which we also write as (22. *The considerations of this section extend in part beyond quantum mechanics and into field theory. Kastrup [145].T(4>). I. Classical Field Configurations 22. Olive [113]. Goddard and D. however.g.g. They may help. f G . Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). This question has important consequences.> 4>a{x) = 4>(ax). A.
80) .76) we see that E(<f>) can be minimized (i.e.. we write • = lim J r . since d2E(<pa) da2 2(2d)T(<f>). i.e. i. (22.e. a=l (22.d)T(<f>) = dP{4>).. (22. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension.71). the second derivative is positive) only for d = 1.xd).79) In order to emphasize that (f> — (p^ in any direction. Consider first the Lagrangian density £[&M = \{d^)(d^) .\ eMo(22. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).74) From (22.. to be finite the field <P ~^ 4>oo £ Mo for x — oo.V(<P) (22. d = 1 minimizes the energy.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}. we set 0 dE(<pa) da J a = i (22. In fact. (22.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.22.73) we see that this implies (2 . x = (x1.. 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.x2. (22. Next we consider more complicated cases.75) Since both T and P are > 0 and d is integral. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. 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}.78) For the energy of a static configuration.
S1. i.e. that we have a set of disconnected classical vacua (</>oo><^oo. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. Table 22. therefore. —00 or equivalently S = + 1 . are the same. is a connected region. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. —1. —00. A correspondingly topologically nontrivial field theory would therefore have to possess a!7(l) symmetry as in the complex Higgs model (recall that in our considerations of $ 4 .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.and sineGordon theories we assumed $ to be real.81) and (since d = 1) S = ± 1 . Thus in this case S^1 is a disconnected set consisting of two points. If we go to two spatial dimensions (d = 2). in fact. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. i. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00.1. . x = 00.566 CHAPTER 22. and. i.38). Then each point at infinity is characterized by a different value of the polar angle 9.e. We see. as is the case for the constant solutions. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . These observations are summarized in Table 22. — 1). (ptx = <f>oo.e.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . of • course. 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). (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a.
assuming invariance of C under the transformations of the group SO(3) in internal space. i. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. i. $(r) : Sl . i. i. 00.82) Now Mo := \ 4>(oc) ^24>2 = a2\. * n reach any other <^(oo) by simple phase shifts. This can be achieved e. 9) would depend only on r = x and hence the problem would be effectively onedimensional. VMx)) with (0(oo) = linvxx) (p(r. if cj> were real and position space twodimensional. map into.• M0.2.e. Thus.22.7 Existence of Finite Energy Solutions Suppose. We see. Thus. we have w e ca {Mx)}> and 1 3 1. Similarly. by allowing <fi to have three components in some "internal" (isospin) space. we would have only two points at infinity. for instance. if <p is real. to have topologically distinct classical configurations. i. If position space were onedimensional. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}.e.e. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. . the one point S = + 1 . i.e.83) .3 (^real) v{4>) = / (22. the field must map position space into field space at infinity. On the other hand.e.e. therefore. This will be "nontrivial" in the above sense only if 0oo is independent of 9. (22. phase transformations 6Q ^> 9Q + 59. that if the problem is to be topologically nontrivial. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial.g.1 . —00 or S = 1 . cj)(r. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}.
The symmetry of the potential (and Lagrangian) — i.e.. it is effectively a surface (and so a continuous set) S2. going once around S%~1 (i. If. Example 22. B. 11 H.V{4>a]. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. sphere.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) £(</>. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo.2.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. $ — $exp(ia) in a U(l) symmetric theory. through each point at infinity). + i ( D " f t W ) a .e.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. i. so that details of these models are irrelevant. the winding number is n. $ — > > $ exp{i Xlo=i Pa. we cover <S^_1 n times.568 CHAPTER 22.3. .e.. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. . a = 1. circle. the subsequent discussion requires only the general form of the Lagrangians. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved.„ where G r = 9 M £ . Summarizing we see that </>(x) is a mapping from S^T1 to <S^1.a M J . 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. % . invariance under replacements $ — —$ in $ 4 theory. i. Glashow [108]. the appropriate rotation.e. Georgi and S. Olesen [220].2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". Nielsen and P. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. A) = T(4>.</>] = ~G^Gai. A) + T{A) + P{<j>). Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space. $ — $ + 27 > • T in sineGordon theory. L. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3).
some contribution must be .A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. A. J 4 ) + A 4 .84) and investigate the existence of finite energy classical field configurations. (22..T{A). negative. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. (22.d)T(A) . >• A„ x ( x ) = AAM(Ax).84) is positive semidefinite and so must be finite by itself. 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. we have {^/»)A^ (22. We now consider scale transformations with A £ R.7 Existence of Finite Energy Solutions where T{<j>. Solution: Each contribution in (22.84) in this way one finds that the quantities involved transform T ( ^ . (22. A A ) = A2dT(^.86) (22.89) P{4>) are positive semidefinite and not all zero.d T ( A ) + AdP(</>).22.<^[x) transforms like a vector.i T(AX) = X4~dT(A).d T ( ^ . <0 >• <£A(X) = <KAx). i. T(A) = f ddxV(4>) > 0. A).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'>.e.e.87a) C?M„(x) — \2G^{\x). (2 . l j ddxGijGai] > 0.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).A).dP(4>) = 0. ^ A ) = A 2 . > Since T>IJ. (22. and so B ( ^ A .d)T(4>. A) + (4 . i.Ax) dX Since T((f>. (22. (i)^ 2d T Considering the contributions of (22. Mx') = E with a scale transformation x — x' = Ax. = 0. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also. Thus we set dE(<fix.
** e. 22.90) and we would have to add to Eq. (22. 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. Tchrakian and H. T^{<t>x) = \4dT(</>). H. W. Skyrme [253].89). tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. One can thus define a scalar field or wave function ty(x. Thus such a contribution could counterbalance T{4>. a = a(T) = a T —T c .92) The expression ^ ( x ) ! is known as the "order parameter. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models.89). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. Landau [111]. 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. In the following we consider a related theory.570 CHAPTER 22. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a socalled pure gauge theory with socalled instanton solution (which we shall encounter in a simpler context later). R. (22. (22. (e) For d > 5: In this case it is not possible to stabilize Eq.89) the contribution (4 — d)T(4>). Ginzburg and L.g. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. of course. a > 0. D.e. We then have a situation of the potential like that described in the context **T.t) such that  ^  2 describes the density of Cooper pairs. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. This case is. MiillerKirsten [268]. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. (22. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. J. exemplified by soliton and sine—Gordon theories. . 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. is temperature dependent and changes sign at the critical temperature Tc. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. ft V . Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). A) or P{4>) for d = 3. L. H.
The parameter 7 serves to put the minima of the double well at V = 0. from x < 0) means — as is familiar from electrodynamics — replacing V by V . but for T < Tc it has a double well shape. i.94) to check the dimension) 1 = I—a. assuming there is no superconducting material in this domain).8 GinzburgLandau Vortices 571 of the complex Higgs model (cf.e.2): For T > Tc the overall potential has a single well (minimum).93) J' (22.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. Variation as before.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . is a measure of the distance from the surface of the superconductor to where .22. Clearly the length (see Eq. 22.^ A n (this is called "minimal coupling of the electromagnetic field^ with vector potential A to the Cooper pair field ip in a nonrelativistic treatment) where g — 2e~ is the charge of the Cooper pair. (22. Sec.(tp) = 0. leads in a onedimensional case (considering this briefly) to the equation ^  = aV + /?^ 3 axz Imposing the boundary conditions (in the domain 0 < x < 00) ^(0) = 0. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. Then the static energy density becomes (cf. (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 < 0.2) £(^. Example 22. ^(00) = a (22.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. i. setting S£.e.
The magnetic field on the other hand.96) ipH. i. J.e. as indicated in Fig. 22. V x M = j s .98) where M is the magnetization resulting from atomic currents. i.M. . (22.e. The density of the Cooper or super current is given by the typical expression of a current. 2mi Setting < / > \P\e vp B = ViA. i. (22. MiillerKirsten [215]. W. exhibits a completely different behaviour — in fact. We can see this as follows. 22. 450.e. (22. Classical Field Configurations superconductor (x>0) Fig.572 no superconductor CHAPTER 22. p. h [^DV'V'DV]. the order parameter \ip\ attains (approximately) its asymptotic value. yJ—a/[3.9 Behaviour of ip and B in the superconductor. In the Maxwell equation See e. 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 ).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 . H.g.9.
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. m xi A B . These vortices therefore carry magnetic flux which is necessarily quantized .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. (22. and V x (V x B) = / J 0 V x j s . This is zero in the present case. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.22.div grad" and V • B = 0.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. Hence 0 = V x H = V x ( — B . K i ™*. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 .. i. In the onedimensional case we have B{x) = B(0)ex/XL. There are two types of superconductors.5 to 10~6 cm).L B = O.99) Inserting (22. Using the relation "curl curl = grad div .M i. Approximating p by its equilibrium value.e.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter.e. this becomes A B = no V x j 5 . V x B = [i0V x M = /lois. Also dD/dt = 0 in a static situation like the one here. (22. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value.
(22. de Vega and F.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. See also P. . Schaposnik [69] after their equation (3.574 CHAPTER 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. i. G.36). in other words. Finiteness of E requires \<t>\ = \<fioo\ on this circle. In the two models of Example 22.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. (22. Thus the asymptotic field </>(r. of two points on a line. 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). 'See the remark in H. In the NielsenOlesen model defined in Example 22. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). in fact a periodic function of the polar angle with period 27r. A.9). 22. we must have that lim V ( H ) = 0.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. where exp(ix(0)) is an arbitrary phase factor. A. De Gennes [67]. *A. Abrikosov [2]. x—>oo Thus at spatial infinity </» has the value of a zero of the potential.101) (p(r. We can think of (e.102) In the onedimensional soliton case of Eqs.9) ^e^ '>\<poo\. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. Classical Field Configurations (see below).e. the space consists of two diametrically opposite points on the circle.g) twodimensional space as being bounded by a circle at r = oo.37).. J. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. i. (22. The existence of these vortices has been confirmed experimentally.e.
The two theories we considered above in Example 22. Misner [96]. W. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. (22. 0 < 0 < 2TT.* 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. . Thus smooth deformations *See D. Finkelstein and C. Naturally an integer cannot change continuously.22. defines the orientation of this line element. abelian (i. The map A : S1 > S1 defined on S1 C R2 by (22.e.10 States of winding number 0. commutative). X is called the domain of / and y its range.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. Suppose <j>.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. (22. 22. i. 22.1. Sl := {zeC\\z\ = 1}. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig.1).10. We may define this as « = ^[X(0 = 2 T T ) .102) is called the exponential map of S1 onto S1. Then a continuous function of (j> on [0.9 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".X ( 0 = O)].e. In mathematical language S1 is therefore a compact. Fig.118)). topological group with the usual multiplication as group operation. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. 2n] remains continuous if the function assumes the same value at the two endpoints. The 1sphere is the unit circle in the space of all complex numbers.
e. i. i.g.e. connectivity. We observe that since time evolution is continuous.105) is described as "pure gauge" since it is determined entirely by the phase x(#).i'ldx(0) 9A Thus for (22. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). the winding number must be a constant of the motion.. The gauge field required in the asymptotic limit (22. r^oo 1 dx(9) n r o0 (22. (22.104) so that d2x hJ Since A has direction e#.) remain the same under homeomorphic transformations and are therefore called topological invariants.e. elements in the same class have the same topological characteristic. The different kinds of connectivity of these topological spaces (e. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). it retains this twist or hole under the continuous distortion) and is described as a homeomorphism.576 CHAPTER 22.. In other .105) the corrections falling off faster than 1/r. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. i. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. no hole. The winding numbers n are such topological invariants..104) to be finite we must demand that g . In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. finite. we can write 1 d h J ee rde w At h r^oo . one hole. if it has one twist as above or a hole. we required that lim r—>oo w o. For that reason n is called a "topological invariant?.g. Thus two homeomorphic spaces have equivalent topological structures.
105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. the vector potential A.22. cannot be deformed continuously into one with two loops (n = 2). if A = V x . An example is the mapping H(t. It follows that the field Fij. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).e. the gauge field. i. i. with n = 1. Therefore. 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".e. the group space of (the phase transformations of) U(l) continuously.9):=t(f>£\9) + (lt)(f)W(9) with te[0. However. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. (22. 9){^ and hence the functions (22. for B / 0 . the higher order contributions in (22. Of course. the number of flux quanta 2irh/g being the winding number n. A field configuration with one loop (or twist).105) insure that B ^ O . it is determined by a gradient (l/r)d/dd.l]. cannot be pure gauge everywhere. we can consider transformations which describe continuous transformations of a configuration with fixed n.109) A W(0) = e^ W. is also zero at r = oo.e.110) . We can calculate the magnetic flux $ through the region with $ = * rd9A0 = .106) J g j r d d g Thus the magnetic flux is quantized. However.<b rd9—4^ = n. we can go n times around S^. into S i . (22. Ffiu.108) map the spatial circle S^. (22. i.9 Introduction to Homotopy Classes words.107) The fields <f>(r. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22. going once around S^. we have B = V x A = curlgradx = 0.
e. 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. Solution: As stated. Reuter [77].e. the unit interval I = [0. n. Two curves which can be related to each other in this way. We verify these in the following example. i.T. it is clear that this is also a homotopy. Dittrich and M. (b) symmetry and (c) transitivity.9) is called a homotopy (emphasis on the second syllable). 8) = cfinl\9). Steenrod [261] and S. Classical Field Configurations which is continuous in the parameter t and is such that H(0. Since the passage of time is continuous. (b) symmetry and (c) transitivity.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).lQSl^S^ with H(0. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. and the space of t.e.8) itself must be a configuration with definite winding number. i.113) . are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).9) = <fiW(9) and H(l. H.g. Example 22. A superposition of classical configurations with different winding numbers does not minimize the energy.0) = $\6). (fin (9) in the above example. (22.1]. S%. H(1. — S\ > (22. We can consider (22. 9) is an equivalence relation. Thomas [270].578 CHAPTER 22. i.e.111) In principle an expression like (22. n' ^ n.9) = ^°\0). i. the phase factors of (f)(9). (22. The relation (22. But this type of combination is excluded sind H(t. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).110) could be written down for configurations with different winding numbers.e. e. A brief introduction can be found in the very readable book of W.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. time cannot make a field jump from one homotopy class (see below) to another.3: Homotopy an equivalence relation Verify that the homotopy map H(t. The map H(t.111) as a map from the space described by 9. these two maps (fin from S^. In fact. to the internal group space of £7(1). Hu [134]. f T h e standard texts on homotopy theory are N. H. like <pn (0). we have to show that the map satisfies the three properties of (a) identity. i.
0) = </>(0) is continuous there. 0). Then H(t. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. 0) = tf2(l.0) for for 0<t<. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent.* 22. 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. On the other hand H(0.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. i. i<t<l. then </>W ~ <t>{0)• We define an H(t. . 0) = ^ (0) and 6(1.10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the onedimensional sphere S1. we see that of the set of these maps some are equivalent and others are not. In some cases the set of homotopy classes possesses group structure (satisfying the group properties).e. 0) = 4>W (0). Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'.9) = <f>(0\6) and H(l. . is isomorphic to) the set of integers Z (positive. 0).e) tf2(2il. U(l)). is such that H(0.9) = . below). at t = 1/2 both i?i and H 2 give c/^1'. (c) We can demonstrate the property of transitivity as follows. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. <f>W .ffi(O. 9) = H(l. 9) := H(l .9) H(0. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0.22. a class being a set of equivalent maps. 9) = H(0. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). ' . Moreover. negative or zero). i? is a homotopy connecting </>(°) with <^>(2'. being the continuous map connecting </>(°) with <j>(1'>. \ M)\ H!(2t. (b) Next we check that if 0<°) ~ <p^. Hence <f>(0^ ~ <p 2 ).e. Returning to the maps \(x) : Si ^ S1^ (i. <^(2).t. 9) connects <pn (0) with itself continuously. since and tf (1. 9) = <f>W (0). as they are called. these groups are then called homotopy groups (cf. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. 0) = <£{2). which therefore demonstrates that the homotopy relationship is transitive.e. so that H(t. Homotopy classes are sometimes also called Chern—Pontryagin classes.0). We now define the function H ( f 0 ff .
and then introduce some of the related mathematics. but TTI[S2] = 0: S 1 is a circle.g. Rajaraman [235]. 22. 22.580 CHAPTER 22.e. Sketch of idea demonstrating that nilS1] = Z. Fig.) why 7i"i[S2] = 0. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors.115) Varying t € [0. 22. p.11 The trivial n = 0 map allowing shrinkage to a point. S 2 the surface of a sphere in a threedimensional Euclidean space. How can we see this? We first sketch the idea. 52. In the diagrams^ of Fig. . 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. Appropriate maps. T (22.1] 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. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. n=0: t=1: t=1/2: i. which is to be mapped into another circle Si indicated by dashed lines.11 the circle drawn with the continuous line is the circle S%. to make the result plausible. T < 9 < 2TT. In order to really appreciate how topology comes into the picture here one should understand (e.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T.
103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x).e. d (22.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. to the case of winding number zero.h.117).116) In fact from this we obtain (see below) 2  n and this means r.117) in ±.22. 2 the maps are wound around the dashed circle. the transformations^ (the result may be looked at as 8n of expression (22.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. i.12. 0 < X(P) < 2vr. is invariant under infinitesimal (continuous) deformations. The maps for these cases are illustrated in Fig.117) 2^ d6X(x)\1(x). . (22. 22. One can prove that the expression (22. (22. 22. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. and hence the winding number. In the cases n = 1. 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. We make another observation at this point.h.e. of (22.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. For winding number n we have the mapping (i.s.s. i.e. Coleman [55].12 The n = 1 and n = 2 maps not allowing shrinkage to a point.
the cyclic group with two elements (since 50(3) is doubly connected). il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. whereas 7ri(50(3)) = Z2 (i. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided.e.582 CHAPTER 22.8. only one homotopy class. In fact. i.e. there is a theorem which says thatH TTg(Sn) = 0 for q < n. which means that a curve connecting any two points of S2 can be continuously deformed into every other curve connecting the two points) and one writes 7T1(S2)=0.e. the identity class (the group space S2 is simply connected. . Thus there is only the trivial map into a point. an integer modulo 2) i. Sec. Steenrod [261]. Classical Field Configurations Under this transformation n changes by 5n Now. 15. One can show that the group space of SU(2) which is the sphere 5 3 C R4 is simply connected and that therefore TTI(S3 = 517(2)) = 0. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. 1 1 See N.
e. 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. finite Euclidean time) classical configurations (which are nontopological). 23. 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!).Chapter 23 P a t h Integrals and Instantons 23.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. We consider such a theory now with action S((p) depending on the real scalar field 583 . in such a way that the factors appearing in the result are exhibited in a transparent way. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. Since we have not yet introduced periodic (i. An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. different from literature.
n = 0.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. and ^2(0) : m* . note that here the mass of the equivalent classical particle is taken to be mo = 1. "For comparison with literature. in particular E.2) Fig. i. ! " ^ j = ^ 2 .2.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . 23. Chapter 6) 1 2 .* /•tf CHAPTER 23. 23.2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq.584 <)>{t). Patrascioiu [110] and Chapter 18.1.e. Fig. (23. Path Integrals and Instantons S(<f>) / dt \i>2 .1)) and hence E0 = mh/y/2. The Hamiltonian of the simple harmonic oscillator (cf. Gildener and A.1 The double well'potential.v{4>) <?VV m4 __ 1 . (23. 2 2 4 ™ = ~. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.
2 The wave functions of the lowest states. this alone does not yield the level splitting. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). e.In mechanics of a particle of mass 1 with ~ position coordinate x(t). The states of the system here must be even or odd under x — ±x.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. i.x) = x2 V(x).23._(0) = 0. but there is tunneling between the two minima if the central hump is not infinitely high. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . 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> ~ * ±</>.e. and in the above 4> corresponds to x. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig. as we did in Chapter 18 — but as we saw. must have definite parity as in a physical > situation. W(0) = 0. we have the Lagrangian L(x. 23. We can develop a perturbation theory around either of the minima. This tunneling affects the eigenvalues. 23. Fig. </.g.2 — and we can deduce from this the boundary conditions.
i. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.5) and iS = dLE d dr[d(d(f)/dT)} i. (23.£'(0) = 0.6) d*l dr 2 .8) V(4>) + const. dr m+™ (23. dLE = 0. But quantum mechanically it can tunnel through the hump from one well to the other.4) J <bi =—(An rTf (ITLE = I JT. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. For small h we can attempt a stationary phase method. (23.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21). Then (0o. (23. (23. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.3) for tf — oo. Euclidean time T = it.tf\ . The transition amplitude for this. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.2 m 2 ( l m z (23.  . i. that given by Eq. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.7. i.0o. In the semiclassical path integral procedure that we want to use here. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. is finite if we go to imaginary. r . Equation (23.e.7) so that (23. classically forbidden domains then become "classically" accessible. dr We observe that Eq.2.586 CHAPTER 23.U) = J'D{<f>}exp[iS(<l>)/h] (23.e.<h.9) .2 V'(<l>) = .3).e. SE.e.
e. in the present case (T — To) = m tanh m(r — TO). timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). .10) where TO is an integration constant. Fig. i.3 The instanton path. and hence kinetic energy gained is equal to potential energy lost. we see that the equation is the same as the classical equation for a static soliton (i.2 Instantons and AntiInstantons 587 In fact. this means the total energy of the particle in Euclidean motion is zero. from fa — —rn/g. Thus we know the solution from there which is the kink solution (22. Its trajectory.e. The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description.e. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. is depicted in Fig.18).3. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). i. — (23. 23.23. 23. the classical path. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".
10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.14) But this is not yet enough.13). (23.Tj) oc exp 4 m3 m^2 (23. The configuration (23. thus indicating the nonperturbative nature of the expression. 23.r 0 ) ' (23. The classical equation (23.13). we obtain lim f+ OO.13) We observe that this expression is identical with the energy of the static soliton of the 1 + 1 dimensional soliton theory which we considered in Chapter 22.11) For this quantity to be finite for TJ = —oo. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.]. the constant must be zero.8) we obtain T f SE dr[2V((j)c) + const.10) is given by (23. Using Eq.7"/! ).7) admits another configuration which is the negative of (23. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23. Path Integrals and Instantons The Euclidean action of the instanton solution (23. (23. This antiinstanton configuration rises monotonically from right to left as shown in Fig.13) into Eq.^ . Inserting the result (23.588 CHAPTER 23. Tf = oo.TJ T fro.5).10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.' .* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r . For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.4. but with the same Euclidean action (23. We also observe that the expression is singular for g2 —> 0. (23.4).
10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. 23.r0). (23.g.23. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. 23.15a) we have a simple superposition.5.15b) is not exactly a solution.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 . from the sketch or from the variation of SE. (23. The sum (23.15b) since we require configurations from vacuum to vacuum.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). . b u t if TQ. 23.15b) 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. 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 —1.15b) a solution of the classical equation? Is it a topological configuration — if so. 0c(r) = &(T). and an antiinstanton localized at (say) TQ. in the case of (23. e.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig. / > Looking at the instanton (23. Questions which arise at this point are: Is the nonmonotonic configuration (23.g. 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 (23.5.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.TO) + ^ c ( r . as we can see e. #(r) = </>C(T .15b) has the shape shown in Fig. Clearly we are here interested in configurations of type (23.4 The antiinstanton.
In fact. since (23. 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 . at oo the variation is <<> = 0).5 Superposition of an instanton and a widely separated antiinstanton.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.16) . we see that in the domain [—oo. g cosh m(r — TQ) (23. Path Integrals and Instantons '. 5^ In order to answer the second question we observe that (23. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig.T 0 ) } ] . 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. 23.15a) is not exactly a classical solution.15a) is a configuration which starts from 4> = —i^/g and ends there. vanishes as T\ —> ±oo (of course. Finally.15a) can be argued to be zero. of course. #(r) dr 1 1 m 9 cosh m(r — ro) cosh m{r — T'Q)_ cosh2 m(r — TQ m cosh2 m(r — TQ) g cosh m{r — ro) m 1 [1 . it also does not exactly stationarize the Euclidean action.e x p { . Ignoring exponentially small contributions the Euclidean action of the configuration (23.2 m ( T o . T\] with LECIT = 0.CHAPTER 23.
23.2 Instantons
and
AntiInstantons
591
and looking at Eq. (23.13) a n d ignoring exponentially small contributions, we obtain (note t h e same values of t h e limits)
<Koo)=(/>(oo) SE dA
Jd> oo)
#A*T
For t h e configuration (23.15b) we have with a similar argument
fTi SE \Joo roo JTx \ / /
rTi
( [ * • • • + [°° •••)LEdr=
( f
\Joo
1
...
f
i— Ti 1
...)LE
J
= 0.
Joo
As a further example we consider a configuration consisting (approximately) of two widely separated instantons (or kinks) and one antiinstanton (or antikink) localized at TQ , TQ and TQ respectively, § # ( r ) = ^ T  T o ^ T  T f M T i  T )
0(r  T I ) ^ ( T  r^VCTa  r)
+^(rr2)^c(rr(S3))0(T3r) with  o o = T 0 < r^ 1} < 2 \ < r^ 2) < T 2 < r ^ < T 3 = oo.
(23.17a)
Fig. 23.6 Two widely separated instantons and one antiinstanton. Again t h e configuration is nonmonotonic and as in t h e previous case one can show t h a t it is only approximately (with exponentially small deviations) a solution of the classical equation. In order t o answer t h e question concerning the associated topological charge of </>(3)(r) we recall t h a t this is determined by t h e boundary conditions of the configuration; thus in t h e present case t h e configuration belongs to t h e instanton or kink sector with topological charge
3
r (3) r
0
( E.g. in the overall range of r = 2T we can choose the locations TQ i ) : 2T/3.
2 T / 3 , T ^ 2 ) = 0 and
592
CHAPTER 23. Path Integrals and Instantons
+1 (like the single instanton or kink). Thus for each of the four charge sectors (corresponding to the two vacua and the kink and antikink sectors) we have a multitude of approximate solutions^ of the classical equation as illustrated in Fig. 23.7 in which (a) and (b) depict the two constant vacuum sectors and (b) and (c) the instanton and antiinstanton sectors respectively.
(a) (b) (c) (d)
Fig. 23.7 Classical solutions in their topological sectors. In evaluating later the Feynman amplitude about such configurations we use the completeness relation of states. Thus in the case of </>c (r) we write
oo poo
/
oo
d<t>2 /
J —oo
d<j>1(<t>f,Tf\(j>2,T2){<l>2,T2\(f>l,Ti)(<l>l,T1\<l)i,Ti).
(23.17b) Evidently the evaluation of this amplitude requires two intermediate integrations.
23.3
T h e Level Difference
The nonrelativistic Schrodinger wave function ip of a discrete eigenvalue problem for a potential V satisfies a homogeneous integral equation which we can write ip(x',t')= dxK(x',t';x,t)tp(x,t), t' > t. (23.18) Here K(x',t';x,t) is a Green's function satisfying the differential equation DK(x', t'; x, t) = iS(x'  x)5(t'  t) with differential operator d Id
2
(23.19)
(23.20)
" B . Felsager [91], p. 143, therefore calls these solutions "quasistationary configurations". Further discussion may be found there.
23.3 The Level Difference
593
(One can convince oneself by differentiation that the integral equation is equivalent to a timedependent Schrodinger equation). In Eq. (23.20) V0(x) = V(x)  SV(x), V = V0 + SV,
where 5V{x) is the deviation of the exact potential V from an approximation VQ. (Since we cannot — in general — solve a problem exactly, we have to resort to some approximation; thus we have to distinguish between the exact problem and the approximate problem which we can solve exactly; the Green's function is constructed from the eigenfunctions of the exactly solvable problem). The Green's function K(x', t'; x, t) can be expanded in terms of the eigenfunctions T/4 (x) of the Schrodinger equation for the potential VQ(X), i.e. (to be verified below)
K(x',t';x,t) =
We have (cf. Chapter 7)
^Y,f 
dE
'•i0)*(x')^0)(x)exp{iE(tlt)} EE&0)
(23.21)
" 1 d2  V0(x>) 2mo dxf2
^<?V) =  i « V )
so that DK(x',t';x,t)
•JL
dt1
V0(x') +
d2 2m 0 dx'2 EE, dE 2^
(0)
^h(x/)^n0)(x)exV[iE(t't)]
n
J
iE(t't)
=
i6(xx')5(t'
t),
(23.22)
as claimed (note that the first delta function results from the completeness relation of the eigenfunctions of the exactly solvable problem). We can go one step further and integrate out the independence of Eq. (23.21). Replacing En in the denominator by (En —ie),e>0,so that the integrand of (23.21) has a simple pole at E = En —ie, and integrating along the contour shown in Fig. 23.8, Eq. (23.21) becomes (with the help of Cauchy's residue theorem) H K(x', t';x,t) = Y, Tp{n}* (x')^n0) (x) exp[i£4°) (*'  t)}9(t'  t).
"This formula is discussed, for instance, in R. J. Crewther, D. Olive and S. Sciuto [60].
(23.23)
594
CHAPTER 23. Path Integrals and Instantons
This is our earlier expression (7.9).
Im E t  t>0
ReE
Fig. 23.8 Integration contour. We now recall from Eqs. (21.10), (21.17), that the Green's function K can be written as a path integral, i.e. that K(xf, tf;xi,U) = J cD{x}eiS/H{tf  ti). (23.24)
Evidently we want to relate Eq. (23.23) to Eq. (23.24) in the case of our onedimensional $ 4 theory. In the derivation of Eq. (23.23) we distinguished between the exact problem with potential V and an approximate problem with potential VQ which can be solved exactly. In an actual application such a subdivision can be a matter of convenience. Thus we could try to approximate V by taking as VQ the harmonic oscillator part of the potential and 5V oc <5 4 then the set {T/4 } would be the set of oscillator eigenfunctions ?>; which is wellknown. Alternatively, we could be interested in a completely different problem, e.g. one with Vo given by Eq. (23.1) and 8V oc <^6. In this case ipn would be the exact eigenfunctions of the Hamiltonian for the (/>4potential (23.1). It is shown for instance in scattering theory that the exact transition amplitude is characterized by poles at the exact eigenvalues (e.g. in the case of the hydrogen atom all discrete eigenvalues are simple poles of the scattering amplitude). Thus if we want to relate (23.24) to the path integral representation of the transition amplitude for the (/)4potential (23.1), we must take in (23.24) the exact (real) eigenfunctions ipn{x) and eigenvalues En for the potential (23.1) which, of course, we do not know. We proceed as follows. From the above equations we obtain K(fi,t';4>,t) = Y,M<i>')Tpn(<t>)eMiEn(tt')/h}
n
= f %>{<!>} exp[iS/h].
•*
(23.25)
23.3 The Level Difference
595
This equivalence of the standard decomposition as a sum on the one hand, and on the other hand with the Feynman integral on the right is the reason for describing the Feynman amplitude also as kernel. We now consider the left hand side of this expression for Euclidean time r = it. Since r' > r, the expression is dominated by the lowest and next to lowest eigenvalues E+,E associated with the ground and first excited state wave functions ip+(<fi) (even) and i/j(<fi) (odd) respectively. Thus the left hand side can then be approximated by (with h = 1)
K(<//,T';0,T)
~
^+(cf>')^+(4>)eME+(T'T)} +^_(0')^(<£) exp[E(r'  r)].
(23.26)
We are interested in the transition with oo and Moreover, V>+W>o) = V+(<fo), so that K(<f>',T';4>,r) ~ {^ + (</>o)} 2 exp[E + (T'T)] {</>_(0o)} 2 exp[£_(T'T)]. We define AE, E0 by** AE == EE+ so that E+ = E0Then
K{<J)Q,T'\<J)Q,T) <t>Q
(23.27)
^(0o) =  ^  (  0 o ) ,
(23.28)
(23.29)
(level splitting),
E 0 =  ( E + + E_),
(23.30)
AE,
E_ = E0 + AE.
(23.31)
~
{V'+(0o)}2exp
 ^  l A ^ ( r '  r ) (r'  r . (23.32)
{V>(<fo)}2exp
 \E0 + AE\
Here the wave functions at the minimum position (fro are comparable to (i.e. approximately given by) the corresponding ground state eigenf unctions of the harmonic oscillator and so (apart from overall normalization) (cf. Chapter 6) tp±{4>) oc < exp 1
(<£ + <
±exp
{<!><t>o?
= ±ip±{<f>) (23.33)
**Here AJE denotes the splitting of the originally degenerate oscillator level. This is therefore twice the deviation of a split level from the originally degenerate oscillator level.
596
CHAPTER 23. Path Integrals and Instantons
with E0 = mh/y/2 ( as we noted at the beginning of this chapter). Hence we can set 1 ± e _2 *° V#o)(23.34) l^±(0o)l = e2*§ ± I Then
K{(J)Q,T'\(J)Q,T)
« =
{V(^o)}2exp[JE;o(r'r)]eA^')/2eAE(^)/2 2{V(^o)}2exp[JB0(r,r)/^]sinh^:AJB(T,r), (23.35a)
where we reinserted h for dimensional completeness (observe that when AE(T' — T) = AEAT ~ h and small, the hyperbolic sine may be approximated by its argument and one returns to (23.25) for Euclidean time). The relation (23.35a) assumes fixed endpoints 4>o, i.e.
d<j>" j #'(V</>X</>" V
foMW.rW
+ 0o)<^'^>
(23.35b)
= <^0o) (</>0,r'  0o, r) (cp0\ip).
v '
Clearly without the delta functions the end points are not fixed. (This situation will be required in Chapter 26 in the case of certain bounce configurations). We now want to deduce the explicit expression obtained with the path integral formula and then extract AE by comparison. In this comparison we require r' — r to be a large Euclidean time interval (—T, T) with T — oo. It is clear from Eq. (23.35a) that we cannot put T = co rightaway; > thus we also require the study of a gentle approach of T to infinity, which necessarily makes the calculation more involved. We also note that since the above expression assumes quantization (discrete eigenvalues), the comparable formula must also involve quantization. Thus we must go beyond the purely classical contribution, i.e. we have to consider field fluctuations about the classical configuration.
23.4
23.4.1
Field Fluctuations
The fluctuation equation
We consider fluctuations n(r) about the instanton (or correspondingly static kink) configuration (pc: </>(T) = MT) + V(T). (23.36)
23.4 Field Fluctuations
597
Of course, the level splitting AE (to be extracted later) must be based on a symmetric treatment of instanton and antiinstanton. Allowing for fluctuations which are not tangential to the path one necessarily introduces more degrees of freedom — in much the same way as in allowing relative motion together with collective motion. Thus the consideration of fluctuations leads to the consideration of a larger (in fact infinite) number of degrees of freedom, i.e. to the continuum. We naturally impose on the fluctuations the boundary conditions 77(T = ± T )  T  O O = 0. (23.37) Clearly we have to expand the action into (23.5) we obtain
SE((J))
about
4>C{T).
Inserting (23.36)
«« />[(K
(ITLE = /
i(g+ £) + m + „ /«/>[K* + £,
dr
oo
1 / d0c \ ~ m"* 2 2 V dr J ' 2g
2
m
2J,2
1 2J,4 re+2*
+
WlA{t) *»**w+tf*
(23.38)
,2J2 J +Sg*<%rjz + 2<?2</>cr?3 +  5 2 T ? 4
Here the first pair of curly brackets {• • • } yields as before the contribution 4 m3
SE(<PC) = V~2
3?
(cf. Eq. (23.13)). In the second pair of curly brackets in Eq. (23.38) we use the identity
s:Am  MW.
dr where we used Eq. (23.37). Hence
—oo •2,
oo
dr2
(23.39)
dr2
•V(j),
+r,2{^924>l ~ m2} + 2</V0C + g2rf
(23.40)
598
CHAPTER 23. Path Integrals and Instantons
The term linear in 7 vanishes as a result of Eq. (23.7). Hence 7 SEW and SE(<j>) = = Here
oo
= SE(<l>c) + SSE(ri)
(23.41a)
JdrLE^c
+ v) + (23.41b)
SE{ci>c)+lJdTr]\T)L'E{4>)W
dr
/
00
\ it) ^^ ~ ^
jvi^eMSE^/h]
T){<p} eMSE(4>c)/h]
+
+2ffW+ .9W (23.42)
\
Hence, since the Jacobian from </>(T) to T?(T) of Eq. (23.36) is unity (with iS = —SE and subscript 1 in the following indicating that the single instanton case is considered),
I
exp[6SE(v)/fi] (23.43)
exp
4 m6 Jv{r)}exp[8SE(ri)/h}, 3g^h,
where we used Eq. (23.13). We consider the approximation in which we restrict ourselves in Eq. (23.42) to terms quadratic in n. This approximation is called the Gaussian or '•'•oneloop''' approximation.* Then
oo r 1 / j„ \ 2
dT
/ oo oo
(23J.0) f 00
\{tr) +'W*S»'>
2\dTj
dT
i/^y
2
+ rf m 2 {3tanh J m(T  r 0 )  1}
(23.44)
1
(7?, Mr?),
where M is an operator.t We set (observe that r acts only as a parameter which parametrizes the classical path; the dynamical — real timedependent — quantity is £n) V(r) = ^2^nrjn(T),
71=0
(23.45)
"The word "loop" refers to internal integrations which this nonclassical contribution requires. ^The factor 1/2 is extracted so that the fluctuation equation assumes the form (23.54b).
23.4 Field Fluctuations
599
where {r/ra(r)} is at every point r of the classical path a complete set of orthonormal eigenfunctions of M with eigenvalues {w^}: Mrtn(T) = wlrin(r), dTrjn{r)r]m{T) = p25nm.
(23.46)
Here we have in mind p 2 = 1. However, for ease of comparison with the literature* we drag the parameter p2 along. Then
(T?, Mr,) (2 = 5) E [°° dr(^Vn, M^Vm) = E l & l W 
(23.47)
Formally, if we insert (23.36) and (23.44) into h of (23.43), we obtain h = I 0 exp with 4 m3' (23.48a)
I0 = Jv{r,} exp
where
2^M??)
(23
^7%
fc=0
exp
n
(23.48b) D{77} = Urn FT dr](rk)
(23.49) (in our earlier discussion (cf. Sec. 21.6): 77(T^) — %, and r?(r) is given by • Eq. (23.45)). With v(Tk) = Y^nVniTk)
n
we have dr](rk) = E d CnVn(r k ),
n
(23.50)
and (the determinant being the Jacobian of this transformation) (23.51)
fc=0 fc=0
Hence I\ = exp 4 m3 3 ^
d
K^)/(n^) e [4? le " 12 ^
•fc=0
(23.52a)
•"•Note that a different normalization of {rj n } as in E. Gildener and A. Patrascioiu [110] (they have in our notation p2 = h/fj,2,fj,2 = 2m2) introduces additional factors on the right hand side of Eq. (23.47) and hence changes some intermediate quantities.
600
CHAPTER 23. Path Integrals and Instantons
Now,
2 2 d^e w P e/2
J
2TT
9 9'
(23.52b)
—c
so that the integral in Eq. (23.52a) can be evaluated formally to give I\ = exp 4 m3 3 ^
•*(%>) fl V C
d
k=Q
2TT
1/2
(23.52c)
provided no w^ is zero. We now show that this condition is not satisfied, i.e. we have one eigenvalue which is zero corresponding to the zero mode associated with the violation of translation invariance by the instanton configuration. This is the case if the zero mode, like all fluctuation modes, is normalized over the infinite time interval, not — however — if a finite interval of T from — T to T is considered, in which case the associated eigenvalue is nonzero (as will be shown below) and Eq. (23.52c) is well defined. Hence first of all we show that the differential operator M is the differential operator of the stability equation or small fluctuation equation. The operator M was defined by (23.44), i.e.
i r<x>
^{V,MV)
2
dr nMrj = \ drr/Mr]
OO oo
\fJTn
'dr]\
K£) + 7 ^ > (23.53) £ + ""<*•> V,
dr
J —(
where in the last step we performed a partial integration and used Eq. (23.37), i.e. n(dtoo) = 0:
dr]
°°
f°°
,
d2
From Eqs. (23.44) and (23.53) we obtain
M v"(4>c
= 6g2(f>c2 — 2m 2 = 6m 2 tanh 2 m(r — TQ) — 2m2 (23.54a) 4m 2 — 6m 2 sech 2 m(r — To).
Thus Eq. (23.46) becomes d2 1 (23.54b)
23.4 Field
Fluctuations
601
We have considered the spectrum {w^} and the eigenfunctions {r]n} of this equation previously; cf. Sec. 22.6. There we found t h a t the spectrum consists of two discrete states (one being the zero mode) and a continuum. In Eq. (23.45) we have to sum over all of these states. We also observed earlier t h a t the existence of the zero mode d4>c/dr follows by differentiation of the classical equation, i.e.
A.
dT2
dr'
:[V
(</>)], implying
dr2
V'itc
dr'
°'
Thus we know t h a t the set of eigenfunctions { ^ ( T ) } of M includes one eigenfunction T)Q{T) with eigenvalue zero. Before we can evaluate the integral (23.52a) we therefore have to find a way to circumvent this difficulty. First, however, we consider in Example 23.1 a gentle approach to t h e eigenvalue zero of the zero mode by demanding the eigenfunction ipo (r) to vanish not at r = oo but at a large Euclidean time T, as remarked earlier. § Some part of the evaluation of the p a t h integral will then be performed at a large but finite Euclidean time T, and it will be seen t h a t this large T dependence permits a clean passage to infinity. Example 23.1: Eigenvalues for finite range normalization
Derive the particular nonvanishing eigenvalue of the fluctuation equation which replaces the zero eigenvalue of the zero mode if the latter is required to vanish not at infinity but at a large but finite Euclidean time T. Solution: The operator of the small fluctuation equation is given by Eq. (23.54b). We change to the variable z = m{r — TO) and consider the following set of two eigenvalue problems: (a) Our ultimate consideration is concerned with the equation ( — j + 6sech 2 z  n 2 J ^0(2) = 0 with However, we consider now the related equation (b) ( —.r+6sech22n2)ipo(z) V dzz J = ^^(z) m2 with ^ n ( ± m T ) = 0, n 2 = 4 and Vo(±°°) = 0.
and our aim is to obtain ui 2 . We proceed as follows. The unnormalized solution of case (a) is (cf. Eq. (23.12)) dt/>c m 2 1 ipo[z) = —— = =—, z = m(T T0). orr g cosh z (Note that this is the zero mode which normalized to p 2 , i.e. / drr/^ = p 2 , is 770 = p(d(pc/dT)/^/SE, where SE is the action (23.13)). We multiply the equation of case (a) by i>o{z) and the equation of case (b) by ipoiz), subtract one equation from the other and integrate from —mT to mT (i.e. over a total T—range of 2T). Then
fmT
JmT
, f . dHo
V
j d 2 Vo\
dz2
t5g fmT
mz
, . r
dz2
j
J mT
This corresponds to a box normalization. trascioiu [110].
See also Appendix A of E. Gildener and A. Pa
602
CHAPTER 23. Path Integrals and Instantons
On the left hand side we perform a partial integration; on the right hand side we can replace (in the dominant approximation) the integral by that integrated from — oo to oo, and since this is the integral over twice the kinetic energy (and hence S B ( ^ I C ) of Eq. (23.13)) we obtain (remembering the change of variable from T to z) , dipo Vo — dz dj)o\ Wo . i mT dz J . diPo
dz
mT
™2 39 2 '
(23.55a)
In order to be able to evaluate the left hand side we consider W K B approximations of the solutions of the equations of cases (a) and (b) above. The W K B solution of case (a) is (c being a normalization constant and the ZQ occurring here a turning point)
^°( z )
=
~TJl e x p ( ~ /
dz 1/2
v J,
v(z)=:
 6sech 2 z
and
V(ZQ) = 0.
Thus for z —• oo we have v1/2{z) — n and tpo(z) —> 0, as required. The corresponding WKB > solution of case (b) which satisfies its required boundary condition, i.e. vanishing at z = mT, is similarly seen to be
*>(*> =
^
exp
Z**1/2)">(* L
/ dzv1/2\ + exp (  1 I
f,V2 dzv
exp
["dzv1/2
J zn
where v = n 2 — 6sech 2 z — (u> 2 /m 2 ). Differentiating the last expression we obtain d ipo(z) = —cv1'4 exp ( dz and so dz But (see ipo(z) above)
•>Po\ z=mT ^ CV 1/4 ' exp rmT
dzv1/2\
exp ( /
,f, dzv1 / 2
1
4'o\z=mT —  2 c 6 1 / 4 exp
rmT
/ Jz0
dzv1!2
dz
/ J z0
dzv1/2).
Hence (with n2 ~ n 2 — (uig/m2)
for T —> oo) and analogously —^o(—mT) ~ 2 — i>o(—mT). dz dz
dz
•4>o(mT) ~ 2 — ipoljnT), dz
It follows that the left hand side of Eq. (23.55a) is
4>o
But
^0(2)
:
mT dj>o mT ~ 2ip0—ip0 dz dz mT mT 4m 2
g cosh z
^
z
for z —> ±00,
so that 8m* e = F 2 i — ^ 0 ( 2 ) — =F — =F2V>o(z) for z —> ±00. dz g Hence ip0(±mT) — i>0(±mT) az = 2 ^ 0 ( ± m T ) — ^ o ( ± m T ) ~ q=4( — dz \ g
2\ 2 4mT
23.4 Field Fluctuations
Equation (23.55a) therefore becomes
603
if,
\
2 c
4mT „
^0 4m4
/
m2 Sg2
We thus obtain the result
wl z, 9 6 m 2 e  4 m T ,
wo ~ voe~2rnT,
vo • 4\/6«
(23.55b)
We observe that this eigenvalue is positive; the configuration associated with the zero mode is approximated by a classically stable configuration. For T — co the eigenvalue vanishes. Note that > here we used a Euclidean time interval of total length IT. This boundary perturbation method may be repeated for higher eigenstates which are then found to possess Tdependent eigenvalues.
23.4.2
Evaluation of t h e functional integral
It is clear from Eqs. (23.52a) and (23.52c) that the infinite product they contain provides a particular difficulty. It is the evaluation of this quantity that we are concerned with in this section. The highly nontrivial method was devised by Dashen, Hasslacher and Neveu* with reference to other sources. ^ We set Jo := det
drj(Tm
din
/(nIW)^',
^
dr
d£k exp
?£i« n\
2
2 2 Wnp
(23.56)
One introduces the following mapping or socalled Volterra transformation for large but finite T at the lower limit
y(T)
=
V
(
T
)

/
Z(~T)=V(T)
T
^ 0,
(23.57)
where N(T) is the zero mode, i.e.
N{T) =
=
HL
1
.
(23.58)
g cosh m(T — r 0 )
One can then show
fvu
*R. F. Dashen, T Integration in I. M. Gel'fand and and W. T. Martin
— see Examples 23.2, 23.3 below — that Vz } Dry exp 1 \1/2
2ith)
lJTAr2(T)]
1/2 T
[N(T)N(T)\
dr
1/2
/
TN*(T)\
(23.59)
B. Hasslacher and A. Neveu [63]. functional spaces and its application in quantum physics is lucidly described in A. M. Yaglom [107]. In particular this paper utilizes results of R. H. Cameron [43], [44] which are required in our context below.
604
CHAPTER 23. Path Integrals and Instantons
Observe that this expression is independent of the normalization of the zero mode N(T). Also observe that the transformation^ (23.57) transforms effectively to a free particle Lagrangian. After evaluation in Example 23.2 of the functional determinant occurring here, the integration requires the introduction of a Lagrange multiplier which enforces the endpoint conditions, i.e. boundary conditions, on the fluctuation rj{r) and hence on 0(T) itself; this latter calculation is done in Example 23.3. The evaluation of IQ with the above formula is then straightforward. We have for T — oo >
m
V K
'
'
1 g2 cosh4 mT
m 24e
4mT
and
T
dr
i
71 m4 m
rmT
/
JmT mT sinh 4z
32
dz cosh z
m°
+
sinh 2z
3z
mT mT
+—
ill
m 32
5
AmT
(
We observe how the largeT dependence cancels out in In and obtain the simple result In = det dr)(rm)
d£n
allfc
n
2n
w
1/2
m \ 1/2
kP2
(23.60a)
Note the variable T at the upper limit of Eq. (23.57) which implies that the transformation is a Volterra integral equation. This is an important point and should be compared with a Fredholm integral equation which has constants at both limits of integration. Consider a differential equation with second order differential operator M and Green's function K(x,x'), e.g. (M  X)u(x) = f(x), MK(x,x') = 5(xx'), a < x < b,
and with boundary conditions B\ [u] = 0, B2M = 0. The solution u(x) of the differential equation can be written u(x) = h(x) + \f
J a,
K(x,x')u(x')dx'
with
h(x) = f
J a
K(x,x')f(x')dx'.
For f(x) 7^ 0 this integral equation is called an inhomogeneous Fredholm integral equation; for f(x) = 0 the equation is called a homogeneous Fredholm integral equation. If one has an integral equation where K(x,x') is a function of either of the following types, K(x,x') = k(x,x')0(xx') or K(x,x') = k(x,x')0(x'  x),
the inhomogeneous Fredholm equation becomes a Volterra equation, i.e.
rx rb rb
u(x) = h(x) + A /
J a
k(x,x')u(x')dx'
or u(x) = h{x) + A /
J x
k(x,x')u(x')dx'.
In the book of B. Felsager [91] this transformation can be found in his Eq. (5.42), p. 190.
23.4 Field Fluctuations
605
Inserting this result into Eq. (23.52c) we obtain for the kernel h the simple expression ' m\ ' 4 m3 (23.60b) irh) 3 hg2 We see that this result is nowhere near to allowing a comparison with our earlier result Eq. (23.35a) for the determination of AE. The reason is that the zero mode has not really been removed, so that all fluctuations, i.e. perturbations, are still present. This has to be changed and is achieved with the FaddeevPopov constraint insertion that we deal with in the next subsection. Example 23.2: Evaluation of the functional determinant
Apply the shift transformation (23.57) to the fluctuation integral (23.43) and evaluate the functional determinant occurring in Eq. (23.59), i.e.
Thj
Solution: We first derive the inverse of the transformation (23.57) by differentiating this which yields
JV(r)
We can rewrite this equation as
Z(T) N(T)
f,{r)
N(T)
N{r)
N2(T)
T)(T)
:
d ( r](r) dr\N(T)
Thus
?j(r) r
N(T)
i= r
J^
T
dr '
Z(T) N(T)
' hand side of this equation yields
When multiplied by N(T),
r) Jj partial integration of the right
r)JV(r) I" K ' ' JT
V(T)
dr'—(—^—)z<T'), dr'\N{T')) y h N(r)£ ,
dr' N(T') 'V \Z{T').
r,{r)
=
z(r) + f(r),
f(r) =
(23.61)
For further manipulations we observe that with Z2(T) (obtained by squaring the expression of the first equation above) we have by differentiating out the following expression:
\ i, M^
dr~
2/ ^NM
JV(r) r,2(r)V"(0c)
,
N
, ,JV(T)
N(T)
, , ,N2M
'N2(T)'
so that since N(r) — V"{tj>c)N{r) = 0 (by definition N(T) being the zero mode)
V2{T)
+ V"(0c)r/2(r) = z2(r) +
^[^(r)
Mr)'
N(T)
606
CHAPTER 23. Path Integrals and Instantons
The derivative term will from now on be ignored since when inserted into the action integral and integrated it yields zero on account of the vanishing of the fluctuation r)(r) at the endpoints T = r 0 =  T , T. Thus 5S,
i
r1
dr
+ v"(<pcW
£
dr
z2(r) 2 v
;
The functional integral to be evaluated is To of Eq. (23.48b) which excludes the classical factor. Thus
h
/^ } exp[/; /T {i(J) 2 + i^(, c )}
jv{z}
D{V} Dr]
f.A\i2{T)]
D{z},
Dz Dr] Dr] Dz
T>z Dr]
Our next step is the evaluation of the functional determinant. Prom Eq. (23.57) we deduce
dr]{r") ) 7
^L=S(TT")+
(T
K(T,T')5(T'
T")CLT'
with
K(T,T')
.
JT
N{r'Y
(23.62)
where we allow for a general r dependence in the Volterra kernel as well. Discretizing the Volterra equation (23.57), i.e. Z(T) = r,(r) + fT K(T,r')r,(T')dT',
or more generally an integral equation of Fredholm type, by setting 5 = Ti+i — T;, i = 0 , 1 , . . . , n with TO = —T,rn — T, we can write the equation
z r
{ n)
= ^2
finiVin)
+
5
5Z
K T
( n>Ti)v(.Ti)
(23.63a)
We observe that K(rn,Ti) = 0 for % > n. The discretization here is a very delicate point. A different and convenient — but not unique — discretization (discussed after Eq. (23.66)) is to rewrite the equation as
n n ..
z T
( n)
=
^2 Snir}(Ti) + 5^2
n r n
K T T
( ^ i)~{v(n)
+
v(ni)}
cn—i
] T SnrtTi)
+  £
K(r,Ti)f,(Ti)
+ ~Y,
K{r,Ti+1)r]{n).
(23.63b)
Equation (23.63a) can be rewritten in the matrix form of a set of simultaneous linear equations: / z(n) \
2(T2)
/
i + i^Cn.Ti)
5K(T2,TI) 5K(T3,T1)
o
1 + <5K(T 2 , SK(T3,T2)
T2)
z( 3)
T
1 + SK(T3,T3)
0 1 +
0 0 0 0
6K(T„,T„)
V{T3)
V 2(r„) /
V
*See also I. M. Gel'fand and A. M. Yaglom [107], after their Eq. (3.4).
TI) K(T2. n )d? (23.23. .e. 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. . Felsager [91].N{T) Finally we add the following remarks concerning the discretization (23.T1)dT1 + J dn J dr2.. i. Cameron and W.64) as above and consider the determinant of the expression.64) Inserting the explicit expression for the kernel given in Eq. T. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. ^This is the method explained by B.n) K(Tj. Thus the factor a inserted above has to be taken as 1/2. r i ) + .n) i=l N lim exp £*(T. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). [44].Ti) i=l : exp \ £ K(T.68) D above. In the limit S —* 0 . r ' ) is to be evaluated at r = T ' . We observe that if we discretize Eq. the value of the determinant is half (or arithmetic mean value) along the diagonal. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) ..64) D (23. as in Eq. are summed over all values from 1 to n. 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.66) •Dr) .lnJV(T)}] N(T) N(T) (23. whereas the determinant arises only once in the original determinant Dn. we obtain exactly the nonvanishing terms on the right of Eq. (23.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. (23.63b).. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1.T2) I exp I K(n. (23. the kernel K(r. p. where a f 1 See R. 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.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. H. Martin [43]. Thus D„ = l + 5 ^ K ( r i . so that (with T *T) 1/2 •Dz N(T) (23.exp • /_:T JV(T')" dr' = e x p [ . X(TI.a { l n i V ( r ) .TI) 0 K(T2. 2! .67) i=l v ) = lim exp V In ( 1 + ' L K(T.7 Y.n) o K(TJ. 192.T')dT' (23.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .64).63a).n) z=l x (23.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits. (23. = 1 52 K(n.j.. Thus in our case one of the kernels is zero and we have to decide what we do when.
72) The endpoint integration dz(T) = dz(rn+i) can be combined with this. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23. (21. f(T) = N{T) J^ dr'^^z(r') (23.61) we obtain the constraint on the function Z(T) which results from the endpoint constraint JJ(T^) = rj(T) = 0 on the fluctuations T){T). In = dzi dz2 • • • dzn e x p . .3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23.— .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . N(T') . (23.N(T) . Eq. T + N(T) / T' = T JT dr' N(T') . and hence may be ignored. (23.71) Hence (with h = 1) l = j . (23. In every integration / dzi the contribution €7.0.57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . With partial integration (in the second step) we obtain (since (cf. rn+i = T. Briefly. Evaluating Ii.e—*0 In.59)) + f(T)] = — f dz(T) f°° dae^W+f™.J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ .l + £7t) «i (23. n J=. Eq. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). represents a constant translation. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. (23.73) .70) and this now requires also integration over the endpoint coordinate z(T). Solution: From Eq.2o) 2 rc^O.( . I2 and using induction one obtains (cf.^ .Z i .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 21.59) the endpoint conditions »)(±T) = 0 and evaluate the integral. Path Integrals and Instantons Example 23.Z ( T ).69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf.(T) + /_ dr — . (23.608 CHAPTER 23. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo.
(21. Popov [89]. / d£o <9r0 5[F(TQ. i." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function.^ 23. / — / / We started off with V(T) = ^CnVniT)..59) and agrees with that in the literature.23.4. 11 L.e.£Q) — aTo]dTod£o.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode. .74) where we reinserted h. (23. 1/2 T W 2 (r)J (23. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. (3. (23. Eq. ^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.12). Faddeev and V.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. Patrascioiu [110].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. j / dTT]n(T)r]m(l r . D. a = const. Ignoring the remaining metric factor with the reasoning explained > between Eqs. Gildener and A.10) and (21. J co § dr Recall that / f ^ doce^ = yfVfp.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. N. This result verifies Eq.11). £n = ~2 dTT](T)7]n(l (23.
i. in the direction of the zero mode.79) (23. (23. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. (23. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).J drm(r)Mr + T0) (23.75) requires.^O) with dependence on ro which the replacement (23. e.83) **The collective coordinate in the present context is also called "instanton time".e. (23.£Q) — aro] contained in Eq. (23.g. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) . .75) enforces the collective coordinate ro to be given by F(ro.81) 2 / dr»7o(r)0c(r) = .78) <t>ij) = Y. where L = in + cn. i /*oo (23.80) i r or •i /*oo dr7?o(r)0(r) = 0.610 this means that CHAPTER 23. (23. (23.77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration.£o)/a. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r).^n{r). at r = ro.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T).e. Thus. The delta function 8[F(TQ. Path Integrals and Instantons _^ ^ w .81) the rdependence is integrated out and hence the integral does not provide the function F(TO. this constraint implies that we go to the subspace which is orthogonal to the zero mode. We establish the function ^(To>£o) with the following reasoning. / dTm{r)n{T). In the integral of Eq.e. The number £o is the component of the fluctuation rj(r).76) we set 4>c{r) = ^CnVnir). i. We consider the following integral / ( r 0 ) := . Then cn = ~2 / dr(pc(T)rjn(T). In analogy to (23. We now impose (i. demand validity of) the constraint £o = 0.
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.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 ) .4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r)..y/S^ro] (23. .84) (23. .75) 1 = AFP f dT06[{F{rQ.85) PJ W i t h Eq. (23. (23.^) ./ drri0(T)<pc(T) + PJ (23. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.y/SET0 = 0. 158.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.£O) OTQ = = £ " . (23.91) " F o r more discussion of this point and related aspects see also B.23.88) ro0 Then in Eq.86) = Jdrrjoir^Mr F(TO.87) l/M) (23. We evaluate this quantity as follows in the leading approximation. so t h a t with F(TQ. Felsager [91]. Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) .<t>c{r)} . We have AFP = /F(T0.89) Here the factor App is called the FaddeevPopov determinant. p.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.(23.
/ dT0AFP. /(TO) ^/SE~T0} V~SET0 c d£. P J AT (23.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. =~ (23. P the relative sign being of no significance. 770(7"). fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. 7 Thus there is no fluctuation along the direction of the zero mode in the sum.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 .612 CHAPTER 23. i. .90). using the relation (23.91) into Eq. Consider. if ^ + d£o = 0.93) where AT = TT — 00.93).0AFP8 4o H P .94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.e. i.e.e. the "static energy" invariant. ~ There is a quick way to obtain the result (23. tangential to (J)C(T)) leaves this and hence the square of this quantity.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. Inserting (23. i. 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. in d(j)c(To) + dr/fa). (23. 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. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. if the coefficient of 770(70) vanishes.e. 0 d r (23.
(23. (23. We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton. We determine I'Q by observing t h a t this differs from IQ of Eq.98) .A T A F P i " o e x p T—>oo p where I'o n^O. (23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.4 Field 23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. Since w\ is for finite T given by WQ of Eq.93).4.23.52a) and replace in this J d£o by the expression (23.96) does not yet allow a comparison with (23. (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.55b).96) lim . We observe t h a t the result (23.60a) for I0 and Eq. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.97) is a constant still to be evaluated.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.43)).60a) in not involving the zero mode contributions. This very important intermediate result is the contribution of a single instanton or kink configuration to the p a t h integral (ii has been defined by (23. (23. 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.
Gildener and A. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 . Gildener and A. 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. ^See E.100) We observe that the dependence on the normalization constant p cancels out. Although we choose p2 = 1. i. (23.i n s t a n t o n contributions As already mentioned. (3. their Eq.e. Patrascioiu [110].55b)) w0 = v0e VQ 23. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 .32).614 CHAPTER 23.2 m T e x p 2 irgh ' T»oo (23. ~ rKo')e{Tx .a n t i .101) 2T > oo (cf. the approximation is then known as the "dilute gas approximation".20).4>C exp h (23. (23. they have to introduce the inverse of that factor on the far right of their expression (3. *For the validity of the calculation the instantons or kinks have to be far apart. In view of their normalization of the normal modes. we drag the factor p2 along for ease of comparison with literature.* Finally we rewrite the expression (23.102) *For comparison with the parameters of E. 0c M.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. .5 I n s t a n t o n .100) in such a way that it exhibits the crucial factors of which it is composed.r). Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.4. This means we are concerned with configurations like* 2n+l d(r . the result becomes h lim 2 m r ^ ^ 1 e .TiMiy+'Mr «>.
(n = 0 implies the single instanton solution).4>i.fo) = Id2 St SE(<Pf. are varied.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. (23. here (f>2. .106b) SE{<t>l.Tl\(j)i.6.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . T2 = T/3.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.H)2 + . The factor ( .r/l&.r 0 ( 1 ) = .104) Considering the case n — 1 of a kink with one additional kinkantikink pair.<t>ff + h .<t>2)eSE{<t>2. T3 = T) d(f>l((pf.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left). 23. 1 0 6 a ) and SE{(J>I. Since these endpoint configurations.Tf\(p2.T2}((t)2. ( 2 3 . 2 .<j>i) . Proceeding as before we now consider first in analogy with Eq. r 0 ( 2 ) = 0.4>l)eSE{4>l. r0(3) = 2T/3. Consider (with T 0 = T.<Pi)2 + • • • .Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f. (23. (23. .36) (/. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.(l>i) (23. 7 i = .4 Field Fluctuations 615 for n = 1. Taking these configurations into account.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.r / 3 . this means we have to consider the amplitude (23. (3) (23. the total Feynman amplitude is given by (^/.4>I) ld2SE SE(c/>i.2 T / 3 . .23.Tl}{4>l.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.T2\(f)l.l ) i + 1 in the sum of Eq.
e. of course.4>i) { 2 " 2d SE . dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 . It has to be remembered.2J.108) where the last expression is obtained as follows.. i.e. the .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). The allowed positions of the individual kinks have to be such that their order is maintained.616 CHAPTER 23. in the case of l2n+i therefore ry2n.2d2SE ~c 4>i d(f)i e x p SE^f. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. i. ±m/g (23.3 9 L 3m 2 and 9SE((/)) ± . It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.e. (23. each of which implies a multiplicative factor 7. the dilute gas approximation. d2SE{4>) d(j)2 2m. On the assumption that the kinks of the instantons are widely separated. i.(01 . each of the three intermediate kernels contained in Eq.4>i) / d(f>2 e x p {<h .12) we have JM~°°) i. that each applies only to a fraction of the overall Euclidean time interval of r = 2T.e. (23.106b) is to be substituted into the second relation of (23. (23. Path Integrals and Instantons where the first expression in Eq. With our remark after Eq.105) can in other respects be treated in analogy with the single instanton amplitude.4>f) d4>\ 4>s72 (23.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations. rn ^2 J2H 9 1m' ±2gcj).
112) We now compare this result with expression (23. In the case of (<^f. 23.101): (^/.111) Finally we sum over any number of instantonantiinstanton pair configurations and obtain for the Feynman amplitude oo 1 oo 1 ( {JATAFPIO 2n+l I = lim V n=0 J 2 n + i = lim . Gildener and A.105) t h a t each of the three amplitudes contributes a factor e x p [ .Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.e. T ] with ] T 3 = 1 A » T = A T = 2T.93) has been applied. then sign reversals TQ i ) .^ .2 T / 3 this product becomes that given by E.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23.m A . (3) ° JrW: 2T/3 (2T)3 .(*) o ax = 2 T / 3 . i.r. a 2 0. Am3 2m J rW ( r ( 0 and this followed by setting aj.7 3 ^ — .(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23.a 3 [110].110) Finally we have to recall from Eq. (23.4 Field Fluctuations 617 replacement (23.V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.35a) which contains AE in the argument of sinh and obtain for the level splitting A J resulting from consideration of the instanton plus any number of instantonantiinstanton pairs A*£ = 2n7AFP/0^=exp 2TT 1/2 4 m3 m V2 / m \ 1/2 4 v / g ^ exp ( 2TT irh _±rrv^ 2h ^With shifts r.)(3) 1 o A T 3 Ao lim .6 the collective (i) coordinate TQ has to be to the right of TQ . Patrascioiu JT/3 ° ir< >2T/3 1 dr. and TQ to the right of TQ This implies t h a t in this example we have the product of integrals^ rp (3) (2) .r/^. we now have to integrate over three collective coordinates.23. and in the case depicted in Fig.T/0j.
AE. here as 1) is also taken into account. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . (23. Liang and H.W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0). The difference between levels. MiillerKirsten [163]. l CHAPTER 23.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S. 23. J. . explicit quantization). To help the comparison we note that — with indices indicating the authors. GP for those of E. K. Finally we realize that the computation of (23. even without explicit specification of the prefactor. has been obtained without explicit use of canonical commutation relations (i.182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — . W. In fact the equivalent to writing down canonical commutation relations is the evaluation of the Gaussian integrals which enter the prefactor of the expression in Eq.Q. J. J. J. Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton). W.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs. — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92.' We see therefore that the classical Euclidean configurations 4>c can be considered to be responsible for the quantum mechanical tunneling which gives rise to the level splitting AE.e. We see that the method — above we included a lot of highly nontrivial details which are passed over in other literature — is considerably more complicated than the method of the Schrodinger equation. Patrascioiu [110].113). (23. Bose and H.113) for qo = 1: A1E „ . .618 and hence AE A.113) is very complicated.^r . Gildener and A. by either replacing A T by 2AT or doubling A*. MiillerKirsten [30].5 Concluding Remarks In this chapter we used the path integral to obtain the level splitting of the ground state of a particle in the double well potential. Wiedemann [4]. We observe one more aspect of our calculation. ^See P. (18. we obtain AE = 2A* and agreement with the result (18. W. Achuthan. MiillerKirsten and A.With this correspondence A\E of Eq. H.
We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues.1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1.Chapter 24 P a t h Integrals and Bounces on a Line 24.2) . ™=£('^° < 2 «> Here the two symmetrical minima of the potential are called vacua (classical or perturbation theory vacua as they are sometimes called because ordinary RayleighSchrodinger perturbation theory would be performed around these points). i.a 2 ) 2 + c^T2.e. and their interpretation and uses. Previously we considered in particular the quartic or symmetric double well potential for a scalar cp. however. Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 . Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling). 619 (24. In order to really appreciate the difference between topological and nontopological configurations (the solitons and instantons we considered previously being of the first kind) we begin with a recapitulation of some crucial aspects of the former. in a somewhat different formulation.
2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a.1 The asymmetric double well. Path Integrals and Bounces on R1 Here dVT{4>) vanishes for . = 4(ACT2 cT 2 ). We see that <fi = 0 minimizes VT(</>) for temperatures T > TC.e. a spontaneous symmetry breaking leads to the double well (i.e.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 .620 CHAPTER 24. > / V[<H 'true'vacuum 'false'vacuum Fig. but implies instability for T < Tc. its symmetry under . at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) .a2)4> + 2cTz 2 0. J 4. 24. We can destroy the reflection symmetry of V((f)) (i. A2 _ a — \„2\ u ) ~r u/\y> 2 (24. T = 2\(<f>2 .A a 2 ) . Thus if we start with the system (like the early universe) at a high temperature and then allow the system to cool.
see e. V(<j> = m/g) = 2e. . perturbationtheory vacuum).g. False vacua play an important role in models of the inflationary universe* and elsewhere. The Lagrangian density of the theory is given by C = dp*d»*U(*).^J >o.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e.) . the vacuum being understood as the classical.24.t In the following we consider first (after some recapitulations) a simple model of a false vacuum. We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions. which (b) are saddle point configurations of E (or SE). Penedones [59] and references cited there. S. a linear term as in w^^'A1^'*0V(<l> = m/g) = 0.5) We see that one minimum of V now lies higher than the other.1 with (24.7) *See e. (24. for instance. the quantum mechanical vacuum).g. the higher minimum is called a "false vacuum". A. L.Y. . and (c) determine the lowering of the energy to the true quantum mechanical ground state (i. H. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE). Pi [126]. Cornalba. also in contexts of string theory. <44 2) ' This is now an asymmetric double well potential as in Fig. We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory. T False vacua and bounces occur. 24. Costa and J.g. M. tf(*) =^ 1 .e. Since the true vacuum minimizes the potential. (24. We calculate the classical configuration — which in this case is called a "bounce" — which corresponds to the soliton or instanton in our previous chapters (but is not topological) and we use this in order to calculate the lifetime of the unstable state (or equivalently the imaginary part of the energy). On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. Guth and S.
8) This equation possesses the kink solution <t>c{x) m tanhm(a. (24. we see that (24. small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention).e. Path Integrals and Bounces on R1 Static fields given by d$(x. which is an important point as we shall see. — XQ) (24.12) Since <fic(x) solves Eq. (24.e.13) For small w\ > 0. so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). We observe that this kink solution is odd in (x — XQ). We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional. Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x). i.t) dt 0 are written (f>(x). t) = <f>c(x) + Y^ iPk{x)ekt (24. dx2 + u"(<j>) 5(x y). The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c.7) yields the following equation called stability or small fluctuation equation. (24. i.622 CHAPTER 24.9) 9 for the doublewell potential given above. (24.8).10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24.14) .
eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue.1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite.e.e. 3m 2 and w\ > Am2.2 The inverted potential.24.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24. .e. We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. the other a positive quantity) and a continuum which are given respectively by w\ — 0. thus the zero mode here is given by d . 24. configuration (for our </>4theory).oo \ x== + CX) Fig. d \m . 22. the zero mode). underneath the particle's velocity. i.8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero. o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = . (24. i.16) V[<H x= oo 0 d2 . In the case of the double well potential (24. .15) cosh mx From our study of the solutions of this equation in Sec. i. kink. 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24.17) .
Below we shall see that the configuration described as a bounce is given by an even function of x. F (24. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function.21) The charge is defined as the space integral of the time component of the current. (24. Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function. so that its derivative representing the pseudoparticle's velocity is odd. In this case the usual Newton equation reappears with an opposite sign of the potential. We define the current k» := e^dvfc. which has the solution 4>c(x).2.e. the associated zero mode. 4 TTT. 4>"{x) = U'(4>).20) The number 1 on the right is the socalled topological charge which is obtained as follows. Integrating the classical equation (24.19) E((f>c) > / J00 dxV2U(f)'c(x x0) = —x &9 1. (24.8).624 CHAPTER 24. Then the energy satisfies f°° . 24.e.3 (24. i. The classical configuration is then called a pseudoparticle configuration or instanton. i. with Euclidean time. i. where it is then again at rest. i.18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U. The kink solution can now be looked at as the classical solution of a onedimensional field theory. and — clearly — we can resort to a classical interpretation of this function. In the pseudoparticle consideration the coordinate x plays the role of time. Path Integrals and Bounces on R1 We observe that this zero mode is an even function of (x — XQ) in agreement with our expectations for the ground state wave function. Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. quantum mechanics. We see that this pseudoparticle starts from rest at one peak of — V at x = — oo and in losing potential energy gains just enough kinetic energy to reach the second peak at x = +00.e. The associated zero mode represents the velocity of the pseudoparticle as indicated in Fig. by Q = \(°° dxf^eoid1^) = l[M°°) ~ M°°)] = 1(2422) .e.
T. H. Coleman [41]. See also S. W. We call this topological stability.11) of this reference the number 35 of 8/35 is misprinting of 3. The requirement of finite E therefore does not permit \<j)c deformations.A2)2 ^ 0 except for A2 = 1. [57] and C.A W + £/(A0C (24. MullerKirsten [166]. We see already here that a configuration which has the same value at both endpoints has Q — 0 and so is nontopological (this will be seen to be the case for a bounce). the kink solution having a monotonic behaviour on the way from one endpoint to the other.5. (4. [19].23) J —( In this expression we have the potential with the behaviour (see Eq. J. Our main objective here is in the first place the recalculation of the result of Bender and Wu* for the decay rate of a particle trapped in the ground state of an inverted double well potential by means of the path integral method.4 and 6. It is important here to realize that the nonzero value of Q results from the different boundary conditions of the classical configuration at +00 and —00. We recall finally that the onedimensional theory with the double well potential allows the calculation of the difference between the ground state energy and that of the first excited state.24) Thus the energy is finite only for A = ± 1 . of course.5 = 15. W. The theory does. possess other solutions which have Q — 0. In Eq. We now turn to corresponding considerations for a potential which allows tunneling to infinity and hence the definition of a particle current. Bender and T.• ±00 : U(X(pc) = ^ ( 1 . . 2. T