•
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
we recapitulate the origin of quantum mechanics. With various techniques and deeper studies.e. in fact. be treated to a considerable degree of satisfaction perturbatively. 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. one problem being that there are few nontrivial models which permit a deeper insight into their connection. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. with degenerate vacua) were first and more easily derived from the Schrodinger equation.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. In the first part. screened Coulomb potentials and maybe singular potentials. but arise also in recently studied models of large spins. Our approach to quantum mechanics is through a passage from the Poisson algebra of classical Hamiltonian mechanics to the canonical commutator algebra of quantum mechanics which permits the introduction of Heisenberg and Schrodinger pictures already on the classical level with the help of canonical transformations. the path integral method of Feynman was soon recognized to offer frequently a more general procedure of enforcing first quantization instead of the Schrodinger equation. Thus this first part . but also about complex energies that he encounters in a parallel course on nuclear physics. again point the way: For scattering problems the path integral seems particularly convenient. in Feynman diagrams in quantum electrodynamics). However. like periodic potentials. These basic cases will be dealt with in detail by both methods in this text. whereas for the calculation of discrete eigenvalues the Schrodinger equation. that is the Coulomb potential and the harmonic oscillator. has not always been understood well. numerous problems could. even though the expansions are mostly asymptotic. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. Excluding spin. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. Chapters 1 to 14. its mathematical foundations. the aforementioned exactly solvable cases. With the growing importance of models in statistical mechanics and in field theory. basic postulates and standard applications. Then the Schrodinger equation is introduced and the two main exactly solvable cases of harmonic oscillator and Coulomb potentials are treated in detail since these form the basis of much of what follows.g. To what extent the two methods are actually equivalent.
This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. and the elliptic or Lame potential — here introduced earlier as a generaliza . and their eigenvalues. 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).XVII deals mainly with standard quantum mechanics although we do not dwell here on a large number of other aspects which are treated in detail in the longestablished and wellknown textbooks. The 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. 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. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). periodic instantons. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. we deal mostly with applications depending on perturbation theory. The most prominent examples here are the double well potential and its inverted form. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). The solution of this case — however in nonperiodic form — turns out to be a prerequisite for the complete solution of the Schrodinger equation for the singular potential 1/r 4 in Chapter 19. In 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. Thereafter the concepts of instantons.e. and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. i. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. The potentials with degenerate minima will be seen to reappear throughout the text. Using perturbation theory. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. After a treatment of power potentials. the chapter thereafter deals with Yukawa potentials. In the second part. which is presumably the only such singular case permitting complete solution and was achieved only recently. Chapters 15 to 20.
except that these are no longer of hypergeometric type. cubic). Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. and whenever available also with the results of WKB calculations.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. This puts Schrodinger equations with e. quartic. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. 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. 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. for instance. 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. from our precise reference to unavoidable elliptic integrals taken from Tables). such as limiting procedures. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. 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. This method is therefore more complicated. anharmonic oscillator potentials on a comparable level with. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. All results are compared with those obtained by perturbation theory. it is. for instance. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. we can make the following observations. the Mathieu equation. With a fully systematic perturbation method and with applied boundary conditions. this comparison on a transparent level being one of the main aims of this text. however. 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. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. an approximation whose higher . The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. We then illustrate the relevance of this in the method of collective coordinates. In fact.g.
as well as applications and illustrations. As a rule. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. K. are relegated to separate subsections which — lacking a better name — we refer to as Examples. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). . shows him that each such formula here has been properly looked up). Nonetheless. in particular Professors J. 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. at the end of a chapter (after troublesome turning of pages). with the source given in the bibliography at the end. thereafter University of St.g. e. T.Q. we also consider at various points of the text comparisons with WKB approximations. N.g. like E. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). Whittaker and Watson [283]. For ease of reading. also for the verification of results. Throughout the text some calculations which require special attention. Andrews). 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. 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. Dingle (then University of Western Australia. The line of thinking underlying this text grew out of the author's association with Professor R. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. Watson [283]. The author is deeply indebted to his onetime supervisor Professor R. 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. Their deep involvement in the attempt described here is evident from the cited bibliography. B. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. Instead a glance at a nearby footnote provides the reader immediately the names of authors. not the least for permitting a more detailed and hopefully comprehensible presentation here. W. Whittaker and G. Park (Masan). Since this comparison was the guideline in writing the text. D.g. Liang (Taiyuan). MiillerKirsten *In the running text references are cited like e. B. J.* H. H. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. D.
1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900.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 . Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. The best approximation to such a body is a cavity with a tiny opening (of solid angle d£l) and whose inside walls provide a diffuse distribution of the radiation entering through the hole with the intensity of the incoming ray decreasing rapidly after a few reflections from the walls. We do not enter here into detailed considerations of Planck. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. very few explain in this context what he really did in view of involvement with statistical mechanics. Instead.Chapter 1 Introduction 1. the amount of radiation absorbed by a body is equal to the amount the body 1 . which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy). 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). Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. Kirchhoff's law in thermodynamics says that in the case of equilibrium.
h = 6. In Eq. i. radiators.. T)du is the mean energy density (i.T) = dv3eC2U/T. A being the wavelength of the radiation. the formula (to be explained) u(u. These expressions are: (1) Wien's law.626 x 10 .e.3 4 J s.1 Absorption in a cavity. 1.e. Let us look at the final result of Planck. energy per unit volume) of the radiation (i. kT kXT (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.2) and the . How did Planck arrive at the expression (1. The parameters k and h are the constants of Boltzmann and Planck: k = 1.38 x 1(T 23 J K'1. Introduction emits. Fig.2 CHAPTER 1.l ) where x = ^ = ^ . and then measuring the increase in temperature of the heat bath. y J c 3 \ex . (1. u(u. i.1) ' y Here u(v.T) = 2*?£(?)kT. v + dv in equilibrium with the black body at temperature T. Black bodies as good absorbers are therefore also good emitters. (1.1) c is the velocity of light with c = u\. It was found that one expression agreed well with observations in the region of small A (or AT). T) were known and tested experimentally. and the other in the region of large A (or AT).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.e. 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.
he searched for a derivation. 3 (1.3) are contained in Eq. e xhv. C2.2) and (1.T) 2^^kT.1) as approximations. that the formulas (1. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. and thus over a discrete number of admissible states. (1. 1. which are indistinguishable) among the N indistinguishable oscillators at . "small" (i. (x large). T) u(i/. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. but — here the discreteness appears properly — only in elements (quanta) e.3) Ci. We see.2 Distributing quanta (dots) among oscillators (boxes).47TZ/ 2 {x small).1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>.T) = av e6"/Ti' where a and b are constants. C3 being constants. Indeed. and he had succeeded in finding such an expression of the form u(v. When Planck had found this expression.e. where the Fig. exp(x) ~ 1+x) and "large" (exp(—x) < 1). in the first place Planck had tried to find an expression linking both. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons".1) in regions of a. This is the point. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. (1. Here W is a number which determines the distribution of the energy among a discrete number of objects. To this end he considered Boltzmann's formula S — klnW for the entropy S.1. we obtain: u(i/. discretization begins to enter. . Considering Eq.T) = 2^C3T.
as in most other Tables. (1. 940. (1.e. Magnus and F. We now obtain the energy density of the radiation. W. u(i>.2. p. h = const.1)!P! (1. e = his. Gradshteyn and I. e. M. v + dv. S. there not called Stirling's formula. with the quanta represented schematically by dots as indicated in Fig. Agreement with Eq.e. formula 8. The Stirling formula or approximation will appear frequently in later chapters. N * oo.6) Fig. 1. . U{T) being the average energy emitted by one oscillator. . one obtains (cf.4) IniV! ~ JVlniViV + O(0). i. and the second law of thermodynamics ((dS/dU)v Example 1. = 1/T).1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.T)dv.g.3 Comparing the polarization modes with those of a 2dimensional oscillator. Ryzhik [122].4 CHAPTER 1. i. with riydu — 2 x —w—dv. 1. p. Oberhettinger [181].343(2). (1. We visualize the iV oscillators as boxes separated by N — 1 walls.7) *See e. Introduction temperature T. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v.2) requires that e ex is.3. Then W is given by w = (N With the help of Stirling's formula* {N + piy.g. as for small values of e). I.
i. so that . We obtain therefore u^T) = Unv = 2^fJ^—i.T) has a maximum which follows from du/dX = 0 (with c = vX)...965 and xmin = 0. (1.UJ = 2KV. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. v + dv.3.T)dX = ^ehc/*kT_idX.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.7).3. dj\l dM . 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. so that the derivative of u implies (x as in Eq. Then L^j = nrii. as in electrodynamics.V 2 ) E = 0.3. KT = I J . L for i = 1. is given by . so that (lvL\A 0 I I = rr. as indicated in Fig. The number with frequency v in the interval v. We obtain the expression (1.1. .1)) The solutions of this equation are ^max = 4. (1. We observe that u(v.. The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.e. (1.. .2.1).8) This is Planck's formula (1.1..2. nvdv per unit volume. rii = 1.7) for instance.i = 1. '''From the equation I \ JW .UJ2/C2 + K? = 0.3 (as for ideal conductors). 2 2 2 r2 L K — 7T n .. In terms of A we have u(X. where^ 2 [2KUY .2.
This is Wien's displacement law.—v <L9) We observe that for T — 0 (i. (1. which can assume the discrete energies en — nhv. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. However. which can not be eliminated without a different approach. Thus we have a rather complicated system here. of course. .l. the behaviour of the system at zero absolute temperature. which had also been known before Planck's discovery.6 The first value yields AmaxT = CHAPTER 1.1. One expects. A. such a procedure leads to contradictions.10). and from which the constant h can be determined from the known value of k. .2 (1. we then have at T = 0 independent oscillators.e.965K = Const. We are not dealing with the linear harmonic oscillator familiar from mechanics here. . 2 . If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu.10) Thus here the socalled zero point energy appears. 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 .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=—. but one can expect an analogy. Lorentz and Planck that Eq. One might suppose now. Since temperature originates through contact with other oscillators. Later it was realized by H. n = 0. . ra = 0. . i. We therefore examine such contradictions next. x — oo) the mean energy vanishes (0 < U < * > oo). which did not arise in Planck's consideration of 1900.8) could be derived much more easily in the context of statistical mechanics. Introduction he 4. that of an oscillation system at absolute temperature T ^ 0. that it is easier to consider first the case of T = 0.
around 1926. d . kT/2. i. with Heisenberg's discovery of the uncertainty relation.In . . 1.kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.1 ' u~ e .= ..k dU 1+U\ f ln(l T + U\  U U k ( e . we obtain T V S =fc[ln(TV+ P .1)! . so that 1 f ds\ T — \dUjy . {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) .i n .1.InP!] ~ kN The second law of thermodynamics says 1+ 7 ln 1 + ( 7)~7ln7 \au)v For a single oscillator the entropy is s = S/N. We "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".e.ln(TV .2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later. where k is Boltzmann's constant and W is the number of times P indistinguishable elements of energy e can be distributed among T V indistinguishable oscillators.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .1: Mean energy of an oscillator In Boltzmann's statistical mechanics the entropy S is given by the following expression (which we cite here with no further explanation) S = fcln W. In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").2 Contradicting Discretization: Uncertainties 7 Example 1.1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. for 2 degrees of freedom.1)! .
8
CHAPTER 1. Introduction
shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = mx2 +  w V .
ZJ
(1.11)
Zi
The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A
A A
^ =\[Ef+l
(i i2)

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

(1.17a)
In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at
1.2 Contradicting Discretization:
Uncertainties
11
any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 Any'=
(1.17a) 2TT/L
since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)plane, we can write more precisely ApAq > 2irh. (1.18)
This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.
Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by
12 the limits 0 < q < q + Aq, 0<p<p
CHAPTER 1. Introduction
+ Ap,
what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is reinforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.
1.3 ParticleWave Dualism
13
the number of such photoelectrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particlelike, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few pointlike spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particlelike nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum
The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f
L. de Broglie [39].
14
CHAPTER
1.
Introduction
1.4
ParticleWave Dualism and Uncertainties
We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a doubleslit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.
^y
Ax
Fig. 1.7 Schematic arrangement of the singleslit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].
1.4 ParticleWave Dualism and Uncertainties
15
figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.
Si
t
Ax
\
»v
"^^J
B ^ ^
*
w
p3
Q
Fig. 1.8 The wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)
Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — rexp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T
5 = 2a =
lim NSN = — A a ; s i n 0 .
(1.21)
If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.
16
The intensity at the point Q is therefore
CHAPTER
1. Introduction
h = h
where from Eq. (1.21)
•K ,
.
a = flisinB = Aisint A 2
.
k
Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is
Ie=Io
sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2
(1.23)
The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 )  , i.e. for A x sin0 = (2n + 1 )  = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.
A
(1.24a)
(1.24b)
The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the singleslit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave Dualism and Uncertainties
17
Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or
1.4.1
Further thought experiments
Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light
Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it
18
CHAPTER 1. Introduction
is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~±. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have L = rp' = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.
Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.
1.5 The Complementarity
Primciple
19
quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by
AE « T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax
vG
.
It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of Ai£.
1.5
Bohr's Complementarity Principle
Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The socalled Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a doubleslit (this being the apparatus in this case) the photons or
20
CHAPTER 1. Introduction
electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wavelike or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*
1.6
Further Examples
Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with timedependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —yEo, w o (1.28)
where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui
u> = u>o a,t t = 0, w = ui at t = T,
.
_
i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w mx • 2\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt
dt The energy of the oscillator is l „ E — mx1 2
l 0 , z l 29 + mu} (t)x , y 2 '
so that
dE 1 — = mxz dt 2
9
dJ1 > 0, dt ~
(1.29) v '
where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain  [mx + mur (t)x\ = 0, i.e.
1 1 1 — —mx 2 H—mx 2 dt u22 2
1 mx 2
2d
— . 2 dt\u)
"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].
1.6 Farther
Examples
21
d ( E\ 1 , d / 1 \ — — ) = mx2 — ( — = dt\uJ2J 2 dt\u>2)
mx2 dw < 0, UJ3 dt ~
(1.30) v
where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, ie. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E
E dw2 < 0, UJ4 dt ~
i.e.
1 dE 1 dw2 < —2 . E dt ~ u) dt
(1.31)
JEo
or
E  Jui
u, '
> °
"o
E0 ~ UJ22
' ^ u'2 — < —
E'
<
^EQ.
Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with
e + wgeo, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2  =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < 5Bo.
E0<E'
Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and
22
CHAPTER
1.
Introduction
the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average
(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence
(L32)
E i.e. in
If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then lmw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (mx2\ =
If we now insert in the equation for dE/dt,
/I 1£ 2\ (  mx ) = , \ 2 / 2u2' we obtain dE_/l
~dt ~ \2mX
and hence
2\dw
2
_Edw2
dE _ 1 dw2 _ du
/ ~dT ~ 2w2~dT'
E — = const. w
°r
~E ~ ~iU> ~~ ~u7'
In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by
~~ p
2R\2nh)
~ AKR'
See e.g. L. Schiff [243], pp. 25  27.
Chapter 2
Hamiltonian Mechanics
2.1 Introductory Remarks
In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and reexpress Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the cnumber valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.
2.2
The Hamilton Formalism
In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
A. (2. The following properties can be verified: *See e. in analogy with Eqs.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.3).p% are therefore again observables. chapter VII. ^As H. (2. 6t>0 5t Real quantities which are directly observable are called observables.1) as d .^ ^ j . Compared with an arbitrary function f(qi. . Hamiltonian Mechanics is wellknown.Pi) is contained implicitly in the canonical variables q^ and pi. (2.1). (2. the equation of motion of the observable u. x^fdu. .P.pi) of qi. as mass times velocity. Goldstein [114] remarks at the end of his chapter VIII. the entire timedependence of observables u(qi. Pi has to be considered as an independent quantity. whereas the velocity requires observations of space coordinates at different times. (2. this expression is simply a functional determinant. H. one obtains the Hamilton equations* • OH . dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi. which can be observed directly at time t. A system consisting of several mass points is therefore described by a number of such variables. i.2) If we have only one degree of freedom (i = 1). The total time derivative of u can therefore be rewritten with the help of Eqs. (2.2) as This equation is. Rather.4) contains as special cases the Hamilton Eqs.1).4) as the generalization of Eqs. Goldstein [114]. qi(t + 6t) .Pi. M.n n. since . chapter VIII. (2. which all together describe the state of the system.) = £ [wm + WiK) = £ [WiWi .. One can verify readily that Eq.g. It suggests itself therefore to consider more closely the properties of the symbols (2. the standard reference for the application of Poisson brackets is the book of P. We can therefore consider Eq. Dirac [75].t).1). One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. du \ x^/dudH du dH\ . All functions u(qi. (2.qi(t) qi = hm f . S »(«.e.
{C. Bx} + a2{A. B} is completely evaluated.B}C or C{A. a i S i + a2B2} = ax{A.5a) (2. If we evaluate the Poisson brackets for qi.5d) above. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. 25 (2. As an example we consider a case we shall encounter again and again. (2.5b) (3) complex conjugation (note: observables are real. The original definition of the Poisson bracket will not . {B.Pi. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. that the very general Eq. (2.B*}.4).BC} (5) Jacobi identity: {A.e. like here.5d) is useful in calculations.B}C + B{A. These are {Qi. but could be multiplied by a complex number): {A.Pk} = 0. B}. If.3) of the Poisson bracket. for example.2.6) We can now show. Since Eqs.6) give the values of these. (2. then the ordering is taken as in (2. we obtain the fundamental Poisson brackets. {pi. Later we shall consider noncommuting quantities. C}} + {B.Qk} = 0.C}. the Poisson bracket {A.B}* (4) product formation: {A.Pk} = 5ik. in other words without any reference to the original definition (2.B} = {B.5d) = {A*. {A. (2. B}} = 0.5e) = {A.A}.6).5c) The first three properties are readily seen to hold. (2. Property (2. (2) linearity: {A.B}.2 The Hamilton Formalism (1) Antisymmetry: {A. As long as we are concerned with commuting quantities. A}} + {C. we wish to evaluate {A. that of the linear harmonic oscillator. (2. it is irrelevant whether we write {A. {qi. i. B2}. which combines the Hamilton equations. where A and B are arbitrary observables.
p}p + p{q. These are transformations qi—>Qi = Qi(q. From Eqs.7) into (2.. Since constants are also irrelevant in this context.26 CHAPTER 2. and P={p. (2.qot2 .4) we have for u = q.t).10) we deduce q = p = q.p.P2} + {q. and so q = q. we consider as Hamiltonian the function H(q. and Pi are ordinary real number variables and that H(q. According to Eq.6). (2. Then we obtain: (2. q = {q.12) .. or q(t) = qo+ Pot .8a) and use the properties of the Poisson bracket and Eqs. Pi—> Pi = Pi(q.8b) p=q. (2. (2.9) and (2.q2}) = = 2i{q. "4' = q.p.p.11b) (2. q + q = o. (2. (2.p) is an ordinary function is also irrelevant.yPot3 + •••• (2. (2. from which we infer that q(t) — qocost + posint. In the evaluation one should also note that the fact that g.p) = ±(p2 + q2).H}.11a) 'q' = q...7) q = [q.t).10) In classical mechanics one studies also canonical transformations.H}.8b) We insert (2.8a) (2.\(p2 + q2)} = l({q. Hamiltonian Mechanics be used at all. (2.9) Similarly we obtain from Eq.p\) P.
those at a time t — 0.qk} = 0. B} of two observables A and B in terms of either set of canonical variables. .B}Q>P.2 below contains a further illustration of the use of Poisson brackets. {PhPk} = 0. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. {QhPk} = Sik.3 deals with the relativistic extension.1 we verify only Eq. for which Hamilton's equations hold.Pk} = 0.t). as {A. (2. {qi. (2.e.e. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation.P.14) With the help of the definition (2.15a). _ax p__dK_ (2.P is canonical in the sense defined above. which means that a Hamilton function K(Q. Goldstein [114].Qk} = 0. Example 2. which transforms qi. (2. Pi^Pi=Pi(Q. (2.pk} = 5ik.e. i. P) exists.15a) °raS {AiB}Q.e. Of course.B}q.15a). .t).e.Pi back to their constant initial values. i. {qi. (2.2. i. i. verify that the Poisson bracket of two observables A and B is invariant.* Hence in Example 2. that (dropping the subscripts therefore) {Pi. Example 2. {Qi.p = {A. *H. and Example 2.13) We write the reversal of the transformation (2.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P). this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.12) Qi—>qi = qi{Q. i.B}q.P.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.P One can then show that {A. Eq.P provided the transformation q. chapter VIII.3) we can now express the Poisson bracket {^4.p <> Q.
( 9A = {Pk.A}q.p = dB dA dB \ rA 1 {A'B^ E b r s r . § See P. Thus whenever q and p are multiplied. (2. OQk Pk}q. dA {Qk.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. Solution: Relativistically we have to treat space and time on an equal footing. .p^v dP.Qj}q. V V °Qk dPk dPk dQk J E x a m p l e 2. p. Qk}q. Hamiltonian Mechanics S o l u t i o n : U s i n g t h e definition of t h e P o i s s o n b r a c k e t a p p l i e d t o ^4 a n d B w e h a v e r „ „•.B}q.p^— + {A.15b) dA {Qk. Replacing here A by Q and B by A. we obtain as claimed by Eq.p(0) = poSolution: The solution can be looked up in the literature. 236.F}r ~di d~E ~ d~E~8t Consider F = H(q.B}QIP.E(t). {A.28 CHAPTER 2.3: Relativistic Poisson brackets By extending qi. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u.pc).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). Thus we extend q and p to spacetime vectors (t. we have —Et.15a) r„ ™ v .pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets.p = J2 d~Fk Replacing in the above A by P and B by A.q/c) and (E. their product Et — qp being relativistically invariant.p) . 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo.Pj}q. Mittelstaedt [197]. § E x a m p l e 2. we obtain analogously dA dQk' Inserting both of these results into the first equation.A}q.P^pr.p^ "Pk. (2.s F 5 7 r ={A.+ dQ. we obtain dA {Qk.
but partial derivatives of F do not vanish. of course._. p= p(qo. one obtains Gi with q = q((lo.. if qo.14) and area A of the phase space ellipse given by Eq.p).. Of course.dt y/l  qZ/c2' 2. we have A = (p apaq — J .p). also boundary points of one domain are mapped into boundary domains of the other. du .. and the space spanned by the entire set of canonical coordinates is described as its phase space.Po. Probabilities 29 (This is.p) around some point qo. i.y.e.p). i. We consider first the a priori weighting or a priori probability. dE . (1.t0.. du q\ P = —. and E as a function of t). and at time t > to in a domain G\(q. But now we consider a system whose phase space coordinates are not known precisely. Since Hamilton's equations give a continuous map of one domain onto another. . rrr . c dt 2 2 w^ du du dr . g oc AqAp.p) . g.p) to G\{q. since H is expressed as a function of q and p.2. so that Go(qo. to .t).to.po.3 Liouville Equation. (1. Then .)•'=(*)**£.17a). (2. it is the equations of motion which lead from Go(q. in the case of the linear oscillator with energy E given by Eq. dudH s l u . 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. H(q. Let us assume that at some initial time to the system may be found in a domain Go(q. (*. This probability is evidently proportional to AqAp.e.16) For example.1 Liouville Equation.Pi are the coordinates of some mass point m. . Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.po) = Gi(q.Po.' ±=j1V c \dtj' u^. I .t). numerically zero. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du .E(t)}r = Hence at Relativistically we really should have clu/dr. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.po is a point on Go. 2TTE UJ and hence g oc — . We distinguish in the following between two kinds of probabilities.3 2.
5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 .)curve for constant E and $ is — as may be seen by comparison with Eqs.4. in view of this independence it can be expressed in terms of the conserved energy E. the (pg.14) t o (1..eded<t> = 8TT2IE./. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. dpdq and similarly dt = — Ap. which means. Je=0 . Pi 2 / \.30 CHAPTER 2. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left.1): d . Example 2.4: Liouville's theorem Show that A q A p is independent of time i.15) — an ellipse of area § dpgdp. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates.p^.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. — dp A A — In(AqAp)s = — + — = Example 2. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. dt dt Aq d ( A p ) dt ' Ap Here d(Aq)/dt is the rate at which the qwalls of the phase space element move away from the centre of the element. and hence g oc %ir2IdE. this has the same value at a time to. Thus g must be independent of t. = 2TrIEsm6. dt dq dq dp dp d2H dqdp d2H = 0. as is demonstrated by Liouville's theorem in Example 2. J0=0 Hence g oc 8n2IdE. Example 2. dt dq and with with Hamilton's equations (2. Hence from the difference: — . Show that the total volume of phase space covered for constant E is 8n2IE. (1.— = — Aq. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' .
however.2.p.3 Liouville Equation.3. The total number of systems N is obtained by integrating over the whole of phase space.p. (2.t)dqdp= 1.t)dqdp = N.e.p.e. (2.17) F ^GT^^ 0 Fig.18) With a suitable normalization we can write this / W(q. it is suggestive to introduce a factor 2KK with every pair dpdq without.20) . whose positions in phase space are characterized by points.Po. Since W has the dimension of a reciprocal action.p + dp. W=p(q.p.Thus we assume a large number of identical sytems. leaving the basis of classical mechanics! Hence we set / «w)^ =. d d 1 P = JJdqidpi.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. (2. dn — p(q. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space.q + dq. i. Thus dn is that number of systems which at time t are contained in the domain q. Probabilities 2.19) Thus W is the probability to find the system at time t at q.t). by jr = P(9»P.*). (2.1 The system moving from domain Go to domain G\. i. 2.p.
a t dt + q+dq.3 The region G. The number of points at time t + dt in domain G. 2. 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. The equation of motion for n or W is the socalled Liouville equation.p / . i.p. we take into account.32 CHAPTER 2.t)dp{ jt Q.p.p) and vp(q. i. i.2 The ensemble in phase space. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. p + dp G P O q qndq Fig. * q Fig.p.P dt p(q + dq. 2. 2. we consider the domain G in Fig.t)dp( — I .p) denote the velocities in directions q and p — p{q. t)dqdp p(q. — if vq(q. how the system moves about in phase space. t + dt)dqdp. In order to derive this equation. p(q.e.p. that in our consideration no additional points are created or destroyed. t + dt)dqdp — p(q. In doing this.3 and establish an equation for the change of the number of points or systems in G in the time interval dt.e.p. We are now interested in how n ov W changes in time.e.
t)vq(q. so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . = K(9.t) dt p(q.t)} with JW(q. p.p.p)dtdp .p)dtdq . .p + dp) dp = or  However.p + dp. Probabilities and thus p(q.t)^ = 1.t)vp(q. (2. . (2.p(q. t)vq(q + dq.P)]K(Q.p) .p.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . t)dqdp = p(q.t)vq{q. p.p.21) with Eq.19).t)vp(q.P *' % = iH>P} ( 2 . (2.p.p. (2. the probable motion of the system under consideration. put differently. (2.t)vp(q.p(q + dq.P.p(q + dq.p.p)=p = 3H —.p. Comparison of Eq.p).p)dtdp +p(q.t)vp(q.t) dt {H(q.p) = q = g^.t)vq(q + dq. .W(q.p) .3 Liouville Equation.t + dt) .t + dt)dqdp — p(q.p.p) dq p(q.21) we can also write dW(q. vp(q. 33 Dividing both sides by dqdpdt this becomes p(q.20) and (2.2.p.p.21 ) This is the Liouville equation which describes the motion of the ensemble or. dH Mq.P)].p(q.p(q. With Eqs. .p + dp.4) shows that p and u satisfy very similar equations.p + dp)dtdq.
q+dq. We deduce from the Liouville equation the important consequence that ^ M = 0. for i<«>!/«<*p>"w>(^)". We now inquire about the time variation of the expectation value (it) of u. 2.po are the initial values of q.24). (2.26) we described the time variation of the observable u(q. i.p).e.34 CHAPTER 2. and hence that equal phase space volumes contain the same number of systems.p. With Eq. Equation (2. (. We define as expectation value of u(q. (2. p.p)W(q.4).p) be an observable. 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. (2.e.t)(^J. Example 2.24) since the total derivative is made up of precisely the partial derivatives contained in Eq.24) implies that p is a constant in time.1 the area Go is equal to the area G\. We shall see that we have two possibilities for this. 2. p+dp.p) the following expression: (u)=Ju(q.p (cf.p.4 Expectation Values of Observables Let u = u(q. that dtj\y v if qo. Thus in Fig. since no systems are created or destroyed. The first and most immediate possibility is — as indicated .4). (2. i. t) depends explicitly on time t (if determined at a fixed .that the density or probability W(q.
and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q. Thus we can write. that (u) = (u)0.8.po.t)W0(q0.p. at t = 0.Po.f(q0.p.p).po. Sec.t). we have Qo = g(q.p) = uo{qo.e. so that Q = g(qo. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.i).t).32) In this expression the time t is contained explicitly in the observable u(q. of course. 8.po. where we used Eq. we can also employ a more complicated consideration.t). We expect.p0)(^^J.t).p. (2.p.t).^ Solving the equations of motion for q. (2.Po.22). (2.33) po = f(q. We verify this claim as follows. Observables 35 point in phase space).0) =u(g(qo. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.p. (2.p.28) = Ju(q.34) S e e also H.p){H(q.t) = = W(g(q0.4 Expectation Values.24): W(q.0) = u0{qo.t).p) assumes certain values) that depends explicitly on time. Goldstein [114]. (2. since W oc p is constant in time according to Eq.t) W(q0.t).W(q.29) The distribution of the canonical variables is given by W(q. and hence u(q.p) = u(q.Po.p. (2. (2. .29). However. ie.0) = W0(q0.Po) at time t = 0.t). (2.2.t). Reversing Eq.31) i.30) p = f(qo. 1 (2. (2. Then Eq.Po.Po.po.f(qo.p0.t)}(^y. we can express these in terms of their initial values qo.t).po.27) becomes !<»>  /«(*P>!"W>(^)" (2.
(cf.p.30) u0(g(q.37) and (2.t) = (2 30) (2. (2. (2.t).p)." V " " ™0' > • • » « » 0 . (2.36) and (2.t). Eq.30): duo(qQ.POit)i p=f(qo.t)\—^\ = (u).t).p. (2. (2. (2.p. (2.t) (2.po.Po)(^^)n.29) q = g(g(q.25) into Eq.f({H(q.f(q.Po. (2.t) d t _ (du\ \ d<i dq / P=f(qo.37) as had to be shown.32).P)}). i.p)W{q.e.t).p).p0) = W(q. (240) .t).p.f(g(q.Po.t).t)W "(q0.p.t).t).39) Substituting this into Eq.31)) W0(q0.p.=9(.36) = (2.f(q.(2.36 CHAPTER 2.38) we obtain £<«)„ = .t) dp M du\ ° q 'P=f(qo.t) v{g(<j(q.(dq0dp0\n We deal with the partial derivative with the help of Eq. we obtain («)o = / u{q.t).p.M 27Th i . (2. .30) Moreover. Hamiltonian Mechanics so that on the other hand with Eq. p = f(<j(q.28).t)J(q.35) Inserting these expressions into no we obtain uo(qo.t). we obtain an expression which is different from that in Eq. Taking now the total time derivative of (it)o.po.po)(~ .t) 0H_ 0 q =  {{H(q.Po.P. / ^fdq0dpc . Inserting Eqs.P0.t) M f 9u\ V °P / P=f(q0.p)})q=g(QO.t).t). [ du0{q0.o.p.) Wo(qo.p.t). (2.Po.PO. d dt {U)0 = dt.u(q.) f / u (q0.35).p.t) .f(q.P0.32).\„.p.po.t) dB__ (du\ °P \°PS P=f(qo. Eq.0) u(q.u{q.0)=u(q.f(q.p). (2.P.t).
we obtain zero after partial integration of I and hence from Eqs.p.t)} .p){H(q. (2. ^ „.45) the Liouville equation.p. from the properties of the Poisson bracket.u(q..28) and (2.4 Expectation Values.p).p)) is attributed to the probability W(q. uW} = {H.p. W}.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions.43) ——uW = 0.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). u}W instead of u{H.34) and use (2.(2.. and the time variation of (u(q. we obtain {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.t) dt {H(q.W} in Eq.p)}W(q. (2. (2.p.t) dqdp\n 2irhJ (2. .p).t)}. (2.25) and (2.p). .uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2.42) the relation d . Observables 37 Here we perform the transformation (2. lim ——uW — 0. the observable u is treated as a function u(q. described as "Schrodinger picture". In the first case..W(q.28).t) of u assuming certain values q and p.45) in agreement with Eq.28).p. This timedependence is described by the Liouville equation dW(q.W(q. so that ~d~t <«>0 = J{H(q. However. Alternatively we could deduce from Eqs.41) This expression contains {H.31). x r „.41) and (2.p). u}W + u{H. Consider {H.2. / d q d p \ n 2irh J (2. The phasespace integral of a Poisson bracket like I (2. dt <«)o v f u(q.
p). and probabilities W. does not have to be commutative.po are constant initial values — we have du0(q0. The time dependence of the expectation values can be dealt with in two different ways. Thus we arrive at a more general theory if we define u and W with the following properties: (a) An addition is defined between the quantities.po.4) of an observable on the other. the probability of the initial values Wo(qo.p)}. but does satisfy the usual associative and distributive laws. the reason being that — since qo. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations.. The equation of motion is then that of an observable. (cf.po) is assumed. i. 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. (c) A multiplication of the quantities among themselves is defined. This formulation deals with observables u representing physical quantities.p)}.p0.38 CHAPTER 2.4)) ^ = {«(.t) dt du(q.F(«.e.rt.H(q. Hamiltonian Mechanics In the other case.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics. for which the axioms of a commutative group apply.t). (2. (b) As usual a muliplication by a complex number is defined.p) It = {u(q.46) We thus also recognize the connection between the Liouville equation. called u Heisenberg picture". as we observed. Koppe [152] at the university of Munich around 1964. ." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. which.po. (2. and the equation of motion (2. " T h e author learned this approach from lectures of H.t) dt __ ~ (2J30) ( 4) du0(qo. as the equation of motion of an ensemble or of a probability distribution on the one hand. " This means: Only in the Heisenberg picture dW/dt = 0. Eq. which describe the state of a system.
it is helpful to introduce already at this stage some additional terminology. Introducing it here assumes. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. (2. ~[q.p} = 0 • — ^ ~[q. (2.) = TV (vu). Eq.48) In matrix theory the symbol "trace" has a welldefined meaning. so that we consider this next. (c) TV (u*u) > 0. also written "TV".p)(^y. which all apply to (2. can be deduced from the following characteristic properties of a trace. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. also called "density matrix".2. Thus we can write the expectation value of an observable u (cf. therefore. px > in—.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. (b) TV(cm + j3u) = oTV u + /3Tr v.p} = l {q. if u ^ 0.q}=0. for a phasespace integral we define the word or symbol "trace".p] = l. y:\p.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. ox In view of our later considerations. B] := AB — BA in the following way: 'i {q.6) a socalled "commutator''' [A.26)) (u) = Tr(wW). that its use here implies the essential properties it has in matrix theory. Moreover.q} = 0 {p.48): (a) TV u* = (Tr «)* = TV^. and to be able to correlate these with the above classical considerations. In Chapter 9 we attempt a corresponding approach for classical systems with . by Traceu:=Ju(q. (d) Tr(ut. That this is the case.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. (2.
These aspects will be considered in Chapter 27. excepting the Poisson brackets. i. electrodynamics. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics.e.e. since gauge fixing (i. it will be shown that in electrodynamics. .40 CHAPTER 2. Hamiltonian Mechanics a wavelike nature. 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. the Poisson brackets require modification to Dirac brackets. However.
3. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. Building upon this. 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. if the elements ipi of M satisfy the usual axioms of addition and K 41 .2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. We found that the Poisson algebra permits extensions to noncnumber formulations. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. we can define the Hilbert space as the space of states of a physical system. i.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. of measurable quantities. and selfadjoint operators in this space as representatives of observables. which turn out to be those of the theory today known as quantum mechanics. i.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. A set M.e.e.
(ipi + ipj) + ijjk = ipi + (ipj + ipk).e.ip2. n is called the dimension of .3) If all on = 0. and n is the smallest such number. if every vector ip E M can be associated with numbers Q . • • • . with the properties (a* G IK): (^2.i = 1.02)* (hermiticity). such that n ^ = YJCi^i.1) aipi + aipj. . i=l (3. (a.^1) = (01. ^ i = 0 if ^ = 0. if any two elements ip\. f > 0 if V T ^ O . x M. 2 . so that n 5 > ^ = 0.ipn are said to be linearly dependent. i. . ..5b) . the vectors il)i.2) Vectors ip\. —»• IK.ipn are said to be linearly independent. E (3.  . are linearly dependent.^i) . Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. ( n (3.tp2.5a) (3.~ip2) '• M. i = 1.42 CHAPTER 3. (V>. . .^) called inner product.M. n. (3. 2 . (3. If n + 1 elements ^ € M. ip2 of this space can be associated with a complex number (V'l. . In each case n linearly independent vectors are said to form a basis.ip2). if numbers eti. fa.4) The vector space . . . n exist (not all zero). (aptyi. . (ipi. (3 5c) + a2{ip. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M).aiV>i + a 2 ^ 2 ) = ai(ip.M is said to be a metric vector space or a preHilbert space. . A + tpj = tpj + i'i./?eK).
(3.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations.2.^ 2 ) =   ^   2 + 2 ^ i . \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).)> (36) i. WA+Ml2 = ll^il 2 + ll^2 2 .V>i + A</>2) > 0 .i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).tp2 £ M. ^ ) + A 2 (V2. In order to verify Eq. (310) (^2. ^ i ) + A * ( ^ .9b) = 0 (Pythagoras theorem).7) In addition the following relations hold in a metric space M.9a) we start from if) = ipi + \if>2 6 M.2 Hilbert Spaces 43 where the asterix * means complex conjugation.^1) 2 h 11 0 <%«e + M . ^ 2 )  . which we can write 0 < ( ^ i . linearity in the second component. (3.V.rh) (3.. for arbitrary A and ip2 ^ 0: (V>i + AV>2.V') +"2*(V. ^ i ) + A ( ^ i . (3.^2)1 < H^ill • 11^211 (Schwarz inequality).*h)=\\AH\(38) (3. »2 2 .9a) (3.is defined by d(ipi.^) = ai*(V'i.9d)   V i   = sup  ( ^ 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.e. meaning oneandahalffold linearity). The first two properties imply ( a i ^ i + a2ip2. for ip\. antilinearity in the first component (also described as sesquilinearity.9c) HV'i + tp2\\2 + \\1p1 . The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2. tp2 £ M: 1(^1. (3. if Wi.3. The distance between two vectors if>i..
i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. Examples of metric vector spaces: (1) Let M.13) the vector is said to be normalized.^)1 <IIV>il. beginning with A = 1 in the second line of Eq.9e). (3.\\ih\\.^1)1 Vi 2 + ^2 2 + 2   ^   .9d). (3. for which CO 2 := V^l^l 2 < 00. Mathematical Foundations of Quantum Mechanics 1(^2.44 or CHAPTER 3.V2) IIV'l2 + V'22 + 21(^2. ^ e M are said to be orthogonal if (</>!. (3.11). V 2 ) = 0 .10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl.ii) thus verifying Eq.* Two vectors i f i .  ^ l l = (HlMI +   « 2 . (3. (3. so that   ^ l + ^ 2   <   ^ l   + ^2.9c). . (3. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. ll^ill + llVdl. HIHI (3.12) We omit here the verification of the remaining relations (3. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. hence verification in each case is necessary.9a). In verifying the triangle inequality we use this result (3. o ^ V i = o. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X .
(3. 2 and so on.5c). i. In order to avoid this difficulty. all squareintegrable functions / which are "almost everywhere equal".5a). (2) Let M.5c)) / ( x ) = 0 = > ( / . Eq. / ) = 0.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). which differ solely on a set of measure zero. = 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.w) :=J2v*Wi. . is defined by ([/]. are combined to an equivalence class [/] (with space L2). Js the space C? is not yet a metric vector space although for (cf.g)= [ /(x)^(x)d3x. With the scalar product (f.[$]):= / JM /(x)* 5 (x)d 3 *. with inner product oo 45 (v. (3. and one defines addition and multiplication by complex numbers with respect to these classes. etc. for which the scalar product.e. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. (3.5b). But this applies also in the case of any function which is nonzero only on a set of measure zero. */ x } = / /o for x = 0. which satisfies relations (3. i. Then L2 is the space of all these equivalence classes. •'^ \ 0 otherwise. Elements of the classes are then called representatives of these classes.e.3.
46 and
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
ll[/]=0=»[/] = [0],
where [0] is defined as the class of all functions which are almost everywhere zero. This means that functions that differ only on a pointset of Lebesgue measure zero are looked at as identical. Unless necessary, we ignore in the following mostly for simplicity the distinction between C2 and L2. Convergence of sequences in Hilbert space is then called convergence "almost everywhere". With the help of the concept of a norm we can introduce the concepts of convergence and point of accumulation. Definition: A sequence {tpn} € M. is said to converge (strongly) towards tp E M., if the distance \\tp — ipn\\ tends towards zero, i.e. lim   ^  V n   = 0.
n—•oo
(3.14)
The vector ip is then called point of accumulation. The point of accumulation does not have to be an element of A4. If M. contains all of its points of accumulation, the set M is said to be closed. A normalized vector space M. which with every convergent sequence contains a vector towards which the sequence converges, is said to be complete, i.e. if ipn e M. with lim \\ipn  i/jm\\ = 0
m,n—>oo
(called Cauchy sequence), there is a ip 6 M. with ip = lim ipn,
n—>oo
i.e.
lim \\ip — ipn\\ = 0.
n—>oo
(3.15)
Every finitedimensional vector space (on IK) is complete in the sense of the concept of convergence defined above (so that completeness does not have to be demanded separately). In order to see this, we consider the convergent sequence
n
<Pa = ^2CmiPi,
where i/ji,...,ipn€.A>i ras)
CaieK,
(3.16)
constitute a basis in M Then (according to Pythagon
nv«  M = i sec™  c0i)A\\ =YI i c  c^2'
2
2
n
(3i?)
3.2 Hilbert Spaces
47
a relation also known as Parseval equation. The convergence of the sequence i/ja implies the convergence of the sequence {Cai} towards a number Cj. Then for the vector
i=l
we have
ll^aV'II^ElC^Cil2,
i=i
(3.18)
i.e. that the sequence of the vectors tpa converges towards tp. We thus arrive at the definition of a Hilbert space. Definition: An infinitely dimensional, metric vector space, which is also complete with regard to (strong) convergence, is called a Hilbert space "K. The given definition of a Hilbert space is that usually given in mathematics.t In physics this is generally supplemented by the requirement that the space be separable, i.e. of a countably infinite dimensionality. Naturally Hilbert spaces with a countable basis are the simplest. We supplement the above by referring to the concept of a dense set or subset M of "K. A subset M of "K is said to be dense in "K, if to every / 6 "K there exists a sequence of vectors fn, fn < M, so that fn — f, S > i.e. fn converges strongly to / , implying that every vector / e Ji can be approximated arbitrarily precisely. We consider next some examples. Examples of Hilbert spaces: (1) The hyperspherical functions Yitm(6,(p) define a complete set of basis functions on the unit sphere. Any function f(9,(p) with
[\f(e,tp)\2dn<™
can be written as a convergent series
oo 1=0 /
f{9, ip) = Y, E
m=l
C
l,mYl,m(0, ip).
(3.19)
For completeness we recall here the definition
(o<p)twi r(2*+iwm)!i x v„ sin ™ 0 (±y +m (cos2e _ lV Yl
*i,mW<P)2in
[
47r(Z + m ) !
J
e
sm
\
d 9
)
^cos V
^'
See e.g. N. I. Achieser and L. M. Glasman [3].
48 so that
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Y0fl = —=,
J Air
1
Yito = \ —cosO,
V 47T
/ 3
Yit±i =
I o
^\—e±tvswt
•
A Hilbert space contains a complete set of orthonormal basis vectors or a corresponding sequence precisely then if it is separable. (2) On the space L2(0,2n), i.e. on the space of squareintegrable functions on the interval [0, 2TT], an orthonormal system, called the trigonometric system, is defined by the functions
1
einx,
±n = 0 , 1 , 2 , 3 , . . . .
(3.20)
/2TT
(3) On the space L2(a,b), where (a, b) is an arbitrary but finite interval, a complete but not orthonormal system is given by
In order to obtain the orthonormalized system, one employs the orthogonalization procedure of E. Schmidt.* The sequence of polynomials thus obtained consists of the Legendre polynomials which are defined on the interval — 1 < x < 1. These polynomials are defined as follows: Po(*) = l, *>„(*) = _ 1 dn —(x2ir, n = l,2,.... (3.21)
These polynomials satisfy the following normalization conditions, i.e. are orthogonal but are not normalized to 1:
l
/_
Pm(x)Pn{x)dx
l
= 0, (m ^ n),
(3.22a)
and
J' [Pn{x)\2dx = 1 ^
l
.
(3.22b)
(4) By orthogonalization of the following functions
x2/2
xe~x 2
/2
x2px
2
/2
one obtains the following functions defined on the space L2{—oo,oo): ^[x) = (  l ) n e * 2 / 2 f ^ e  * 2 = Hn{x)ex2'2 dxn (3.23a)
''This procedure is wellknown in analysis, and hence will not be elaborated on here.
3.3 Operators in Hilbert Space with
/*C5tJ
49
j —(
bn(x)<t>m(x) = 2nn\yft6nm,
(3.23b)
where Hn (x) is the Hermite polynomial of the nth degree (which we do not go into in more detail at this stage).§
3.3
Operators in Hilbert Space
Having defined the admissible states of a physical system as the vectors which span a Hilbert space, the next step is to introduce quantities representing operations in this space. This is our objective in this section. We begin with a number of definitions. Definition: Let T>A, the domain of definition, be a (dense) subspace of the Hilbert space "K. Then one defines as a linear operator A on "K the mapping A : VA > H with (a,/3 G C, ^ , ^ e ^ c M ) A(cnl)i + 0rfo) = a(A^i) + /3(At/>2). (3.24b) (3.24a)
Definition: One defines as norm (i.e. operator norm) of the operator A the quantity sup l i M . (3.25) ipevA\{o} WW Definition: An operator A is said to be bounded, if its norm is finite, i.e. \\A\\ < oo. Definition: Two operators A : T>A — "K and B : T>B —>"Kare said to be > equal if and only if Atp = Btp, for every ip G VA and Z>A = T>BExample: An example of a linear operator is given by the differential operator D:= —
d
A:=
_.:.LU ^ f./. _ T1 ^ rwith VD = iil>eL*,jeL*>.
T2
(3.26)
Definition: We define the operations of addition and multiplication of operators by the relations (A + B)tp:=Aip
s
+ Bip, \/i/j<EVA+B = VAnVB,
(3.27a)
Hermite polynomials are dealt with in detail in Chapter 6.
50
CHAPTER 3. Mathematical Foundations of Quantum Mechanics (AB)ip := A(Bip),
V^GPB,
B^eVA.
(3.27b)
Definition: We define operators called commutators as follows: Let A : T>A —> "K, and B : T>B — "K be linear operators in the Hilbert space "K. Then we > define as commutator the expression [A, B] := AB  BA with V[A,B\ = D A n £>B » ft. (3.28)
Definition: If .A : X^ — IK for i/> 6 T>A\{0}, then ^ is called eigenvector with • respect to the eigenvalue A € C if and only if AV> = \i/>. (3.29)
Very important for our purposes are the concepts of adjoint and selfadjoint operators. Definition: Let A : T>A — IK and ip € P A C IK. Then A^ is called adjoint » operator of A if for A^4> := <jj, <f> € T>Aj, the following relation holds: {A* 4,il>) = (<!>, Ax/,). (3.30)
Definition: The operator A : X>A — ft, is said to be symmetric if and only if >
0 M ^ ) = (MV), V ^ i e P
A
c%
(3.31)
Definition: The operator A, A : X^ — IK, is said to be hermitian or selfad> joint if and only if A = Af, One can verify that: (A*)* = A, (A + S j ^ A t + (AA)+ = A*A , (AB)t = S U * , (A" 1 )* = (A*)" 1 .
f
i.e. D A = D A t
and A ^ = A</>.
(3.32)
(3.33a) fit, (3.33b) (3.33c) (3.33d) (3.33e)
We are now in a position to construct relations between operators A and B in K, which correspond to the relations (2.5a) to (2.5e) of Poisson brackets, however, we omit their verification here: [A,B] = [B,A], (3.34a)
3.3 Operators in Hilbert Space [A, aiBi + a2B2] = ax[A, B{\ + a2[A, B2], [A,B]* = [A\B\ [A,BC] = [A,B}C + B[A,C], [A, [5, C]] + [B, [C, A}} + [C, [A, B}} = 0.
51 (3.34b) (3.34c) (3.34d) (3.34e)
The last relation is again called a Jacobi identity. Comparison of Eq. (3.34c) with Eq. (2.5c) requires here in the definition of the corresponding quantity the introduction of a factor "i" (see Eq. (2.47)). As important examples we consider the following operators. (1) q3 : Vqj  L 2 (R 3 ). We write sometimes the application of the operator qj to <f> G L2(IR3): (qj4>){x), and we define M)(x) :=arj0(x). (3.35) We can read this equation as an eigenvalue equation: Operator qj applied to the vector <f> yields the eigenvalue Xj multiplied by <> in the present case on /, 3 R . Since Vqj = {<£ G L 2 (R 3 ) : qj* G L 2 (R 3 )}, we have
^ = ^ t 
Furthermore, for instance for ip, <p G L2(IR1), we have (ip,q<p) = / ip*(x)(q(f))(x)dx = / ip* (x)x<p(x)dx
=
(#,«•
(3.36)
Since for the adjoint operator A^ of A:
it follows (with Pg = V t from above) that qj — q3^. (2) P j : VPj > L 2 (R 3 ) defined by (p 3 »(x) := ih^<t>{y). In this case we have VPj = <<t>€ L 2 (R 3 ) : c continuous, ~ G L 2 (R 3 ) I = / > Vpj]. (3.37)
52
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Here for ip,(f> e L2(Ul): (V,W>) = /V(aO(p0)dx
(3
=7)ifi
f^*(x)^(f>(x)dx
= inr(x)<i>(x)\f00J^r(x)<i>(x)dx
— — I [ ih—i/}(x) \ (f)(x)dx = J (pip)*(x)<f>(x)dx = (p^,0) = (pty,0), (3.38)
so that p* = p. Something similar applies in the case of the following operator which represents classically the kinetic energy T: (3) T:VT^ L 2 (R 3 ) and
™ ( x ) := "£ A<Kx) = ( ^ & 2 V ( x ) '
As a further example we consider the commutator. (4) Let the commutator be the mapping
(339)
Then for <p E L 2 (R 3 ): \Pj,Qk]<l>(x) = i.e. formally The following commutators which define the Heisenberg algebra \Pj, Qk] = ihSjk, \pj ,Pk] = 0, [qj ,qk] = 0 = (pjqk ~ 9fePj)^(x) = (pjqk(t>)(x)  (qkPj(t>)(x)  i ^ f ^ X f c 0 ( x )  xfe—</>(x) J = ih6jk(j)(x.),
are called canonical quantization conditions with respect to a theory whose classical version possesses the fundamental Poisson brackets {Pi, ?*} = Sjk, {Pj,Pk} = 0, {qj,qk} = 0.
The simplest example to consider is the harmonic oscillator. We postpone this till later (Chapter 6). We add, that the quantization must always be
3.4 Linear Functionals and Distributions
53
carried out on Cartesian coordinates. Moreover, the above relations assume that the three degrees of freedom are independent, i.e. there are no constraints linking them. Systems with constraints can be handled, but require a separate treatment.^
3.4
Linear Functionals and Distributions
We now introduce the concept of a continuous linear functional on a socalled test function spaced Our aim is here, to provide a frame in which the formal Dirac bra and ketformalism to be developed later finds its mathematical justification. We require in particular the delta distribution and Fourier transformations. A subset of a Hilbert space "K is called a linear manifold D, if along with any two elements (f>i,4>2 G ^ C 'K this also contains the linear combination of these, i.e. « i ^ i + a2<^2 £ ^ , ai,a2€C. (341) A linear functional [/] in the Hilbert space "K is a mapping of the manifold T> into the set of complex numbers, i.e. [/] : V  C and is written [/]<<£> = / < 0 ) : = / with the property of linearity, i.e. f(<l>i + <h) = f(<l>i) + f{<h), (i.e. the expression (3.42) is antilinear in the first component). If / ( x ) is for instance a locally integrable function (i.e. for a compact set e R n ) like exp(ik • x), then it is clear that the function </>(x) has to decrease sufficiently fast at infinity, so that the expression in Eq. (3.42) exists. Functions which provide this are called test functions (see also later). Instead of the Hilbert space we therefore consider now a space of test functions (vector space of test functions), and on this space linear functionals. Definition: The compact support of a continuous function <fi : Rn — C is > defined to be the compact (i.e. closed and bounded) set of points outside
f
/(x)0(r)dx,
(V<P€V)
(3.42)
S e e Chapter 27. "We follow here to some extent W. Giittinger [127].
54
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
that of (ft = 0. Test functions cf> with compact support are exactly zero outside their support; they define the space D(Rn). A different class of test functions <j> consists of those which together with all of their derivatives \Dn(f>\ fall off at infinity faster than any inverse power of x. These test functions are called "rapidly decreasing test functions" and constitute the space S(Rn): D(Rn) S(Rn) := {(f> G C°°(DRn > C) : support of </> compact}, := {4> £ C°°(Rn > C) : \x\m\Dn(j>\ bounded, m,n,... e I > 0}. N (3.43) Definition: Distributions f{<f>) are defined to be the linear functionals on D(Rn) and tempered distributions the linear functionals on S'(IRn). A subset of distributions can obviously be identified with ordinary functions, which is the reason why distributions are also called "generalized functions".
3.4.1
Interpretation of distributions in physics
It is possible to attribute a physical meaning to a functional of the form
f (</>):= J dxf(x)<j>(x).
(3.44)
In order to perform a measurement at some object, one observes and hence measures the reaction of this object to some tests. If we describe the object by its density distribution f(x), like e.g. mass, and that of its testing object by <t>(x), then the product f(x)(p(x) describes the result of the testing procedure at a point x, since f(x)4>(x) = 0, provided f(x) ^ 0 and <p(x) ^ 0, i.e. if object and testing object do not meet. The expression then describes the result of the testing procedure in the entire space. If we perform the testing procedure with different testing objects and hence with different test functions <pi(x),i = 1,2,..., we obtain as a result for the entire space a set of different numbers which correspond to the individual 4>i{x). These ^dependent numbers are written as in Eq. (3.44):
m = j f{x)<t>{x)dx.
If f(<j>) = 0 for every continuously differentiable function <f>{x), then f{x) — 0. In general one expects that a knowledge of the numbers f{<j>) and the test
3.4 Linear Functionals and Distributions
55
functions <p(x) permits one to characterize the function f(x) itself, provided the set of test functions is complete. In this way one arrives at a new concept of functions: Instead of its values y = f{x) the function / is now determined by its action on all the test functions <f>(x). One refers to this as a functional: The functional / associates a number /(</>) with every test function (f>. Thus the functional is the mapping of a space of functions into the space of numbers. f{4>) is the value of the functional at the "point" <f>. With this concept of a functional we can define quantities which are not functions in the sense of classical analysis. As an example we consider in the following the socalled "delta distribution".
3.4.2
Properties of functionals and the delta distribution
The delta distribution is defined as the functional 5(<j)) which associates with every test function <fi(x) a number, in this case the value of the test function at x = 0, i.e. 6(4>) = </>(0), (3.45a) where according to Eq. (3.44) 6((j)) = f 8(x)(/)(x)dx. (3.45b)
The result of the action of the "delta function" 5(x) on the test function 4>(x) is the number 0(0). The notation J 5{x)(j>(x)dx is to be understood only symbolically. The example of the delta function shows that a function does not have to be given in order to allow the definition of a functional. In order to insure that in the transition from a function f(x) defined in the classical sense to its corresponding functional /(</>) no information about / is lost, i.e. to insure that f(4>) is equivalent to f(x), the class of test functions must be sufficiently large. Thus, if the integral J f(x)4>(x)dx is to exist also for a function f(x) which grows with x beyond all bounds, the test functions must decrease to zero sufficiently fast for large x, exactly how fast depending on the given physical situation. In any case the space of test functions must contain those functions <f>(x) which vanish outside a closed and bounded domain, since these correspond to the possibility to measure mass distributions which are necessarily restricted to a finite domain. Furthermore, for the integral (3.44) to exist, the test functions must also possess a sufficiently regular behaviour. For these reasons one demands continuous differentiability of any arbitrary order as in the case of 5(0?") above. Certain continuity properties of the function f(x) should also be reflected in the associated functional. The reaction of a mass distribution f(x) on a test
56
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
object <p(x) is the weaker, the weaker cp(x) is. Thus it makes sense to demand that if a sequence {(pi(x)} of test functions and the sequences resulting from the latter's derivatives of arbitrary order, (p^(x), converge uniformly towards zero, that then also the sequence of numbers ((pi) converges towards zero. We refrain, however, from entering here into a deeper discussion of convergence properties in the space of test functions. The derivative of a distribution f((p), indicated by a prime, is defined by the following equation f((P):=f(cP>). (3.46) This definition suggests itself for the following reason. If we associate with the function f(x) the distribution
oo
dxf(x)</>(x),
/
oo
then the derivative f'(x) becomes the functional
oo
/
dxf(x)cP(x)
oo
= /($'),
(3.47)
as in Eq. (3.44). Partial integration of this integral then yields, in view of the conditions <p(±oo) = 0,
oo
dxf(x)<P'(x) = /(</>'),
/
oo
as in Eq. (3.47). Equation (3.46) defines the derivative of the functional f(<p) even if there is no function f(x) which defines the functional. For instance in the case of the delta distribution we have 6'(cP) = S((P') = <j>'(0) according to Eq. (3.45a). Formally one writes, of course,
oo
(3.48)
/•oo
/
oo
dx5'(x)<p(x) =
[<J(a;)0(a;)]?foo  /
J—oo
dx5(x)(P'(x) (3.49) (350) (3.51)
= 5(<P>) = <P'(Q). For an infinitely often differentiable function g(x) one has apparently (5 •/)<</>> = / ( # ) , so that (x6)((P) = S(x(P) = [x(P}x=0 = 0 x (P(0) = 0,
3.4 Linear Functional and Distributions or ' x6(x)</>(x)dx = 0,
/ '
57
x5{x) = 0.
(3.52)
Thus formally one can operate with the delta distribution or delta function in much the same way as with a function of classical analysis. As a further example we consider the relation f(x)S{x) = f(0)S(x). According to Eq. (3.50) we have f{x)5{x)4>{x)dx = «J(/0) = [f(x)4>(x)]x=0 = f(0)<f>(0) (3.53)
/
=
f(0)5(<f>) = J
f(0)6(x)<j>{x)dx,
as claimed in Eq. (3.53). Formal differentiation of the relation (3.53) yields f(x)6\x) = f(0)S'(x)  f'(x)S(x). (3.54)
One can convince oneself that this formal relation follows also from the defining equation (3.46). In particular for f(x) = x we obtain the useful relation x5'{x) = S(x). (3.55)
A very important relation for applications is the Heaviside or step function 9{x) which is defined as follows:
From Eq. (3.56) we can deduce the relation 6'{x) = 6{x). (3.57)
For the verification we associate with the step function the following functional
oo
roo
/
oo
0(x)<f>(x)dx = / <f>(x)dx.
JO
/•oo
For the derivative we have according to Eq. (3.46) e'(x)(<p) = 8(x){4f) = dxc/)'(x) = (f)(0)  0(oo) = 0(0)
= 8{x)(<i>),
58
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
or symbolically 0'(x) = 5(x), i.e. Eq. (3.57). After the introduction of the delta function it is customary to consider briefly Fourier transforms, i.e. in the case of one dimension the integral relations f{x) g(k) = =  =
V Alt
I
1
\ /
f°°
dkg(k)eikx dxf(x)eikx
=
F~lg{k), (3.58)
Joo
roo
 =
= Ff(x).
\Z2TT JOO
We assume here some familiarity with these integral relations. It is clear that the existence of these integrals assumes significant restrictions on the functions f(x),g(k). As a formal relation in the sense of the theory of distributions we deduce from Eq. (3.58) the important formal integral representation of the delta function, i.e. the relation
I
/•OO
S(x) = — /
dkeikx = S(x).
(3.59)
One can see this as follows. According to Eq. (3.58) /(0) =   L fdkg(k) yzir J and g(k) =  L y lis J fdxf(x)eikx.
Inserting the second relation into the first, we obtain
/(0) =
Jdxf(x)^jdke^,
and comparison with Eqs. (3.45a) and (3.45b) yields (3.59). We close this topic with a comment. The singular delta distribution was introduced by Dirac in 1930. The rigorous mathematical theory which justifies the formal use of Dirac's delta function was only later developed by mathematicians, in particular L. Schwartz. Thus today the singular delta distribution is written as a regular distribution, i.e. as an integral operator, by writing, for instance, 5a= f <5(x  a)^(x)dx = ^(a) V 0 e S ( [ R n ) , (3.60)
and one derives from this that the delta function <J(x) has the (impossible) properties of being (1) zero everywhere except at x = 0 where it increases so enormously that (2) the integral over <5(x) is unity:
f S(x)dx = 1.
Chapter 4
Dirac's Ket and BraFormalism
4.1 Introductory Remarks
In this chapter we introduce the (initially position or momentum space representationindependent) notation of Dirac* which is of considerable practicability. We also introduce those properties of the notation which make the calculations with states and operators amenable to simple manipulations. The notation will be used extensively in later chapters. The integral representation of the delta function discussed in Chapter 3 can be considered as a formal orthogonality relation for harmonic waves exp(ikx) which finds its rigorous justification in the context of distribution theory. Thus we have — again for simplicity here for the onedimensional case —
1
/"OO
S(x x') or 6(k k')
= — /
27T
i
dkeikxe~ikx'
(4.1a)
J_00
/oo
= — /
27T J_00
dxeikxeik'x.
(4.1b)
Since k and similarly x here assume a continuum of values, Eqs.(4.1a) and (4.1b) are described as normalization conditions of the continuum functions exp(ikx) as distinct from orthogonality conditions for functions depending on integers m,n as, for instance, the trigonometric functions um(x) = cos(mx), vm{x) = sin(mx),
'See P. A. M. Dirac [75].
59
60 for which
CHAPTER 4. Dirac's Ket and BraFormalism
1 fn  / dxum(x)un(x)
^
= 6mn,
(4.2a)
JK
i r
 /
^ J7T
dxum(a;)i;ri(x) = Smn,
(4.2b) (4.2c)
i
 \
^
r
dxum(x)vn(x) = 0.
JK
In the following we shall therefore "orthogonalize" continuum states represented by vectors of a Hilbert space which is no longer separable (i.e. which has at least a subset whose vectors are characterized by a continuous parameter) in the sense of the relation (4.1a) to a delta function instead to a Kronecker delta as in Eq. (4.2a). With this formal use of the delta function we can manipulate continuum states easily in much the same way as discrete states which implies an enormous simplification of numerous calculations, t The Fourier transforms introduced in Chapter 3 permit an additional important observation: We have several possibilities to represent vectors in Hilbert space, since the Fourier transform describes the transformation from one representation to another (" configuration" or "position space representation" <> "momentum space representation"). A position space Schrodinger wave function ?/>(x), which we shall consider in detail in numerous examples later, corresponds to the representation of the vector ip of a vector space (as representative of a state of the system under consideration) in the position space representation, i.e. as a function of coordinates x. In the momentum space represenation the vector tp is the Fouriertransformed V>(k) of ip(x); this representation will be used in particular in Chapter 13.
4.2
Ket and Bra States
In the following we introduce the notation of Dirac.* We define ketvectors as elements of a linear vector space and bravectors as those of an associated dual vector space. The syllables "bra" and "kef are those of the word "brackef. The spaces are linear vector spaces, but not necessarily separable Hilbert spaces, unless we are dealing with an entirely discrete system. More as an excercise than as a matter of necessity we recall in the following some considerations of Sec. 3.1, expressed, however, in the notation of Dirac. Hopefully this partial overlap is instructive.
'Formally this means that we have to go to an extended Hilbert space, which contains also states with infinite norm, see e.g. A. Messiah [195], Vol. I, Sec. 7.1.3. *P. A. M. Dirac [75].
4.2 Ket and Bra States
61
In order to achieve a representationindependent formulation we assign to every state of the system under consideration a vector, called ketvector and designated ), in a vector space V. In the symbol \u), for example, u is an index of the spectrum, i.e. a discrete index in the case of the discrete spectrum, or a continuous index in the case of a continuous spectrum. The linearity of the vector space implies, that if £ is a continuous index, and A(£) an arbitrary complex function, then with £),
\w) = J\(0\Odt
(43)
is also an element of V. As mentioned, the space can be finitely or infinitely dimensional. In the vector space V of ketvectors a set of basis vectors can be defined, so that every vector can be expressed as a linear combination of these basis vectors. With the vector space V of ketvectors we can associate a dual space V of bravectors, which are written (x. The bravector (% is defined by the linear function <X{«» of ketvectors \u). For a certain ketvector \u) the quantity x has the value (xit), which is, in general, a complex number. For a better understanding of the concept of the dual space, we recall first the difference between a linear operator and a linear functional. According to definition, the linear operator which acts on a ketvector \u) G V yields again a ketvector \u') G V. On the other hand, a linear functional x i s a n operation, which assigns every ketvector \u) G V linearly a complex number {x\u}, i.e. X= with x ( A i K ) + A 2 « 2 )) = A i x O i ) ) + A 2 x ( M ) The functionals defined on the ketvectors \u) G V define the vector space V called dual space of V. An element of this vector space, i.e. a functional x is symbolized by the bravector (x.§ Then (x\u) is the number that results by allowing the linear functional (x G V to act on the ketvector \u), i.e. X(\u}) = <xl«>One should compare this with our earlier definition of the functional / on the space of test functions <> We wrote this functional /. «)eV>x(l«»
m = w».
By construction every ketvector is assigned an appropriate bravector. The reverse is not true in general. See e.g. C. CohenTannoudji, B. Diu and F. Laloa [53], p. 112.
62 CHAPTER 4. (4. if (u\v) — 0. or (4. i.e. if x\u) = 0 V kets \u) E V.5) 4. the norm squared.(t.4b) (4. Dirac's Ket. if \u) = 0 (see (3. Two ketvectors \u). Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . (2) (u\u) > 0.and BraFormalism Furthermore we have: ( x  = 0. V \u) (in the case of £) with infinite norm!). (3) (u\u) = 0. that the vector \u) E V is associated with a vector in V. It follows that we can now write the Schwarzian inequality (3.3 Linear Operators.5c)).\u) = (u\0*. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. 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. which we write (u\. The dimension of V is the same as the dimension of V. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. \v) E V are said to be orthogonal. This expression is linear with respect to \u) and antilinear with respect to \v).4a) \v) = J\(m<% and (4.4c) (44d) (v\ = J\*(0(m One defines as the scalar product or inner product of \u) E V and \v) E V the cnumber (v\u). As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*.9a): \{u\v)\2 < (u\u)(v\v).
we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. This product is commutative. AB^BA. Finally we define the tensor. In this way the operations of operators A.or direct product of vector spaces. The operator A.. BA = 1. It follows that (r. Then the following rules apply: A + B = B + A. on bravectors are similar to those on ketvectors. Multiplication by unity. if \v) — A\u) i. defines the unit or identity operator. the space Vi <S> V2. Furthermore.e.4. let (771 6 V be the bravector defined by this functional. i. the space Vi <S> V2 has the dimension . with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) . ltt) = \u).or Kronecker. and one writes: (V\ = (X\A. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. If Vi has the dimension A/i. Let the following bravector be given: (x\ £ V. i. Then one says: (771 results from the action of A on (x. This does not always exist. Two operators A and B are said to be inverse to each other. as we can see as follows. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). The inverse of A is written A . if AB = 1.e. Furthermore linearity applies.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u).3 Linear Operators.. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. The product vectors l?^ 1 )^ 2 )) span a new vector space.. The product of such vectors is written: \u^u^) = \u)^\u)^. 1 and \u) = B\v). B. denned as a linear operator on V. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). The inverse of a product is (AB)'1 =B1A~1. also acts on the dual space V.e.
e.e. [A^. then \v) is a linear function of \u).7a) (4. Every operator A^> commutes with every operator A(2\ i. here designated with the same symbol.7C) («)(u)t = \v){u\. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). we obtain the conjugate relation <tAt«> = (u\A\t)* (4.e. (v\t) = (u\A\t). Next we construct the following scalar products: (t\v) = (t\Ai\u).g.A^] = 0. Every such operator is then also associated with an operator. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. i.and BraFormalism an operator in the Mi A/2. Since we demanded the scalar products to have the property (t\v) = (v\ty. in the product space. AW\U)W = \V)W. 3.e. AW\uW)\uW) = \vWUW). i.3). and (v\ = (u\A.7d) .64 CHAPTER 4. Dirac's Ket. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.6) for all \u). Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. also Sec. \t) G V. If (v\ = (u\A. Let A^> be an operator in the space Vi and A^ space V2. i. As a consequence we arrive at the following properties (e. we write \v) = A*\u).7b) (4. the operator A^ is called hermitian conjugate or adjoint operator of A (cf.
i. It follows that a is real. [H. Verification: Since A — A^ and A\u) = a\u). .K\.3 Linear Operators. or (u\A\u)* — {u\A^\u) = (u\A\u). if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. The commutator of two hermitian operators is antihermitian: [H. The operator H is said to be positive definite. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . An important theorem is the following. so that (a)(a)t = a)(a. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa.K]^ = = [HKKH]* KHHK = = K^H^ [H. so that (n^4n) is real. if H is hermitian and (u\H\u) > 0 for all ketvectors \u).K}. Obviously the quantity \a)(a\ is an hermitian operator. as is also (u\u). it follows that (u\A\u) = a(u\u). for we had (\u){v\)i = \v){u\. K] = 0. if H — H' and antihermitian. Analogous considerations apply in the case of bravectors (u.e. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. 65 The linear operator H is called hermitian. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A].e. i.4. An hermitian operator A has the properties (a) that all its eigenvalues are real. but only if these commutators commute. hermitian The product of two hermitian operators is not necessarily hermitian.
Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.9b) (4.e.u'). so that (v\u) = 0. The above theorem remains unchanged. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). If these form a complete system. i.e. Ps is idempotent. i.9a) . then the hermitian operator defined on this space is an observable. The entire set of ketvectors spans the extended Hilbert space. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. Ps\ua. (n\n')=5nnl. so that (u\Ps = (us\. Ps2 = Ps(4. Dirac's Ket.Au) = a(v\u) — b(v\u) — (a — b)(v\u).e. (u\u') = 5{v . but the continuum vectors are normalized to a delta function.8) The projection operator has the following important properties: Ps is hermitian. as we can see as follows. (u\Ps = {us\. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). Next we introduce the very useful concept of projection operators. i. \u8*) E S*. we have 0 = (uArt) — (u. (4.66 CHAPTER 4.) = 0. (v\A = b(v\. Let S be a subspace of the Hilbert space !K. since a is real. since a ^ b. {us\us*) = 0. i.and BraFormalism Moreover. a ^ b. which includes vectors of infinite norm. A continuous spectrum can immediately be included by passing to the extended Hilbert space.e. Assuming A\u)=a\u). and let S* be its complement. ( n  z / ) = 0 .
Very similarly we can construct operators which project onto the subspace spanned by the continuum states.1. . (1 — P)\u) are orthogonal. Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace.. i. iV) is a set of orthonormalized states.e. = (p2p)\p). Let a continuum state be written £). i. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. Then 0 = (P2P)\p) and hence p = 0. c = (a\u). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN.e. P\p) =p\p). if the set of vectors 1).9c) This property can be seen as follows. i. i. where £ is a continuous index.e. \u) arbitrary. \ua) = c\a).3 Linear Operators.e. (4. p 2 . so that (Ps2 ~ Ps)\u) = 0. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1.9d) Thus the vectors P\u). Ps is projector onto S. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). 2). 67 (4. Set \u) = \ua) + \ua*) with (a\ua*) = 0. . The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw).4. The quantity \a)(a\ is called elementary projector. (1 — Ps) is projector onto S*.. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. as can be seen as follows.. so that \ua) = (a\u)\a) = \a)(a\u). Let p be an eigenvalue of the projection operator P.p = 0.
Coulomb potential. (11. n=l In the case of continuum states (which are no longer countable) one has correspondingly as projector of all states in a domain £ e [£2>£i] or in the case of the differential domain d£ of £: dp = j '£ de'ion This latter operator is called differential projection operator. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.e. Observables are representatives of measurable quantities. which are orthogonal to each other (i. If there are r such vectors. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). Degeneracy will be discussed at various points in this text.4 Observables Operators which play a particular role in quantum mechanics are those called observables.114c).6. one says the degree of degeneracy is r . ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.r\ "An example is provided by the case of the hydrogen atom. However.6. (11. see Eq.114b). 4. In general it is possible. i. Dime's Ket.. Examples 8.. u2). P = £n)(n. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors. . are linearly independent). be a system of basis vectors in this space. which we introduced earlier.. Sec.r)(ui. in this case the spectrum also has a continuous part.68 CHAPTER 4. that one and the same eigenvalue Aj is associated with several eigenfunctions. the operator P is correspondingly P2\u)= f2\0m\u).1 and 11. Let \u\).g. See e. and Eq.e. 8.
we obtain i i i i i. the operator A is completely determined by specification of the eigenvalues A. If Aj ^ Aj'. Let us set {A . ^ A i P i = A = ^Aiui.e. The operators Pi are linearly independent. (4. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r\ = l. and APi = XiPi.10) expresses the completeness of the orthonormal system.\i)Pi = 0.r \(ui. the relation (4.r)(ui. i. (4. i.e.r. Together with the orthogonality condition.r =^ i. and the eigenvectors \ui.10) This expression is known as completeness relation or closure relation.e.11) i.10) gives the linear combination it) = ^ i. i i (4.r (4. then PiPi' = 0.12) .4. in the case of degeneracy with (ui. i.r \ui.10) follows from the fact that i Applying the operator A to the projector (4.10). the projection of this operator is the entire space. which we do not enter into here.r\u) i. The uniqueness of the expression (4. PA = y£pi i = Yt\ui.r)(ui. is always implied). Applied to an arbitrary vector \u) G "K Eq.r>(ui.r)(ui.r'} = Sii>6rr>. or also as subdivision of unity or of the unit operator.r) (the convergence.r\u).r\u)\2.r\ui/.e.4 Observables The dimension of this subspace is that of the degree of degeneracy. 69 i If A is an observable and if the spectrum is purely discrete.
i. (4.70 CHAPTER 4. > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv.v'). written in Dirac's notation.and BraFormalism This is the expression called Parseval equation which we encountered with Eq. 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. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1. (4. Dime's Ket. for instance exponential functions.14) . for instance. In a similar way we can handle functions f(A) of an observable A. (u„zv) = 8{v . the eigenvector \ui): f(A)\ui) = f{\i)\ui). we have the corresponding generalizations.e.17) previously.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). If the spectrum of an observable A consists of discrete as well as continuous parts. Let v be the continuous parameter which characterizes the continuum. as explained earlier. The ketvectors \uv) are orthonormalized to a delta function. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv).e. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. i. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). (3. One defines as action of f(A) on. now however.
[A. In the onedimensional case we write {x\x') = S(xx'). (3.5 Representation Spaces and Basis Vectors 71 This relation expresses an arbitrary function f(A) of an observable A in terms of the eigenfunctions of this operator. their commutator vanishes. Eq.e. with energy E.59). Therefore we want to reexpress the Fourier transform (3.15) and (4. if (a) they all commute pairwise. i.16) represent subdi . In the following we consider more generally ketvectors \ip) as representatives of the physical states. (4.. Since both expressions (4.C.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. B].17) The Fourier transform provides the transition from one representation and basis to the other. We shall return to this theorem. \k)€Fk. again later. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem. For many practical purposes the use of the Fourier transform is unavoidable. form a complete set of commuting observables. Extending this we can say: A sequence of observables A.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. / dx\x){x\ = 1. where the symbol ip is already indicative of the Schrodinger wave function ij..5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum.. and (b) if they possess a uniquely determined system of basis vectors. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies).and bravectors.58) in terms of ket. First of all we can rewrite the integral representation of the delta function.B. if and only if they possess at least one common system of basis vectors. x) € Fx. (4. which all commute with one another. here presented without proof.4.. (4.e. as a formal orthonormality condition. 4. i. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute. for which the completeness relation is f dk\k)(k\=l.
The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. (4. Dime's Ket. . (4. (4.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy.18) According to Eq. (x\k) = ^=eikx.21) provides the ketvector \ij)) in the A. <a#) = [ dk{x\k)(k\ip). Obviously we obtain this by inserting a complete system of basis vectors of Fk.2a) etc.1a). i. (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.e. il>{x) := (x\i/}) : "K * C.e. V 27T (k\x') = . Rather \x) and \k) serve as basis vectors in the representation spaces Fx.Fk. called wave function.19) Comparison with the orthonormalized system of trigonometric functions (4.and BraFormalism visions of the unit operator. ${k) := (k\if>). The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. (3. which are representatives of the states of our physical system. we can rewrite Eq. 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).x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). (4.15): 5(x . i.L e " * * ' = {x'\k)\ V27T (4.space representation.72 CHAPTER 4. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n).e.59) or Eq.
if we do not take into account the complementary part of the universe.1 Introductory Remarks In this chapter we remind ourselves of the measuring process briefly alluded to in Chapter 1 and of the necessity to recognize the abstraction of a physical situation that we perform. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.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.e. 5.p. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. Schrodinger [244].Chapter 5 Schrodinger Equation and Liouville Equation 5. Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). We mentioned earlier (in Sec. the calculation of which for specific cases is the subject of Chapter 7. with the complementary system or *E.2 The Density Matrix We shall establish a matrix p. 1. 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. i.t). which is the quantum mechanical analogue of the classical probability density p(q. 73 .
where T means transpose: p = CTC*. (a\i)=0 = {j\b). we have to deal with a socalled "mixed state".j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. a (5. The actual and real state of a system is then a superposition of two types of states.3b) The quantum mechanical expectation value of an observable A. are called "pure states". which in Dirac's notation we can write as = J]C a »i) a.b.74 CHAPTER 5.^ Since p is hermitian. p" ( C C * ) t = CTC* = p. that is. \a). the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. i.e. \i). 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). of an operator that acts only on the states of the system we are interested in. These are the quantum mechanical states which are also referred to as microstates. called the density matrix.. we have (a\b) = 5ab.....e..3a) and N J  a ) ( o  = l in the complementary subspace. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. \a). T .b. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete.i. where i represents collectively all quantum numbers of the state of the system under consideration.i. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. i. (52) (5. but also of the states of the complementary set of the universe..j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix. Schrodinger Equation and Liouville Equation the measuring instruments. The states \i) £ "K.
This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. The ensemble can be represented as a set of points in phase space. (5..4) we had the expression (A) = Y/Pij(j\A\i) i. Gibbs introduced the concept of ensemble.9) is diagonal. i.e. quantum states.10) and is therefore.j(x'i)(ia. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. This transition to a new set of basis vectors in which the matrix becomes diagonal.3a). a complete orthonormal system with properties (5.) = YJc t .ji(x / )i*(x).5..e. e. i i (511) In Eq.e. This ensemble is interpreted as a large set of identical systems. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.x) := (x'\p\x) = yju. which today is a fundamental term in statistical mechanics. P ^ ) = i)(#) = (#)i) (5.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j. In the following we write simply \i) instead of \i'). which are all subjected to the same macroscopic boundary and subsidiary conditions. seen macroscopically of Z copies of the system).e. .. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i). a (512) *In his reformulation of the statistical mechanics developed by Boltzmann. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). as we saw earlier.* We can specify the state of our system.\p\i")(i"\ = Y. in a specific representation. in the position state representation or xrepresentation. (5. a projection operator. where x represents collectively the entire set of position coordinates of the (particle) system. i. (see below) eM. or (5. (5.3 with Pi^YtCaiCaj*.g.6) (i'\p\i") = Wi'Si'vi = real. If the ensemble corresponding to the system under consideration consists of Z elements (i.2 The Density Matrix 75 \i') 6 "K.7) p = ^2\ii)(i.2) and (5. as we discussed previously. but occupy in general different microscopic. In this sense we have p(x'. Thus the real system is in the pure state \i) with probability uii/ J2i Ui.
) := (mm = (#') W> = l(#')2 > 0. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) . (5.16) (5. (5. (5. i t (5.18) Substituting this into Eq. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf. To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). Eq. (5.3 = pvv = uv. (5.4)) (5.14) which we can write (A) = Tr(pA). according to Eq.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. On the other hand.5) above) (i\p\j)=Pij. as claimed by Eq.76 CHAPTER 5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. Then the expectation value of the observable A is (cf. 3 (5. In particular for A = 1. Hence Wj/ > 0. Eq. (A. We can also show that Ui > 0.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). (5. we have Tr(p) = l.17) Hence the sum of the classical probabilities is 1. Thus Tr(j>ii)<i)=l. (5.18).4).
We now consider the latter expression.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. Equation (5. and in the other states with probability zero. We see this more clearly by going to the position space representation. so that p{x. i. For the case n we have TJ=J (520) (5. (5. in all other cases the state of the system is a mixed state. If all ui but one are nought.t)\2. not in a mixture of states. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. by representing the vector \i) by the wave function i(x. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights. (Question: Is the state of the universe a pure state?) 5. ^2ui\i){i\.t) and instead of (x\Pi\x) we now write p(x.3 The Probability Density p(x. thus enters these states. We observed above.24) s (x\x') = 5{x .t) we now write tp(x.21) p=\i){i\=Pi.23) Instead of i(x.3 The Probability Density 77 We see therefore that p defines the mixed states. one says. but we see here. the system is in a pure state. which is contained in the coefficients Cai. i. that — different from u>i — P.e.t) = \^{x. . t) p= i The expression (5.t).22) In the following (x\i) is always to be understood as (x\i(t)}. In the following we are mostly concerned with considerations of systems in pure states. (5. since they do not take into account the (interaction with the) rest of the universe.8).x').e.t) = {x\i). represents a quantum mechanical probability. that these are actually not sufficiently general. (5.e.20) expresses that the system can be found with probability 1 in the state \i). the only remaining one is 1. Thus we have a system in a pure state.5. The effect of the interaction of the system under consideration with the surroundings. defines the statistical operator. i.
but the linear operator states are moving — even if the system is observably stationary). (5. or the wave function tp(x.t) = / dx{i\x)(x\i) = (i\i) = 1. We now want to obtain the quantum mechanical analogue.28) Differentiating Eq. We therefore postulate first an equation for the time dependence of a state vector. and hence the integral over all space dxp(x. To this end we have to differentiate the density matrix p.22) can also be written (xi) 2 = (i\x)(x\i).t) = \rp(x. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).e. 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). and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. This differentiation requires the time derivative of a state vector \i) E "K. i. (5.4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec. it is not "derived").27) is known as the Schrodinger equation. (5.^ The equation which is the conjugate of Eq.27) where H is the Hamilton operator of the system. (5.8) and multiplying by ih. (5.3. the expression (5. 2. which is chosen in such a way that the desired analogy is obtained.26) 5. t). the Schrodinger equation is always postulated in some way.8). . Equation (5.25) J Thus the particle whose state is described by \i).t)\2.e.78 CHAPTER 5. We shall convince ourselves later that the postulated equation is sensible (in fact. Schrodinger Equation and Liouville Equation The relation (5. can be found with probability 1 in the space R1. 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. (5.27) is iHJ\ jt= m . The generalization to a threedimensional position space is selfevident: JK 1 / p(x.
(5.32) Comparing Eq.e.34) . M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5. (5. we observe that here on the left hand side we have the total time derivative. dp/dt ^ 0.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H.27). The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.Xi)\j).4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.5.28) we obtain (see also Eq.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.24)). (5. (cf. p also contains "quantum mechanical probabilities" (i.p}. (3. such a formulation is described as Schrodinger picture. We can now consider the Schrodinger equation (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). We thus obtain the operator form of the Liouville equation. i. (2.p} + ihJ2^\J)(J\.e.28) the states are considered as timedependent.27) and (5. (5.21). Eq.e. (5.37)) (5. i.32) with its classical counterpart (2. (5. With Eqs. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. not the partial derivative. The reason for this is that in addition to the "classical probabilities". both uji and i)(i). Eq. the relation ih^ = [H.
since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.35) Here we can look at the Hamilton operator as a "time generator3''. t ) .Xi){x\En) = En(x\En). 10. in Sec.38) (x\j)t=o = Yt(x\E^(En\J)t=o. With the help of Eq.. we write V>(x.. „ / ith—.36) however.8) for p. T meaning temperature. **R.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. the state \i) is still timedependent. however. (5.and temperaturedependent Green's functions. it is plausible to refer to this equation at this stage. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. or H\En) = En\En).35) can be written'! ip(x.0): where H(ih—.80 CHAPTER 5. a (5.4.4. with respect to time t.x 1 ^ ( x .e.. 5. P. in p = ^2i^i\i}(i\.** In view of the close connection between time. appearing in the Boltzmann distribution. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. but with respect to the parameter /3 = 1/kT. Feynman [94]. i. (5. n (537) This equation is described as the timeindependent Schrodinger equation. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. (5.36) in in The initial wave function or timeindependent wave function ^(x. . since the solution of Eq. which we shall need later. 0) or \il))t = \il>)t=o exp (5.t) = exp ip(x. In expression (5. operator" the exponentiated oper . For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1.t) = Hld .e. to) = exp[—iH(t — to)/H\. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. we can replace these states by corresponding timeindependent states. we shall describe as "time development ator given by U(t. i.t): > ih—</>(x.
42) Inserting here Eq.4 Schrodinger Equation and Liouville Equation 81 Example 5. (5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x').1). i (5. (1. we can rewrite p. (5. (5.40). i. i ( 5 . (5. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.x')^p(x.41 ) or with 4>i(x) := (x\Ei) this is p(x.44) r (TrePH)' (5. the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. Eq.e.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT.40) 1. we obtain p(x.^ n tne position space representation Eq. or ^ = ^ e 0Ei _m. x') = ^ i LUiMxWix').5.39) Without proof we recall here that in the socalled canonical distribution the weight factors uji (similar to those of the Boltzmann distribution) are such that (cf. (543) Since H(f>i(x) = Ei<f>i(x).45) since on the one hand •PEi i i i and on the other hand .x'(3) = Si e ~^W*V).
82 CHAPTER 5. B] + L [A.50) Differentiating this equation with respect to j3.48) = 1.45) without normalization as PN(J3) := e~pH.(3). .X'. (5./?). B}} .49) In the position or configuration space representation Eq.47) (5.51) where the subscript x indicates that H acts on x of PN(X. whose calculation for specific cases is a topic of Chapter 7 (see Eq. (5. (5. We now rewrite the factor exp(—(3H) in Eq. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.35).x'. we obtain 9pN gpP) = SijEie^ = EiPNij(J3). (5.45).46) we can write the expectation value of an observable <^Tr<„. Differentiating Eq. [B.tt tt For a more extensive discussion see R.55)). (5.4> = ^ g ^ .46) is pN(x. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5. Example 5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.L [B. A}} + • Solution: The relation can be verified by expansion of the left hand side. (5.51). M. Wilcox [284].46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij.47) with respect to (3. (5.52) Hence this expectation value can now be evaluated with a knowledge of pN. we obtain = HxPN{x.x. Schrodinger Equation and Liouville Equation so that p is given by Eq. Tr(pN) = / dxpN(x. (7. (5. Equation (5. This solution is the Green's function or kernel K(x. With Eqs.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5. (5./3). [A.x'.51) is seen to be very similar to the Schrodinger equation (5. (3). (5.45) and (5.x'.(3) := (x\pN(P)\x') = (x\e~0H\x').
we shall see on the way. 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.. 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). Noz [149]." Quite apart from this basic significance of the harmonic oscillator. This is a fundamental topic since the harmonic oscillator. can — in fact — be quantized in quite a few different ways. However. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space.53)). (8. * Comment of Y. Kim and M. that the oscillator. its quantization here in terms of quasiparticle creation and annihilation operators also points the direction to the quantization of field theories with creation and annihilation of particles.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. like the Coulomb potential.51) to (8. An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs. Since the spectrum of the harmonic oscillator is entirely discrete. E. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ".. 83 . S.the present form of quantum mechanics is largely a physics of harmonic oscillators.
in the present case + [EmooJ2x2)ip = 0. For instance we can go to the special configuration or position space representation given by q^x.84 CHAPTER 6. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x).2 The OneDimensional Linear Oscillator In the case of the onedimensional. Also observe that H does not contain terms like pq or qp. To this end we introduce first of all the dimensionless quantities rriQUi fl q.' ^mgLOh by setting and *~.2) which for mo = 1/2.p. which is called Weyl ordering. which is — in the first place — representationindependent. like taking half and half.p]=ih (6.^ The first problem is to determine the eigenvalues and eigenfunctions of H. h = l. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. hence we do not use a distinguishing notation. so that there is no ambiguity due to commutation. p^ih—. However. 2mo dx \ 2 2 (6. Quantum Mechanics of the Harmonic Oscillator 6. of which the only nontrivial one here is [q.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. When such a term arises one has to resort to some definition. . linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. This can be done in a number of ways.1) for the hermitian operators q. This equation would then have to be solved as a second order differential equation. we can also proceed in a different way.to = 1 reduces to the parabolic cylinder equation.
A^] = A\ (6. With the help of Eq. + 1).8) The eigenstates of H are therefore essentially those of N := ^ f A (6.6.14a) (6.5) Reexpressing q and p in terms of A.C]. (6. We now use the relation (3.14b) [A* A.10) then a = (a\A^A\a) = \\A\a)\\2>0.e.7) (6. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . C}B. C] = A[B. (6. (6. (6. if A*A\a) = a\a).13) ( ^ A ) i t = A\A^A From Eq. C] + [A. (6.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. q are hermitian). for ^ A  a ) = a\a).11) Thus the eigenvalues are real and nonnegative. A\ we obtain h A + A^ rriQLU y^ (6. A] = [A\ A]A = A.B}C + B[A.1) we obtain immediately the commutation relation [A. i. Prom these relations we obtain ( ^ = ^ .6) and p = y/motoh AAi =^. and [AB.34d): [A.1 ) .12) in order to obtain the following expressions: [tfA.14a) we deduce for an eigenvector \a) of A^A. (6. A*} = A^A.e.BC] = [A. i. (6.A^] = 1. . (6.2 The OneDimensional Linear Oscillator 85 (p.
or Ata) = v ^ T T . (6. (6.l)A\a) = A(A^AA . A 2 a) is eigenvector of A^A with eigenvalue (a — 2). unless A 2 a) = 0. (6.l)\a) = A(a .17) (6. since this equation implies that the eigenvalues cannot be negative. we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).2)A a). If we continue like this and consider the vectors An\a) ^ 0 for all n.20) . (6. unless A^\a) = 0. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. Similarly we obtain from Eq.A)\a) A{AAU .e. Next we consider the vector A 2 a).16) This means. in view of Eq.86 the relation CHAPTER 6. In this case we have (A*A)A2\a) (6 (6. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. or \\A\a)\\ = Va.. This would mean that for sufficiently large values of n the eigenvalue would be negative.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1).l)\a) = (a . we must have (a) Let \an):= Ana)^0. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . (6. (6.21) . unless A\a) = 0.19) i.l)A\a).11) this is not possible. (6. Thus for a certain value of n > 0.18) = '= a ) A(A^A . However. but (b) An\n>\ llA An+1a)=0.2A)\a) 2 (a .14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a).  a) is eigenvector of A^A with eigenvalue (a + 1).2A)\a) = A(Aa .
25) (6. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a . (6.20) that A0) = 0.5) to shift operators A to the right which then acting on 0) give zero. (6. (6. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n.17) a by (a — n). \ I 2 2 x 87 Replacing in Eq.28) . so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1. For a — n = 0 we deduce from (b) of (6. In the following manipulations we always use the commutator (6.22) With relation (b) of Eq (6. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0.e.n  a .20) we have for a = n = 1: A2\1) = 0.20) we obtain from this that the right hand side vanishes. From relation (6. (6.n ) = " A»a>=» ' = A«a)F = 1 . (6. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. \ 9 \ I / II 4 T ) . I „ .2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.. (6.27) According to (b) of (6. that an = 0.' " /.6.I n U .24) This is a very important relation which we can use as definition of the state vector 0).. The state 0) is called ground state or vacuum state. Moreover.18) we obtain for a = n: p t  n )   2 = n + l. v. i. or a = n > 0.
(6. (6. as we can see as follows.11). l>ocAt0).32) we have (nm) = .34) (6.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2.^ ( O l A ^ n o ) .33) . The states \n) thus defined are orthonormal. we have 2> = ^=dAi\0).11) we have [A. According to Eq.e. (6.32) (6.88 and CHAPTER 6.(A*)n] = n(Ai)n\ (6.31) (6.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1). A^} = 2A\ and [A. and in general [A. (6. A^A* + A*[A.5) and again (6. But using Eqs. (6. i. (At)2]A+ + (A*)2[A. Similarly we find 2)oc^Ut0). and in view of Eq.26) the equality l) = A t 0). (A*)2} = [A. (A^)3] = [A.29) Hence we obtain and in view of Eq. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0.(6.29). (0^ 2 (At) 2 0) = 2. In general we have \n) = ^(A^n\0) (6.
however. Here. the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . whose number is given by the integer eigenvalue of the number operator N = A^A. i. .39) A^A\n) = y/nA^\n .e. The terminology is. ldnm = n\5nm. 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) (6. l .L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.1) = n\n).1 (A + ) m . With Eq. •(» (6.'n\ and = v n ![ ( A t ) M i= + (6.32): rf\n) = . (6.34) A\n) = ±=A{Al)n\0) . In view of the properties (6.1 0).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 . (6.35) Inserting this result into Eq. (6. we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle").8) we obtain therefore H\n) = hu (A^A + ].36) + 1) (6.33).6. n = 0 .38) (6. chosen in close analogy to the postulated "second quantization relations" of field theory. .40) (6. .37) n^)711}^) = v n  n . .J \n) = hw (n + i J \n). In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv. . we obtain (n\m) = 5nm We also deduce from Eq. .41) The contribution hu/2 is called zero point energy. (6. 2 . in quantum mechanics. (6. .1 ) .34) and (6.35) the operators A^ and A are called respectively raising and lowering operators or also shift operators.
n = 0.e. is defined by projection of operators on eigenfunctions of energy.90 CHAPTER 6.. other elements zero.1.39): {n\A.46) v^ .. (6.38) and (6. i.. ( 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. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn.g.44) yfn. (5. also called Heisenberg representation. other elements zero. and — in fact — we have used this already previously (see e.3 The Energy Representation of the Oscillator The energy representation. A^ from the relations (6. Eq. Quantum Mechanics of the Harmonic Oscillator 6.6) and (6.7) the energy representation of the operators q and p. (6.. En = hu(n + ).'\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 .45) Correspondingly we obtain with Eqs. (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. (6.42) We can deduce the energy representation of the operators A.2.43) = = y/n(n'\n — 1) = y/n5n^ni.39)).. Vn + 1.
24). (6. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). i.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}.3)) ^ (q 2n \ + (6. The ground state wave function (/>o(x) is defined by Eq.e. 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.47) LP 0) = 0 .5) are also satisfied as matrix equations.1) and (6. (6.4 The Configuration Space Representation We saw that the eigenstates are given by Eq. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w . 6. (6. Eq.32). ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.6. 2 . by (cf. (6.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs. mow / (6. A\Q) = 0.
l ) > n ( x ) . Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. so that <t>n(x) = ( .2 n n! The functions Hn(£) are obviously polynomials of the nth degree.56) < ^ = We have as an operator identity the relation .l ) n f f „ ( 0 .e. (6. (6.L ( x  ( A t ) »  0 ) . they are called Hermite polynomials. i.32) to apply the appropriate number of creation operators A^ to the vacuum state 0).0 = ( . ) e"^2.92 CHAPTER 6. Vn! Now {XlA (6. (6.54) (6 55) i/4 rzri7r 1 / 4 vo.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 .51) \l 2h {X q \ m0uP)\l 2h \X m0u. c/>n(x) := (x\n) = . In order to obtain the wave functions of higher states. These polynomials are real for real arguments and have the symmetry property # n ( . it suffices according to Eq.
Magnus and F.12.g. Oberhettinger [181]. .12(20 .57b) • Generalizing this result and inserting it into Eq.e. 93 . (2036(20. e. ( 2 0 4 .l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.4 The Position Space Representation i. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.en_ dt.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( .2)m = (±t)m for some function VKO.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. p. (6. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6. (6.58a) ' T h i s definition — apart from the usual one — is also cited e. d£" (6.6. (202"2. 81. 2£.58b) by W.
81. Magnus and F. £. (6.^ + n) He n(0 = 0 (6.94 CHAPTER 6. (6.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) .59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.59b) = (l)^2/2^.6 £ 2 + 3. Oberhettinger [181]. (6. pp. £33£.60b) J—< "See for instance W. . = nHe^iO.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.He'n{£). Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. £ 4 .(e^/2) or Hn(0 = 2n/2Hen(V2ti). 80.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. (6. £2l. = ffen+itO+ntfeniCO.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. (6.
s) = eM?(Zs)2}(6.g. s) as a Taylor series.. F(S. Magnus and F. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.0) 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(0.61) e?l*2nl2Hn{Z). 93. and so on. (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£) = ( . Their normalization is therefore given by OO /"OO / OO J —OO z (6.64) n\ See e.. Thus the generating function can be written F(Z.l ) ^ e^HeniO 2 / 4  . .s) = J2 n=0 (6.6. p.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.s)=F(£. Oberhettinger [181].^ = e^/42~n/2Hn(aV2).63) The meaning is that if we expand F(£. W.
But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. has parity + 1 . for x(t) = 0 and p(t) = mox(t) — 0. 2 . i. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku).e. In classical mechanics the energy of the oscillator can be zero.96 CHAPTER 6. We see t h a t the lowest state is symmetric. 6. We also see from the eigenvalue t h a t the energy of the lowest (i.e. in order to estimate the lowest energy of the harmonic oscillator. i. and we see t h a t even and odd states alternate. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. ground) state is fko/2. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 .1. Quantum Mechanics of the Harmonic Oscillator Fig. In fact we can use the uncertainty relation AxApx ~ H/2. 6.e. mow' Apx = ^m0Luh/2.
Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. in corrected version instead with n* = n + 1/2. like twoatomic molecules. Example 6.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6.: 27 T mov2a2. 14. classical orbit i. n*h. setting q = 1/z. which is the complete orbit. original path q . and is directly observable. In Example 6. to 0. mass mo. It is clear that the uncertainties are not due to deficiencies of measuring instruments (we are concerned here with a system in the pure state 0)). /classical orbit Here q goes from 0 to a. But it has a pole at infinity.1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf.plane pole at z=0 ^ . The kinetic energy is always positive or zero. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u.plane Imz z .6.4) to obtain by contour integration the eigenvalues of the harmonic oscillator. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). E . back to 0. 27rmoi/ <f> V'a 2 — q2dq = n*h. 6. In the plane of complex q the integral has no poles for finite values of q. to —a. we obtain the integral .4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators.2 The original path and the transformed path at infinity. Thus.e. Sec.1.. pdq. The analogous Example 6.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity". This observability can be taken as direct proof of the uncertainty relation.1/a O l/a Rez all at infinity Fig. According to "old quantum theory" with integer n.
In the following we investigate in more detail the important differential equation of the harmonic oscillator. After removal of a Gaussian or W K B exponential. and obtain with this all properties of the solution. z = 0 Vi 2Kmov(Ka2) = n*h = E/v. Example 6.) hv. (m . 6.1)! (6..e. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o . i. 6. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i. / " ^ .V62 a2). .2b = Ze2.2TTI a ^ = . harmonic oscillator and Coulomb interaction. a J \d Solution: The solution proceeds as in Example 6.1. E = Z2e4/4(l + n+ l ) 2 (cf.5. i.( 6 .5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials.98 CHAPTER 6. Eq.2: Contour integration along a classical orbit Setting q = 1/z.114a)). (11. With a2 = E. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). are the most basic and exactly solvable cases in quantum mechanics. but here we leave this . we start with a Laplace transform ansatz for the remaining part of the solution.a) m = 2 .65) f ^Va2z2 .5 . in q at infinity).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. hence we consider the following differential equation in which n = 0 . since these two cases. « qdq + I2 = f dq . .3.c2 = (I + 1/2) 2 (natural units.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . 2ft a2 + 9 c 2 ! / 2 27T In f j — (ac — b) and I3 = <t \a2 r 2b _ c^n/2 = ^( a 3 6 _a2c2). 1 . £ = n*/w = ( n + . for instance — without deriving all properties of the solutions. .e. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . 2 .
(6. fp{ps{p)) = {v2 + n)s{p). (6.i ^  (ps(p)) P n P Integrating this expression. i.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.p t dp = \ps(p)ePt} . Eq. E.68) § = J p2s(p)e*dp. We demand that C(p) := — \ps(p)e~pt] = 0.g. Then (with partial integration) *dt = ~*/Ps(p)e. / dpf(p)e~px the Laplace transform of f(p). p. (6.6. . . 80.111)) the Mellin transform of e~p is n!.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 .g. and / dpf(p)pn the Mellin transform of / ( p ) . Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. (11. (cf. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated).t*ij& + ny{t) = O. W. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. This has to be true for all values of t.J ±(ps(p))e*dp. Oberhettinger [181].69) **See e.e. so that the integrand of the remaining integral vanishes. Magnus and F. (this determines the limits). Substituting these expressions into Eq.66) First we have to remove the £2term. (6.
(2) If n < 0. This exponential assumes a maximum value when pt \—p2 = minimal. The value of the exponential at this point is ~ exp(£ 2 /2). The quantum mechanical wave function is then (cf. (3) We can choose a contour once around p = 0 in the complex pplane. so that y(t) ~ exp(i 2 /2). Hence a possible path of integration > is from — oo to oo. We now investigate possibilities to satisfy the boundary condition C(p) = 0.71) . we can choose a path from zero to infinity (the condition will vanish at p = 0). so that p cannot go to infinity. Eq. consider the exponential in the integrand. For singlevaluedness.n+l between the two limits of integration.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. To see this.70) = 0 P. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6.3. as indicated in Fig. of course). since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. i. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. We can show that if \p\ is allowed to be large. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6.. when p = —t (which is possible. Therefore we take ptp2/2 y(t) . (1) This will be satisfied if \p\ — ±oo. n must then be an integer. This can be attained if p is allowed to be large (t can be of either sign). = / • Im p Rep n not an integer Fig.e.100 and hence CHAPTER 6. 6. then for large t we have y(t) ~ exp(i 2 /2). (6.3 The path when n is not an integer. 6.
l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp.. We can also obtain the derivative form of this Hermite polynomial. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt).5. if n is even and v = 1.0) p p»+i v ' Hen(t) = (l)^ e * 2 /2 / 2vri /(9=t) dg " ^(gi)"+i e ?2/2 = e i 2 / 2 ()ne"i2/2.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0.if n is odd. . Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. (676) .P 2 / 2 00 C—tY ( . . n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n . this becomes jfj.. . 4 . pt p2/2 (6. In y(t) above. 2 . . The exponential function here is described as the generating function of the Hermite polynomials Hen(t).6.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . We have 27T! Setting q — p + t.„y2 v l { ' ^ r (^) \ 2 ' where v = 0.3..
7 that we use throughout this text) and their orthogonality and normalization to Examples 6. we obtain tHe„{t) (l)"n! ~ i.77) = t.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t).g.1 '2dt using Hen(t) = (l)n27 r . Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. for the evaluation of expectation values. Hen(t). n — l)Heni(t).)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . (6. n + l) = l. one can construct the tower of polynomials Example 6. 8.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. Starting from Heo(t) = 1.102 CHAPTER 6. Hei(t) Henj. in radiation problems. Reinterpreting the remaining integrals with Eq.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.n + l ) f f e n ^ i ( £ ) + (n. as in the normalization of asymptotic expansions of Mathieu functions). Solution: We multiply the generating function by a factor t and perform a partial integration on the factor exp(— pt) contained in the following integral (the differentiation of the other factor leading to two contributions).59a).4. Example 6. (n. Solution: We have to evaluate °° 2 . Pptp 2 / 00 /2 dp.70). (6.3 and 6. (6.75) as Hermite polynomials. and elsewhere. Example 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+<.n— 1) = n.e. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6. ( n .
66) is \Zn\V2w Example 6.5 The Harmonic and hence Oscillator Equation 103 / = (!)" But (2iri)2 { J Jp=0 J q0„ pn+ i0m+l P I e d d P 1 1 / epq p" <k rrdq = coefficient of qm in epq = — 27ri / qm+1 m Therefore (IT ^n+m ^ n l I Jp=o ' v d P ' 27U pnm+1 =2ni if Ti=?7i.5 nm . (6.5: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy.^ + s 2 ) / 2 f°° 2 e(P+9+«)*~t /"GO 2 /2di 2 J — oo o = e " /2eP9e"(p+9) / e(t+p+9+a!) /2d4 J —oo ^/^Tga /2e9Qep(q+«) ^ gl^ The coefficient of p n in / ( p .6.q). Hence the normalized wave function of Eq.78) \/27rn!.n)\ i s an+2rij.82) ^ .jQ) n n!I ^—' ^Q fi\(n .q) = = e . and 0 otherwise \/2n(—l)n+mn! if n = TO./(n __ 2r — /i)!. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (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). (6.80) f(p. zero otherwise (6. so that /2 Q2(n+rrf in / ( p . g) is n! Next. 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. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.
ra + 3)(n + 3. 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. n . n + 2){n + 2.79). .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .A o?(.e. i.' .( n + Jr). /i = (n — v) _ / r o o e .78).l ) ( n . n + 2)]ife„ + 2 (t) + ••• .r . n)(n. .^ / 2 * .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h .A i ) ! (6. so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.v)\ ^ Multiplying both sides by the value of the normalization integral.n + 2) + (n. i / ! ( n .i / ) ! ( 2 r + i/)! > ..v)\(s . i.e. The result then follows with half the normalization and orthogonality integrals (6. n. n — 1) are given by the first index. n + 2) + (n.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!. 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.(t)e«2/2dt Now. A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ . 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.n+ l)(n + 1.— — ^ . n + l ) ( n + 1.104 CHAPTER 6.r .u)\ 2s<—" v\(n . n ) ( n . n + l ) ( n + 1.^ ) ! ( n + 2 r . K is the coefficient of a / ( 2 s ) ! in the expression J .e. n + 3)(n + 3. we obtain the result (6. the relation (n. 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) + .Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . n + 2)(n + 2. i.77).v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .83) = (n + 2r)!e a / 2 V „ „. 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." ) a2^) r coefficient of (2s)! in f^ (s .l .3 we know that upgoing coefficients are 1 and downgoing coefficients (n.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . by n!(7r/2) 1 / 2 .e '' ^ o^M ! ( n . n + 2) + ( n . n + l)(n + l .2s _oo ( a 2 / 2 ) i . Prom Example 6. n + 4)_ffe n +4(t) + [(ra. the coefficient of T 2s »{sr u)\ As we observed.
In particular we derive the timedependent Green's functions for the case of a free particle and for that of a particle in an harmonic potential. in Chapter 21 (with a different calculation). Let H^ = E^> (7.1) be the timeindependent Schrodinger equation which has to be solved.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. and those which are timeindependent. we consider the sojourn time of a quantum mechanical particle at a point. This example also offers a convenient context to introduce the inverted oscillator potential. which we write here K(x. more generally. which we write G{x. of a differential equation of second order. which is a point of unstable equilibrium for a classical particle. We 105 . and is thus an important point to be noted here. With reference to this case.x'.Chapter 7 Green's Functions 7. which is normally not discussed.x'). in Feynman's path integral as the kernel or free particle propagator. We consider first timeindependent Green's functions. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5.t). 7. The first of these cases will reappear later. These are the Green's functions of the timeindependent Schrodinger equation or.
the relation X') = 5{x . since the generalization to higher dimensions is selfevident.g. Eq.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix).x') can be written t G (7.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case.x'). and Hi is some other contribution. for instance. .x') by recalling the completeness relation.EW)xG{x. e.e. (7. of the kinetic part p2/2mo and some part of the potential. although the explicit calculations can be much more involved. We can obtain an expression for G(x.13). cf.g. The boundary conditions which G(x. x') has to satisfy correspond to those of * (to insure e. aq2.x'). i.x') is defined by the equation* (Ho . e. (7.e. 'The case of £(°) equal to some Ei is considered later. where £ ( 0 ) = lim E.g. appropriately decreasing behaviour at infinity). The timeindependent Green's function G(x. square integrability. Green's Functions H = H0 + Hi. a perturbation part like (3q4.106 assume a Hamiltonian of the form CHAPTER 7. i. (7.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . (7.x') or ^2^(x)^*(x') i = S(x . We can readily see that G(x.2) where Ho consists.
£(0) .7. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei. i (7.E^ of the eigenvalue. x'\ t) (7.t) (K for "kernel") is defined as solution of the equation ih—K(x.EM)X*(X) ( .Hi{x'))^(x').E<®)xG(x.x.#.x'.x'.x'){E .8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).11) = = 3) = fdx'(H0 .x'. .t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x'). which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .7) with the initial condition K(x. x'){E . For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a. (JL x'\ t) = H0K(x.t) since ih^K(x.x'.10) The contribution on the right hand side is the socalled perturbation contribution.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x. (7.t). The timedependent Green's function K(x. (7. We can see the significance of the timeindependent Green's function for the complete problem as follows.(*'))*fr') Hi{x'))^(x') f dx'5{x .x')(E .) = f dx'G(x..E® (EE{0) Hi{x))m{x). (7.9) = i H0YieE^i%(x)^(x') = H0K(x. .0) = S(xx'). We see that the initial condition (7.E^ We can check this by considering (H0 .
W. that (E . Examples in various contexts have been investigated by L. 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!). Booss and D.108 CHAPTER 7. (7.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution.x'. (7. H.12) Equations of the form of Eq. then at best such a contribution would be something like cipi(x).11) are called (homogeneous or inhomogeneous) Fredholm integral equations. In general. see E.14) We can now see the significance of the timedependent Green's function K(x. Green's Functions Can we add on the right hand side of Eq.6) is not defined.. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). All such methods of solution are based on the analogy of Eq. (7. Merzbacher [194]. MiillerKirsten and A.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). however. This will be different. (7.e. .t) for the complete problem as follows. if we assume that E^ = Ej. 17. The theorem is also known as Fredholm alternative. D. i. But then the Green's function (7. J. Bleecker [33]. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. (7. Sec. integral equations are more difficult to solve than the corresponding differential equations. Wiedemann [183].6) we would have (7. Maharana. But if we assume that E^ ^ Ei for all i.11) with a system of linear equations of the form Vi = Mijyj. 2nd ed. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). see B.15) ''This problem and its circumvention can be formulated as a theorem. Instead of Eq. In cases where an inhomogeneous contribution is given.
i.g.The solution of Eq. We use this in Sec.5. this relation provides the probability density \ip(x.17) as initial condition is obviously W)t = Y.21) dx'K(x. p. (7. 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. (7. t) with the Note that when K(x.20) the Green's function K(x. E. (6.t) = J2 fdx/^n(x)^*n(x')^(x'. for the oscillator potential.t) describes the evolution of the wave function from its initial value ^(a^O).17) is the initial condition of \4>)t.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. t > 0. (7.0)eEnt/in. (5. x'.15) which contains (7.19) i/>(x.18) or. We can write this expression also as§ ip(x.Q) = 6(xx'). *>0.x'. 2nd ed. (7. 1 See e. (7. (7. 109 (7. 7.t) = ^ e £ " f / « V n W < ( ^ ) . in different formulations. (7.' which obviously satisfies the initial condition K(x..22) According to Eq.x'.2 for the computation of the sojourn time. \En)(En\ip)t=oeE^\ n (7. t)\2.7.t) is known. x'. H\En) = En\En). Comparison of Eq.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.t)il>(x'.t) = where if(x^'.158.20) is the timedependent Green's function. .2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. e. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7.g. Eq. Merzbacher [194].7) with Eq.0).e.x'.16) As usual.
Green's Functions difference that (5 = 1/kT plays the role of it.25) in agreement with the timedependent Green's function . between (with E^0' = E) G(x.x'.t) K(x.t). we obtain / W : = . As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. (7.1 The contour of integration.«^«2 W . x'.x'. t > 0. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.e. Inserting for GE+ie the expression above.1. t) = J2 e £ " f M i ( x ) < ( x ' ) .i /f e «/.110 CHAPTER 7.E . With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. i. we obtain then also the corresponding density matrix.23) along the contour C in the plane of complex E as shown in Fig.x') = GE(x. 24 )  ReE Fig.x') and K(x.ie (7 . Jc 27T „ hn . We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. In the following we shall derive the Green's function for the case of the harmonic oscillator. 7. x')6(t) (7. 7.
K(x. In this case the Green's function is the solution of the equation d hK(x. i. (7.t). (7.7.7). (7.x'.e.26) This is the case of a free particle with mass mo.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.28) B(xx')2 t2 DB(xx') 2 /t (7.x'.30) into Eq.x>.X>.29) tJJ 2 2B(x . v ' 2m 0 v h i.0) = 5(xx').— d2 —2K(x. we obtain *I£I = h2 —(2AB). B m0 n _ 2ih 2iK (7. which is moving in one space dimension.30) Inserting Eqs.31) The constant A has to be chosen such that K(x.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.3 The Green's Function of a Free Particle 111 7. It is clear that it is nontrivial to solve an equation like Eq.e. / dxK(x.t) t h2 = .0) = 1. . 2mQK h h2 ih(AB) = —{4B2A).29) and (7. and identifying coefficients of the same powers of t on both sides. A and B being constants. (7.27).x'. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7. Thus we try the ansatz.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.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.
t) into Eq.H We can insert the expression for K(x.33)."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. Green's Functions For parameter values such that the following integral exists. t) = /jJo_ e mo(««') 2 /««. (7. p. (7. (7. j3 = i(xx'). v27r En^^. from K{x.zEnt. 2m0 (7.21).33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. we > have to make the replacements V ^ fdk.oo 27T (7. (7. Mx)^M^) „ n so that J =^ .. 32) It follows that x'.ihMxWn{x').x'.(7.x't) = Y. 158.e.38) . "See the excercise in E.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. i.t) — for instance for a wave packet given at time t = 0 of the form ^(x. we have This is 1 provided A=JB= ^ (7 .x'.34) K(x.20) and can then obtain ip(x. *>0. Merzbacher [194]. (7.112 CHAPTER 7.35a) K(xy.0)oce"Qx2+ifcox.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. n For a free particle moving in the onedimensional domain \x\ < L — oo. 1st ed.t) = — / dkeak2+Pk 2vr J.
. x'. x'. z . 5{ax) = ± J eikaxdk = ArS(x). x'. .x'.7) is the solution of d ih—K(x.t) + ~x K(x.3 The Green's Function of the Harmonic Oscillator 113 The result (7. / ) = ~^K(x. (7.t) 1 + m0u2x2K(x. 1 2mo u>n ox n with the initial condition (7.e..——K(x.X .e. §jK(x.x'.40) Then Eq.4 Green's Function of the Harmonic Oscillator The next most obvious case to consider is that of a particle subjected to the harmonic oscillator potential.8). (7. i.t).x'. We rewrite this initial condition in terms of £ and use for a = const. (21. / ) (7. —ih—K(x.x'. (7. mQLu .7.e.37) will later be obtained by a different method — see Eq. .40) becomes 2 . the relation* 6(x) = o<y(os). .0) = 6(xx') at / = 0. K(x.42) .25) — in the context of Feynman's path integral method.39) In this case the timedependent Green's function K of Eq. (7.43) 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.x usn ot i.d r./ ) + fK(x.t) = h2 2 d2 TW . —KK(X. i. in this domain *Recall that S(x) = ± f eikxdk.e.. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.x'. 7. . i.33).t) We now set h2 d2 = . I (7. .t). (7.. HQ = ^—p2 ZrriQ + \m0uj2q2.
(7. f i.42) and obtain the equation (with a1 = da/df etc. c(0)^.46) If we interpret Eq.Z'.47) We insert this ansatz for K into Eq. b' = Aab. that is. p. (/ = 2a.50b) (7.f) with a(0)»y.e. Feynman [94].48). Green's Functions to the particle behaves almost like a free particle. P. we must have /o = 0.e. it • is suggestive to attempt for K the following ansatz^ m.^ .50c) (7. as the limiting case / — 0.b2.x'. so that for / — 0: > a^—.33). (7.4a 2 )£ 2 .51) See R.4a&£ + 2a . and we expect K(x. (7.50a) we obtain (7.4 a 2 .c o t h 2 / .45) in accordance with (7.45) in this sense. 50. (7.. a = . Integrating Eq. (7.114 CHAPTER 7.50a) (7. . K { ^ f ) ^ ^ e ~ { ^ ' ) 2 / i f f r ° ^ ' (7 45> The same approximation is also valid for large energies E and for t or f small (near zero) in view of the relation AEAt ~ h.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.t) become similar to expression (7. Identifying coefficients on both sides. i. W ) .48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.b2.) a'£2 + b'i + c' = (1 . (7.47) becomes (7. we obttain a' = l .
45).51) for the density matrix PN(X. we must have A = ('. c(0) = £' / 4 / . i. Felsager [91]. as one can verify. In order to satisfy Eq.7.t) = . we must have (besides A = —£') = /m 0 u. (7. (7. (7.c(f) K = 5 =exp Vsinh 2 / into Eq. as we shall see in the following. and to ensure that we obtain the prefactor of Eq. (7.X'. B independent of / . x . x'.t) to obtain this element (x. (7.e. Finally Eq. p. we can use K(x. 174. 2TT^ Inserting a(f). / —————rexp y 27rmsin(u.48) 6(0) = — £'/2f.55) 'For an alternative derivation and further discussion see also B.27).47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. .In B. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . (7. t. Comparing the Eqs.54) For t —• 0 this expression goes over into the expression (7.x'. (7.b(f). 115 To ensure that in accordance with Eq.x'. with A.50c) yields for c(/).i) (7. K(x./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.* With this result we have another important quantity at our disposal.0). i.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. in particular for the derivation of the sojourn time in Sec.2. (5.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq.e.53) or.33) for the Green's > function of a free particle. 7.48).5.50b) Kf) = —r^77) smh 2 / A independent of / . (7. if we return to x.40) of the timedependent Green's function with Eq. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x.
56a) {q2) = = Y.52): Tr (PNq2) Jx2pN(x. <2 (7.. Green's Functions With this expression we can evaluate (cf. )x Inserting into Eq. at temperature T): {A) = Tl{pA) = ^ > . = jdxp(x.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf. 11 For T ~+ 0: u>0 .x.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.P) 2. P. Thus the system is in a mixed state and the expectation value "Cf.p)dx W ) = —^r = —p . we skip the algebra here.x.uij>0 >• 0. (5.40)) 1 epEi 1 Ui = y. (5.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. (5. eP E i = i + y.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. p.55)..56a) the expression (7. (7.52)) the expectation value of an observable A in the canonical distribution (i.x. Eq. Feynman [94].» l. we obtain § < Z> = ^ c o t h ^ ^ ° — .e. ^—• Irpiv J pN(x. .116 CHAPTER 7. 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"}. Eq. For instance we have with Eq. R.
e.21).x'.QUJ (7.54).7. 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. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.J (w — ie)< e (n+l/2)Et e i(n+l/2)u. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . x'.54) is K(x. x'. We return to the Green's function (7.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. In this case the Green's function K(x. 2 + x'2)le^2xx' to—>w—it t—>co 2 .59b) n+ l \ ftwt here 2hf exp _ i ( n + .t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . which means in the oscillator state \i) with eigenenergy fku(i + 1/2). This is the first and hence dominant term of the expression (7. the expression for (q2) would be:'I (fti = J(i\x)x2{x\i)dx j(i\x){x\i)dx (Q2)o = h 2m. If we consider the system in the pure state \i).3 The Green's Function of the Harmonic Oscillator 117 (7. and we replace to by LO — ie.56b) is that with respect to this mixed state (whose cause is the finite temperature T). (7. (6. of K(x. We assume t > 0 and t —> oo. i.50) we obtain. e > 0. setting a = mooj/h. t) of Eq.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.59a) E 0 = ^fiw. 1 mo^ exp lmow / 2 2 ft X (7.
i. to infinity) the contribution with n — 0 dominates.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. In the following we want to calculate (more precisely estimate) with the help of the Green's function the time interval T which a pointlike particle can stay at the maximum of the potential before it rolls down as a result of the quantum mechanical uncertainties. (7.1 we estimate T semiclassically. 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 = 0.62) . k = k. the velocity of planes of equal phase.59a). as in Eq.e. i. 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)].x'. the particle will stay there only for a finite length of time T. is defined as vv = . In a very analogous manner we obtain the solution of the equation for the density matrix. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. of dpN as pN(x. In Example 7. Considered classically.e. Green's Functions For t large (i. which are planes.118 CHAPTER 7. (7. (7.5. However quantum mechanically in view of the uncertainties in position and momentum.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. lie on surfaces.e. The wavelength A is given by The phase velocity v<p..
e. (7. / oo J—oo . the fundamental postulate on matter waves in Chapter 2): E = hu. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. ' t) dk'. i. The wave packet describes a wave of limited extent. e . or dco' da The centre of mass determines the particular phase. t) is the spatial Fourier transform of this Gauss distribution — as we know. i ) = /*/(k')e i(k '" r '' .7. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.<p(k"))}dk'dk". i.)ll/(fc")l[cos(¥'(fe. to the function ip given by oo / fik'y^'x^dk'. xt— + — = 0. We now ask: How and under what conditions does the time variation of the function ^(r. then ^(r.k ') 2 . If we assume for / ( k ' ) a Gauss distribution. duo' da . u = w(k). oo (7.4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf.Q ( k .Jt + a. i. (7.e.63) The relation u = w(k) is known as dispersion or dispersion law. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.e. ^ ( r . essentially again a Gauss curve. for which \ip\ assumes its largest value: OO . e. a > 0 .)¥'(*:")) +ism(<p(k') .65) Let f(k') = \f(k')\eia and <p := k'x .64) where / ( k ' ) differs substantially from zero only near k' = k.
(7. we obtain the deBroglie relation p — hk. In the present case. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. Example 7.120 CHAPTER 7.69) In general one uses the theory of functions for the evaluation of this integral.1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e .— g(k). 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 = . dE v 9 = . For E = hio. (7. = — (768) It is instructive to consider at this point the following examples..fc w> vv = ~a^ = S r a d P E «i7 Sd u . 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)]. we can use a simpler method. . p = hk the group velocity is equal to the particle velocity v. however. The threedimensional generalization is evidently du . .e.^ g=r a rad. i. Solution: The function to be calculated is the integral dxeax OO ezkx. as claimed for =0. Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. a > 0.
69) and (7.. g . 2 f + .72) Solution: From Eqs.2: Representations of the delta distribution Use Eqs. J. Example 7. (7.69) and (7.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x . \a — b\ (7.r „• With the help of Eq. (7. with partial fraction decomposition.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. (7.73) 27T 7 .73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . Appendix A..0 0 7T 7 0 7T e From Eq. XZ (e > 0).&)] for a + b. .. MiillerKirsten [215].75) ' E . (7. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 .4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a ..a) + 6{x .7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example. where c = 9(0) is a constant. (7. or see H.Afc = 81n2. (7. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 .3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.71) (7. W.74) Example 7. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. (7.1 verify the uncertainty relation Aa..70) we obtain 6(h) = lim — f°° dxeax2eikx = .7.
In quantum mechanics the square of the modulus of the wave function ip(x.* o ) 2 / 2 . Solution: In Fig.78) It follows that the product of the uncertainties is AxAk = 8 In 2. 7. when Ax is very small. Physically .2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. The wave function corresponds to the function f(x) in the above considerations (e.« 2 ( * .e. 7.CHAPTER 7.e. Thus a sharp maximum of the function f{x). is Afe = 2 V 2 1 n 2  (7.76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. Green's Functions 1/(27i)1/2a — 1/2(2.g. According to Eq. at time t = 0).77a) small values of a. (7.) leads to a very broad maximum of the Fourier transform \g(k)\.77b) (7. where g(k) = max<?(fc)/2. The breadth A/c of the curve g(k) around k = kg.77a) g(k) = e .70) the Fourier transform of f(x) is (7.2 The Gaussian curve. Thus a slim maximum of the curve of /(a. i. the width Ax is a measure of the uncertainty of the probability.r)1/2a Re f(x) Ax Fig. where  / ( x )  = m a x  / ( x )  / 2 . and hence implies large values of Ak. (7.t) is a measure of the probability to find the particle at time t at the position x. i. requires according to Eq. A simple calculation yields A i = 2v/2~In2a.
2 ) and U)t. 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. O ) .".u> = l.. The result (7. which means all values of the momentum are equally probable.. As G. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. 7. For reasons > of simplicity in the following we set in addition t mo = l. E. ^In Eq.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. the less precise is the determination of its associated momentum. the wave packet has its centre of mass at the origin and has the following Gaussian shape as the pure initial state: iP{x^) = ^L=ex2l2b\ (7.h = 1.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. so that Ki0(x. (7. <7*» We assume that initially. 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. Barton [15].e. The function g(k) is therefore described as momentum space representation of the wave function f(x).81) *G.71) > and (7. (7. (7. In the same limit the momentum uncertainty Afc grows beyond all bounds.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x. .75) implies therefore: The more precise the coordinate of a microscopic particle is determined.79) According to Eq.5.7. Thus G. In the limit a — 0 we obtain from Eqs.«> = / ^ C ' W .76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . This means the particle is localized at x = 0. i.54) with the substitution ui — iu.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} . (7. Barton mentions at the beginning of his paper.x'. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. at time t = 0.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . nr = S(x). (7.
80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .124 CHAPTER 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 (7.86) Evaluating the Gaussian integral of Eq. (7...81) and (7. 9 (7. + cosh 2 1.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .79) into (7. A .t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.82) i/f(x.t) = J oo P .cosh t sinh t' Wisll2b ^ (7.84) we obtain* ip(x.88) .(Ax' + Bxf \/27risinh + (B2 + ^cothi)a. = .83) ip(x. Green's Functions Inserting (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 .(i/2) sinh 2t 2 2 2C 2 (i) + *c^.85) = —r= o2 7 2 sinh 2i. .87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7. (7. ' 2\Aswht % cosh t (7.2 (7.
t)\' 2 125 (7. t) the phase ip drops out and one has (x.4 The Inverted Harmonic Oscillator The real part of Eq.94) / dxf(x)=g'(y)f(g(y)).89) = exp[x2/C2(t)} 7rV2C(t) (7. The probability for the particle to be at time t still within the distance I away from the origin is plausibly given by Q(l.£ 2 (7. In taking the modulus of ip(x. •&(t)=//c(t) = / or. (7.95) . _2 1 exp C 2 (i) = l + 2sinh 2 i.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1.t)= where Co(t) / dx\^(x.t)\2 Ji == 71 / V " Jo dO ? 6' (7.. We have therefore . since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.90) with the expected limiting behaviour \im \ilj(x.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .91) C(t)' (7.92) The sojourn time T is now defined as the mean time T given by T:= r*° ioo 2 (M)s r*(f4 d^e.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).
96) °° l Jr.126 We set I C(t)' so that dT ~d~l Now. 136.From this we obtain C(t) C'(t) Z2 ^(r?2_.97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.(4 2 . V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.98) We insert this into Eq.1 ) 2 \n J ^"•""ff^^V^K?. B. + 0(e'2) (7.e ~' V^~i Jo Vn~i Jo ^ rx=VM _Jl 2 Using the following expansion of the integral* _ _ r e~*' dt L 2 « 1 2e~*2/2 7T X /2^7.= \ / l + 2sinh 2 t.=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'. (7. C(t) = .99) .2)(p_7?2)(7. Dwight [81].=0 V^ C'(t) I Jt=0 V^ (7.. we obtain dT _ 1 ~dJ~l 'See for instance H..97) sinh 2 1 = . p.
Explain the parameters entering the calculation.102) 'As a matter of interest we add that in the theory of a free scalar field x. 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.g. U) (7. See e. 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. Fig. B. . 238.4. It follows t h a t T~lnZ. 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 7.100) Reintroducing the dimensional parameters we had set equal to 1. p. 7.7. T h e following Example 7. also motivated by this paper. (7. this is T ~ ]n(ly/mou>/h).101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". 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. Zwiebach [294]. The equation there describes the classical "rolling down" of this tachyon.
Ap= v /(p 2 )<p)2. i. Quantum mechanically x(0). 2V ' Let A and Bio be the values of x and x at time t = 0.e. (x 2 ) h 2mou 2mou> and so l^J^ULl). the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). / 7 ^ 2 =2J(x }. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. so that 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 . A p are defined by the mean square deviations given by (see Eq. v A x A p = \ (x2)(p2) v = m0u(x2). Thus quantum mechanically x(0). so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.102)) p(0) = ^/(p 2 >. 2 x(0) ^ where A i . the smallest macroscopic length. 2 10 x(0) = B u . Eq.128 with x2 = LO2X2 + const.5)) Ax = <J(x2){x)2. ^(x2)(p2). CHAPTER 7. i. x(0) = A.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. (7. and for simplicity we take (cf. (p2):=mW(x2). and solution x = A cosh Lot + B sinh cot. We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I.p(0) cannot be determined with arbitrary precision. For h —> 0 the time T —• oo in agreement with our classical expectation. so that e. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T.g. . (10. (7. but instead are subject to an uncertainty relation of the form AxAp > h. Green's Functions x = Aco sinh cot + Bco cosh cot. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*).e.
but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4. For mathematical purists the question of convergence of the series is an even bigger challenge. 129 .(i)+/^(2) + ". An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. In general perturbation series do not converge.E^ is already a bigg problem.2h2 cos 2 a # = 0. Frequently even the calculation of the next to leading contributions ip^.Chapter 8 TimeIndependent Perturbation Theory 8. in closed form) only for very few potentials. one would set tt E = = ^(°)+/ty. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. 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).e.e.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. i.
2) n—»oo n It is interesting to compare the behaviour of the series (8. 0.130 CHAPTER 8. TimeIndependent Perturbation Theory In the case of this equation. It will be seen later (e. p.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . the expansions in descending powers of h2.g.= 0 < 1. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. . Schafke [193]. concerning also the uniform convergence of the exponential series. W. ^For further details. 8. T. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified.3) *See e.g.. 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. In the following we explain the difference between convergent and asymptotic series..g.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. h hl are in the strict mathematical sense divergent. E. Meixner and F . see e.+ . 581.. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. Whittaker and G. J. (8.2 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. at the end of Chapter 26) — and by comparing the results of WKB. which is known as the Mathieu equation. e. (8.2 However. This convergence can be shown with D'Alembert's ratio test since''' lim .. Watson [283]. .e. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . E = E^ + h2E^+hAE^ 71 72 7 + . the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence.g.* i. N.
E.. .33333 . i.4938. The theory of asymptotic expansions claims that if the expansion is truncated at the least term. .000 002 00 .. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. + 0. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. the series (8.3) diverges for every value of x. and in fact increase indefinitely.0.222.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.. Whittaker and G. N.4) for every value of x. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .2. + 0.000 000 06 + • • • . .22222 . Thus e.l n ( l + ^ ) n=l L ^ ' (8. . . for larger values of x.. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . after reaching 6/27 = 0.376 .e.. + •••...001 + 0. +0. . 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.. We observe in the first place that since lim n—>oo co (8. e.g. p.29629.8. Comparing the asymptotic series (8. / ( I ) = 1 — 1! + 2! — 3! H • However. and we see that the moduli of successive terms first decrease and then. 235.22222 . This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20..3) with the convergent series (8. /(1000) = 1 .0.0..9876. However.0. begin to increase again.g. We can see this as follows.0.3) 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. in spite of this the series (8.. It is evident that the larger the value of \x\. for x = 3. T..5) Cf. Watson [283].1). the better the approximation obtained.
T. It is interesting to observe the comments of various authors on this point. it is sensible to make the assumption. 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. 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. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. 236. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent.132 CHAPTER 8. Whittaker and G. In fact. **A. whereas applications of the convergent series (8.41) and (16. ip) can be represented by rapidly converging power series in e. E. Watson [283]. In many respects Messiah aims at more rigour in his arguments. This is not appreciated by mathematical purists. § However. in fact. 152. I. that E and \ijj) (i. 11 L. Merzbacher [194]. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small.e. Vol.1. TimeIndependent Perturbation Theory where 7 = 0." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. This latter observation hints already at the importance of asymptotic series in applications. . 16. Messiah [195]. meaning convergent. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z.1 between Eqs. the dominant approximation is." These "rapidly converging power series" in physical contexts are most likely very rare cases.5). p. although this has not been investigated except for a few simple problem^.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function.5) are practically unknown. p. (16. ^E.J In2. 371. Schiff [243]. Sec. N. the leading term of an asymptotic expansion. p. 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. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems).
= / e~t2dt." Of these three authors t h e first.. we can rewrite Eq. Vol.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T .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. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . II. p. B. Dingle [70]. He says. (8. It is therefore suggestive to write iy poo <t>(x) = 1 . Replacing t h e exponential in Eq. 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. . which have been studied in great detail.. (8. 8.10) Vn Changing t h e variable of integration to u = t2x2.2. solutions of second order differential equations. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. Messiah [195].8. *R.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. Merzbacher. there is no need for such bias.. 2 (8. 198 (German edition).7) by its power series and integrating term by term one obtains t h e absolutely convergent series Considering for t h e time being only real and positive values of x. (8.* and we follow some of t h e considerations given there.11) A . We know from books on Special Functions t h a t all of these functions.7) This function has been considered in detail in t h e book of Dingle. (8. ft Jx du = 2tdt.7) as / e'* dte* dt = 1 . In fact. (8.^=e~x2 / e^^dt.9) JO Jx J V 71" Jx The integral is expected to b e dominated by t h e behaviour of the exponential at t h e lower limit.
e.i)[(_!)«' Then.15) < 1 we can insert this expansion into Eq.^ X e U ~ ( 1 + * \ 2 .14) sin7T2. (8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.134 we obtain CHAPTER 8. 7T (ni)!sin{7r(n+^)} ( „ . we obtain 1+ i.13) presupposes t h a t < 1. (8. 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion . Using the reflection formula (*!)!(*)! = for z = n + i . du.e. since (—4)! = y/n. Then e (n —\)\( 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.' we have 2 nH 7T 7T (8.^e~x \Ar i.16) . « \~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). TimeIndependent Perturbation Theory <P(x) = 1 .
12) ignoring the fact that the latter is valid (i. .17) 0(X) = i. and in fact so slowly.19) implies either that we ignore the remainder or the correction term given by the integral in Eq.—yS!L^LLZ e x OO / _ 1\ X2 /CO dueu L \ / V n = Q 1 and we write 0(X)~1 V 7 n=0 v 2^. (8. whereas in a convergent expansion the terms first increase in magnitude.19) originates. (8. ' arg. As in the case of the gamma function. (8.e.8.r<7r. Frequently "=" is written instead of "~".e. It is fairly clear that the above considerations remain unaffected for argx < 7T.18) or that we insert the binomial expansion (8. In an asymptotic expansion the first few terms successively decrease in magnitude.l)! = / Jo Then 2 cHnxdt.e.19) approximates <f){x) the better the larger x is. convergent) only in the restricted domain u < x2. We now see how the asymptotic expansion (8. the asymptotic expansion of the error function is. much more useful than the convergent expansion. argx = 0.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. /•oo T(n) = (n . i. The expansion (8.15) into (8. Whichever way we look at the result. In the foregoing discussion of the error function we assumed that x was real and positive. We can actually understand the deeper reason for this in practice. in general. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. i.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. (8. 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. We now want to relax this condition and allow for phases argx ^ 0.
e.2 y V . dv = 2sds = . thus with \v/y2\ < 1. The situation becomes critical. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.22) we obtain 2 4>{iy) = . VK Jo V^ Jo (8. (8.7) and obtain 4>(iy) = —= j e~l dt = = / es2ds. when argx = 7r/2.vds.15). V7T Jo Changing the variable of integration to v (8.20) where we set t = is.21) = y2s2. we set x = iy in Eq. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . however.^ = .24) We can read off the binomial expansion of the factor in the integrand from Eq. (8. i. 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. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds.23) Throughout the range of integration 0 < v < y2 the integrand is real. (8. =dv.^ f V e ^ . (8. (8. We therefore consider this case in detail.26) or else return . we can try to proceed with Eq.136 CHAPTER 8.
However. These integral contributions are of order 1/y since the integral behaves like e~y and is multiplied by a factor e+y in front. (8. Eq. and so the integral is real. (8. .. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 .e. e ^ ( i y ) ^ ^ .24).30) yvn f Jo V y J Cf. (8. If we proceed with Eq. (8. the better the approximation expressed by Eq. (8. and the larger y.27). for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions. y y 2 ( n _i)! y arg(* = iy) = i * . — (8.2 yir (8. We have therefore <j>(iy) = ^e^TL fV e~v ( l . Returning to Eqs.^] * dv.29) where 5t means "real parf.8.28) because then the correction term oc f°£ becomes smaller and smaller.27) Then .19).28) ignores the integral contribution of Eq. Suppose now we consider Eq. i. (8. i (8. We know that for v < y2 the integrand is real.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! .^ ^ p .28) n=0 We observe that this expansion differs significantly from (8. for large values of y. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! .27) and (8. which will be studied in more detail below. i.e.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv.24).2 Comparison of Asymptotic Series with Convergent Series to Eq.28) we observe that the expansion (8. (8. (8. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.17): J£° e~vvndv = n\. ey is much larger than something of order 1/y.
. (8. (8.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.138 CHAPTER 8. B. *Cf. (8. i. (8. R.26).33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. We have therefore found that XTT n=0 ^ ' and . and • higher order terms of the associated series all have the same sign.2 OO rz=0 V .28).33)) would have • a maximally increasing exponential e+x . (8.e.31). The corresponding phase lines are called Stokes rays. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. With a similar type of reasoning one can show that the same expansion. since (8. e xZ ^ (n — £)! „ ^ ) .* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself." — E ^ ^ forarg. 7r = . Since (cf.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. y.31) and (8. Chapter 1. Dingle [70].25) and integrate from 0 to oo. Looking at Eqs.3!) which is (8. is obtained for <f>(x) with arg x = —ir/2. (8.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. TimeIndependent Perturbation Theory If we insert here the expansion (8.e.. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. . Eq. This result is identical with that obtained previously.
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. 91. 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. (6.2 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. See also Tables of Special Functions. 5. Dingle [70].34) the contributions 1.e. W. R. though. Equation (8. i. Thus if we set x = XR + ixj.32) and (8.2).g. . Eq. 9 . pp. Oberhettinger [181].37) Cf. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0.35) 8.e.13. (8. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . apply predominantly to asymptotic series of the solutions of second order differential equations. and so may not be universally applicable. —1 cancel out. p.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. in the sum of the right hand sides of Eqs. 1/1 ipv{x) v1 vl (±ix). for arg x = 7r/2. i. B. Magnus and F. Chapter 1. '"Cf.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. e. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x).8.2. These rules.
ei<v+VxvHv{x\ argx = £ . 8. . B. 0 < argz <  (8. Eq.37). R. 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. x~l/~1ipu(x). 1 1 Cf. Hence Dil){x) = xv(j>v(x).e. i. Dingle [70]. v — — f — 1. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix.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.7. We consider Dv (x). R.38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. Thus D{J\x) (i.40) § Cf. Chapter 1.36) is invariant under these replacements. ti asymptotic series (8. Chapter 17. We also observe that the late (large i) terms of the series in (8. TimeIndependent Perturbation Theory i. as can most easily be seen in the case of the Mathieu potential.39) (as in the example of the error function).e.x 2 oo «*)  — E^fi=0 v . with a real proportionality factor.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. 9) and XXI. The continuation of (8. i. ' S u c h symmetries are extensively exploited in the perturbation method of Sec. B. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. Dingle [70]. = 7r/2.e. (14). Chapters I (in particular p. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a.140 where^ CHAPTER 8. (8. (8. > > The equation (8.
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)] (8.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). On reaching ir the part containing ^ ( x ) . as yet an unknown real constant. where in the first line on the right hand side the second term contains the real proportionality constant a/2. for xj = 0) and therefore possesses a Stokes discontinuity there. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i.e. < 7T. The value of Du (x) on the other side of the Stokes ray. (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv).40).44) + ^iap) = 0 . 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. Since Dv (x) is real when x is real. i. we must have at arg x = 7 in the dominant factor on the right of Eq.e. = ir. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. In (8.8. < 37r/2.41) the Stokes multiplier a is.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. = ir.42) T ^(eiwu i. (8.e. with \x\ = (—x) for arg a. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. beyond arg x — 7r/2. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. We therefore proceed to continue (8. (8.41) to arg a. another half of the discontinuity appears on leaving the Stokes ray on the other side.e.41) xuct)v{x) + a{x)~u~lipu{x). (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). i.
or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. We can see that the equation is satisfied by (3{u) = ±a(u .1) = ±a(y).48) We compare this with the reflection formula (8. Hence the left hand side must have the same property. (8. (z)\(ziy. V i=0 \U0 v 2w . . Eq.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.51) .1)! TT.47) Equation (8.14).42)) for argx = ir (in the second line of Eq. i.1) = 2sin7ri/.45) is therefore given by a{v)a(v . (8.46) For a given function a(y) this equation determines /3(u).e.e. (8.42) that a multiplies ipv(x). TimeIndependent Perturbation Theory Now a and (3 can still be functions of v.v . i. a{y)P{y) = a(u .1) (8. (8.e. (8. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .1). (3{v .48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . (8.1) (8. Dl \X) is (cf.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. argx = vr.142 CHAPTER 8.l)0(v . i.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .—^ ON • — ^ > + ^ ^ L .e. i.42) we insert from Eq. 8.45) that af3 = 2sin(7r^)) D^\x) = (8. (8.
I / . (8. For details we refer again to the book of Dingle. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.. (8.35)..52) (argx = 7r). Chapter I.*"ei°> f ( ^ _  _ L _ = (!)««(.e.. ' ^ (i/l)!z! ^ .. i.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.50a) because this enables us to define normalizable parabolic cylinder functions.7r). We observe that for v — n = 0.36).7). Eq. the second contribution vanishes and we obtain £>«(*) . ' (X) (cf. (8. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .1.e^ .. (8.e. i. It should be noted that in Eq.1 ) ! 4~L i\{2x2Y i=0 V2^(x).8.)(. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator.A / ± sin^Cx)"M* 2 V ^ .1 ) ( . i. which vanish at x = ±oo.) (8. 8.2. u 1 y> {2i + v)\ 1 n 0.3 Asymptotic Series from Differential Equations In the case of the error function we obtained the asymptotic expansion from the integral representation of the function. —f..36)) in the domain 0 < arg x < IT. solutions of Eq. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di. the error .3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. (2x2)1 = COSTTU(X) v . (8.„ _ix2^(2ivl)l e 2 > —. However.e.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.
Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on. ^ + 2 .38) can be obtained.55) The large x asymptotic expansion can also be obtained from the differential equation.37). Solving these equations we obtain = = 0. Eq.. • • • can be determined. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. i. (8. Instead of dealing with Eq (8.e.59) and equating to zero the coefficients of powers of x2. 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.36).60) into (8.36) in the form (coefficient of — x2 here chosen to be ~. x—>±oo (8. i. how series of the type (8. + — + ^ + • • • . S\.144 CHAPTER 8. % = {.. Eq.^ + 1 ) . x°. \ + U (8.54) we consider again the important equation of the harmonic oscillator. (8. (8. S2. cf.. and demonstrate. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . (8.54) (8.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. previously we had ^.60) 2 x x l Inserting (8. 5„=4 Sl=o.36)) 0 where + x2Q(x)y = 0. We write Eq.e. 0. (8. x.
37). 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. (8. » 1. 2 X 2Wi.35) and (8. and it is clear t h a t higher order terms follow accordingly.38). Magnus and F.61) (the dominant terms of (8. do not enter here into their derivation and simply quote the result: W i t h \x\ . T h e expansions (8.91. (8. function. etc. i. .3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8.63) **W.+ X M W = Q. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . 4~l~2 ' i i _i. / i N .37) with (8.8. T h e parabolic cylinder function normally written Dv(x) is frequently defined via the Whittaker function WK^{x) which is a solution of Whittaker's equation d2W dx2 Thus* Du(x) = 24 + 1 K 1 + T+ .19). In the following we are mostly concerned with asymptotic series in some parameter.z) is a confluent hypergeometric T.b. p. We. therefore. n 22e"T JziL (8. Oberhettinger [181]. The derivation of the asymptotic series from integral representations (cf. are asymptotic series in a variable x. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). Nonetheless we shall need asymptotic expansions like (8.e.62) 4 \ 2 XiFi 2'2 .36) have correspondingly e±x / 2 ) .38) in a variable for the solutions of approximated differential equations.
0. whereas e~x '4 increases exponentially. E. .4x2 (8.1 a2 an a0 + — + ^ + . Whittaker and G.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially. N. 349. 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.+ — + ••• 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. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . Watson [283]. 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. 8.146 CHAPTER 8.64) (c) for —7r/4 > a r g £ > —57r/4: Dv{x) e 4x !*£=£2 2x i/i + +• (I/1)! + xe 4 x (i/+ !)(„ + 2 ) 2x 2 " ' (8. 4 (8. 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. Thus there is no contradiction^ between (a) and (b).65) We observe t h a t the series of (a) and validity 1 1 7r < a r g x < 4 In this domain 'Qx > 9ffcr and so e (b) have the common domain of 3 3 7r.67) Cf. T.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.
g.Sn{x))  = oo n—>oo (8. in particular with regard to possible exponentially small contributions to the expansion for which e. In fact. (8. 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.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) .68) is (effectively) simply a statement of the observations made at the beginning. Chapter I. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV.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). thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. A critical discussion of the definition can be found in the book by Dingle [70].73) Vn + q/#> + e2T/i2) + • • • .8. i. possesses a whole chapter on asymptotic expansions.69) and the consequent nonunique definition of the expansion. first published in 1902. lim xneW x—>oo = 0 (8. The definition is due to Poincare (1866). we set E^En * > ^ n = En°1+eEnV = + e2EnV + .e. .68) The definition(8.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. (8. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". which is assumed to be small. (8. 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.
78) into Eq. so that we choose / • dxrpM^O. We insert the ansatz (8. The second equation can be rewritten as {Ho .As a consequence the state vl/) is a vector in this space. TimeIndependent Perturbation Theory We insert these expansions into Eq.e. °° i J or 0 = EWJdx{rnVil>n). ( ^ . J f°° dxrn E(^ 0) . (8.e. .i4 0) )V4 1} = {EM .148 CHAPTER 8. we write in the position space representation ^ ^ E ^  (878) A contribution to ipn can be combined with the unperturbed part of the wave function \P. (8.n = 40)V>n. (879) We multiply this equation from the left by ip^ and integrate over x. as are also the states Wn )j Wn )i — We therefore write these vectors as superpositions of the basis vectors provided by the unperturbed problem.4 0 ) M = / ^C(4 1} .75) HoW + vW = E^+E^+E^. i. i.e. (8.n. (8.70) and equate on both sides the coefficients of the same powers of e.74) (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K. Ho^+V^n = EW^+E^n. i.w » .V)i.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . ^ ) = 0.76) The first equation is that of the unperturbed problem. (8. With the help of the orthonormality condition (8. We obtain then the following set of equations: Hrj.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.
and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . (8.Viprt e (8. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n. the coefficients a\ .5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. (8. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates.83) makes sense only if Ef] ^ EnV for all i + n. multplying Eq. i. i.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n. (883) Thus the procedure forbids the equality of any E\ '.e. (8. ^ n ) + 0(e 2 ).8. We therefore consider Eq.Vil>n) (880) In order to obtain ipn .i ^ 0.m ^ n and integrate: oo /•oo / oo dxrmY. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution.84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. (8.79). we return to Eq.^(EW ~ V ) ^ + E& / J—oo dx^miln . (il)i.82) *> = ^ + E .e. S ' ti^+Q^)• We observe that Eq.V)^ Setting + 42tyn. with En .4 0 ) M 2 ) = ( 4 1 } .76) which we rewrite as (H0 .84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above). We multiply the equation by ipm. (8.
En0) (8.89) • • • where we guessed the term of 0(e 3 ). (8.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .^w .^ ) + $& w £ (^fc.Vlpn] M _ P (0K 2 ^r .88) into xjjT we obtain similarly * r i + f ^ i^n (lpi. Inserting the expression (8.150 CHAPTER 8. (8.2 (V> m .(1) (i&0) .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .87) (i>k. (8. FVn) + e 2 ^ ( V > n .V^n) .86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq.Vlpn)E. V ^ i ) X (Q) (Vi.81) we obtain (2) _ _ {lt>m. n) £f) (8.VV . ~Ej E E.V^j)(^j.85) together with (8.B i 0 ) ) ( £ f .e. j.4 0) ^+ e 2 E (lpk. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„. ipn ) = 0 i. ^ n ) j&iift + ^n .88) (Q) Inserting Eq.V4>n){lpn.^(8.79) •/—oo oo / Using (8.Vlpn) E^ .£i°») 2 Hence +^£o (£«? .^j)(V .V'>pTl (E^Eny V'fc (8.Vlpn)(lpn.
.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0). We then obtain (lpk.e.90) is the term with only one summation J2 m the contribution of 0(e 2 ). i. One would prefer to lump it together with the other contribution.n(x) = J2crtn+ii (8. We can therefore rewrite tpn ': The perturbation theory described in Sec. Merzbacher[194]).5 RayleighSchrodinger Perturbation Theory 151 The expressions (8. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e.8. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i.90) can be found in just about any book on quantum mechanics.g.7 is based on this procedure. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. An ugly feature of the expansion (8. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. (8. The functions ipn(x) are eigenfunctions of some secondorder differential equation.Vlpn) = y^Ci(lPk. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution..lpn+i) = ^ Cj5k.89)..91) The coefficients Cj follow from the recurrence relations. 8.
. i.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. (8.152 CHAPTER 8.97a) and (8.94a). Equations (8.70). —I.96) are effectively the known equations of ipnitpm and so yield no new information. all n. Equations (8. For example the magnetic states enumerated by the magnetic quantum number m. Em = 4 ° ) +eE$ + e2E$ + .97a) (8.95a) (8.V)(c2l^n + ci 2 ^ TO ). Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.97b) with ipn y£ ipm. .e.94a) into Eq. (H0 + eV)^n and (HQ + eV)Vm = Emym. (8.96) (Ho .93b). In order to deal with the problem of degeneracy we consider the simplest case. .93a). (8. .Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . m = I. (8. so that En = 4 ° ) + eE™ + e2EP + ••• .4 0 ) ) ^ } = (Em] .95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. + c22^m).We now insert Eqs.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . (8. . . (8. I — 1 . Em can now be linear combinations of ipni'ipm. (8. i.ipn).93b) The lowest order eigenfunctions belonging to En.94b) with (ipn. TimeIndependent Perturbation Theory 8. in a central force problem all have the same energy unless the system is placed in a magnetic field. (8.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '.e. is not uncommon. degeneracy..e. that of double degeneracy En = Em .ipm) = 0 = (^m.97b) are . i. however.93a) (8. (8. are allowed to be equal since otherwise the expansion becomes undefined. Equality of eigenvalues.
V^n). Wm. We now multiply these equations from the left by ip* and integrate. Vi/>n)] . (8. ^ ) ) = CU[EW . H0lP$) = C22[E$ ~ (Vm. ^ o ^ ) ) = 0.£ < « = ag)^) . Vlf>m)] .(lPn.EJ®(lPn./>£>) = C2l[E£> .(Vn.98b) + Culpm) + cuV^m).y>m. V^n)\ .8. Similarly Wn.m). £ aW^°Vi) .H0^) .6 Degenerate Perturbation Theory 153 inhomogeneous equations.ci 2 (V„.H0xl>$) .^ ( ^ .99) and ( ^ ^ O ^ ) " ^ ^ .99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .ipm ) the equations (iPn.V^n).J#»a&> = 0j since ii4i = En (degeneracy). Vl/. we obtain with (ipn. H0lpV) .C22(Vn.cnVipn and tyi. ^ ) = ( V .m.(^04°))41)) = EM^CuTpn (ipl.ipn n) = 0 = (ipm.H0il>W) = cn[EW . But (8. (^m. t f ) = a k 1 } 4 0 ) " Eg>aU = 0.C 21 (^ m .^ ) ( ^ m . (8. (^n. . c2iV^n + c22V7pm). Then (^.Vil>m)] V^m). Equations (8. cn(ipm.98a) Setting I = n. and ( ^ m .
1.5).e. A onedimensional system. Example 8. ip = c<p. In this particular case therefore nondegenerate perturbation theory can be used (cf. 11.^ ' V = — (•>!>'<t><t>'i>).. splits the problem into two independent onedimensional systems.100a) determine the coefficients Cn.E)tp = 0.ci2. d>" E=^. if)' (/>' — = —. i. i.154 and V (^m. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . the secular determinant. in particular in parabolic coordinates (see Sec. 9} ip. It is then possible that a special perturbation.100b) determine C2i. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. like that providing the Stark effect (proportional to rcosO). dx i. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. i> 4> Hence there is no degeneracy.E™ (Vn.6. TimeIndependent Perturbation Theory (l/>m. Example 11. Then / > i>" 2— = V ip Hence 0 = V ' V .^Vm) 0. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. as we saw.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. However.Vrl>n) .e. \nip = ln(j> + c o n s t .C22. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. (V>V</>V>) = const.3).Vlpn) CHAPTER 8. If one wants to perform a perturbation calculation in such a case. (8. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). It follows that ip and <j> are linearly dependent.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy. the degenerate perturbation theory must be used. .101) This secular equation determines the first order correction En • The equations (8.Similarly Eqs. . At x = ±00 the constant is zero. 4. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8. (8.e.
106) (8. Applications of the method are. and the solutions then have to be matched in their regions of overlap.7 DingleMuller Perturbation Method 155 8. J. (8. and such that one can even obtain a recurrence relation for the coefficients of the expansion. (8. with some restrictions on the function V. cf. of course. Muller [216]. J.102) where D is a second order differential operator and E a parameter.103) such that if E is written correspondingly E = E0 + e.104) are chosen such that the equation Dcf) + (E0 . Thus the method is a systematic method of matched asymptotic expansions. i. B. Eq.103). H. W.t Since in general a solution is valid only in a limited region of the variable. V = V0 + v. We consider an equation of the form Dcj) + (E . The subdivisions (8.8. J. Dingle and H. the considerations given below would be valid even if D contained only the derivative of the first order.V0)<f> = 0 (8. T For instance spheroidal functions can be studied with this method. See also R. W. W. a potential. also possible in areas not of immediate relevance here.e. the method has to be applied in each domain separately.V)(p = 0.e)<t>. in particular in applications to periodic potentials. We assume that V can be expressed as the sum of two terms. It is immaterial here whether D does or does not contain a first order derivative — in fact. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term. Dingle [72] and H.V0)<t> = (v. (8. (8.* In various versions this method is used throughout this book. . anharmonic oscillator potentials and screened Coulomb potentials. Muller [73]. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.105) (8. Muller [210].102) can be written D<f> + (EQ . B.104) "This method was developed in R.
TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.110) If we write the next contribution <f>W to <p(°> in the form m+st (8.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u . (8.E0\ (the latter in a restricted domain). s an integer: (ve)(pm = 'Y^{fn. (8.Em+s)(/)m+s. Then if ]e < \E\ and \v . m + s)0m+s.108) with coefficients (m. and so v (8. since D<f>m + (Em . m + s) which are determined by the recursion formulae.111) s then (see below) 0(1 = E (m. 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.) This result is obtained as follows.107) represents a first approximation to the solution (f> of Eq.V0)<f>m+a = 0.Vo)(j)m+s = (Em .m + s)4>m+s s (8. v = {D+EmVo)<t>m.m + .V0)4>m = 0.e\ < \V0 .156 CHAPTER 8. where m is an additional integral or nearintegral parameter (depending on boundary conditions). the equation Dcf>m+S + (Em+S .109) D(j)m+s + (Em . Considering (frm+s we have.V0)<f>^ .m + s)" in order to relate them to "steps" from m to m + s. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .+s ' (8.113) .102) with E ~ Em.e ) ^ = 2 ( m . We write the coefficients "(m. the function 4>(0) = <t>m (8.
m) + > ^ ..m)4>m in (8. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8. . .m + s)(m + s.m + s)(m + s.m + r)(rn + r. (8.117) „ . V ^ V ^ {rn. the coefficient of the term (m.Q (m.Em+S) = 0 (second approximation). (8.m) E J^ ( m ~ Em+S) ' • fa.119) is effectively the secular equation).m + s)(m + s. (8.m) — —. Thus (m. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. (8.„N This procedure can be repeated indefinitely. .e.n\ ^119) + .m + s)(m + s.TP \(T? . .110) and are therefore related to the inverse of a .m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m. Eq.116) m ^ v (m.m) 1^1^ IW .8. (8.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. (8. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.m) — 0 (first approximation). We then find altogether ^  l^m  Em+S) + y^y^ SJ. We observe that the coeffcients of expansions (8.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .113) must vanish.^ (Em .102) to order (1) of the perturbation (v — e).118).119) follow a definite pattern and can therefore be constructed in a systematic way.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.T? S The first expansion determines the solution 4> (apart from an overall normalization constant). /01_. (8.7 DingleMiiller Perturbation Method and so (m. .
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. Miiller [211]. The following Example 8. The result is the following expansion. for instance. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. § H. (f>m in the above. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. (8. .[ 4 ( 8 5 g 4 + 2q2 .e. TimeIndependent Perturbation Theory Green's function.158 CHAPTER 8. This can in fact be done and will be dealt with later. Other applications can be found in the literature. for the investigation of the behaviour of the late terms of the perturbation series. i. this is — so to speak — the price paid for obtaining a clear systematics which is essential. and is typical of a large number of cases. has been treated by R. J. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders.38). i. W.120) n4 .* Example 8. ^The potential A r 2 / ( 1 + gr2).6q + 1)1 + 24(5? .I2q + 64) + 256Z 3 (41 g . S. The eigenvalue expansion obtained is an asymptotic expansion. 2 .9) + 4096J4] + 0 ( l / g 3 ) .[q(llq2 + 1) + 2(33g 2 . for instance. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. as will be seen in later chapters.e. Kaushal [146]. . in which q = in + 3. . n = 0 . does not occur in any of the later contributions. since it will be used extensively in later chapters. and when the solution is multiplied by this factor one obtains the contributions involving </>m in the higher order terms (for an explicit demonstration of normalization see Example 17. (8.423) + Z(2720 9 3 . 1 . .7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 .2). This means the solution is not normalized.1)( 2 + 64J3] — . An explicit application of the method (not including matching) seems suitable at this point. as already mentioned after Eq. as will be seen later.
i/> with Dq = d2 . (8. <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.b. . 2 . 1 . a)ip(a) + (a. .a + l)if>(a + 1) + (a. multiply the equation by onehalf and set h = —a/g. 2 The solution XQ(S) = $ ( a . (fc2 + <?2) 4ga and b = I \—.2a2 A.V ! 4 I z S *! V2 ipq(z). .123) the contribution fl<°> Ah + .6 and 16. 3\ 2 V ' 2.2) where 1/ . Sees. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a.122) where A is of order 1/g. „2 J„2JL. Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.. 1 1(1 + 1) 1 .123) The first approximation (note the last expression as convenient notation) ^ . The solution i. We now insert this equation into Eq.2 '(l + l) .0(z) = zl+1ez2^fa. . \2 . ._ „!2 ^ ( " « 2 r r 2 > .l)ip(a . 11. Setting q = in + 3 implies k2 + g2 ga(2l + q). (8.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . S) is a confluent hypergeometric function. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. a . (8. b.1). Therefore in the general case we may set k2 + g2 = ga(2l + q) . 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar.b. (. .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a.8. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ .z2 is a normalizable function if a = —n for n = 0 .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq.121).
one could now use the RayleighSchrodinger method and rederive the result (8.j^0 ' + jU+1 J ^(a + j) with h[a. a + r) = 0 for r > m. a + 1) = a = In general CHAPTER 8. „ .120). (a. 0 = h[a. a ] i + 0(/i3) [a. a]i + h fa. a + r — l)(a + r — l . Hence the next order contribution to tp^ becomes oo i+2 r [a a . The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T (i+2) [a. a ] x + L _ 2 J 1 [ « . a + j)i/>(a + m).b = (g + 3) . i. Now proceeding as explained in the text one obtains finally the expansion ip = ^(°) + v ( 1 ) +y>(2) H — along with the expansion from which A is determined.a + j}i+1^(a + j).a + 2]i P .1) = a .120). If we now evaluate the various coefficients in the last expansion we obtain the result (8. . a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a .a]1=0.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". In fact. o + r) + S m _ i ( a .l . fa. a + r) + S m _ i ( a .a]2+L ^ J1[a + 2 . Now we make the following important observation that Dqip(a + j) = jip{a + j). a + r) satisfy the following recurrence relation: Smi(a. a + i) = 0 for i ^ 0.2a = I + q. As a check on the usefulness of the perturbation method employed here. of course. a . a + r) with the boundary conditions Sn(a. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context.160 where (a.e. a ] i + L ' ^ J 1 [ a + l . (8. (a. a + i)(a + r.a]i = . So (a. a) = b .a—lli.Z. This equation shows that a term fiip(a + j) on the right hand side of Eq. _ J [a .a — 211 . where [a.^ A + J7^S2(a'a)' K a +J]i+l = 4 / i + 2)!' S ''+ 2 ( a ' a + j) for i and j not zero simultaneously.S m (a.i ^(i)=£V+i i=0 J2 j = (i+2). ( 5 m ( a .3). a + r + l)(a + r + l . a + 111 r [a. the latter determining to that order the eigenvalue. when j = 0. a) = 1. .123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. fa. TimeIndependent Perturbation Theory (g .2. The coefficients S m ( a .
proceeds along similar lines. 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 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.Chapter 9 The Density Matrix and Polarization Phenomena 9. i.e. although not exactly straightforward. 161 . 9. Regrettably this has to be left out here in view of lack of space. light. as we shall see in Chapter 27. and this means electrodynamics. the generalization to many particles. We therefore begin with the appropriate classical considerations with a view to generalization. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets.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. and we shall then see that these lead to analogous results as in Chapter 2.1 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.
t) given by A(r. In general the constant Ao is complex. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized. This is the equation of an ellipse as indicated in Fig. A 0 = A0R + iA0I. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.3) (9.162 CHAPTER 9. The Density Matrix and Polarization Phenomena potential A(r.1 The vector potential ellipse.i). 9.e. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0.4) (9. (9. (9. It follows therefore that A(r.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized.t) = 3ftA.6) . t) = A 0J R cos(cji) + A 0 / sin(u.1) Fig. (9.2) (9.1. i. 9.
the connection must be linear. We set therefore A 0 = Ai + A 2 . i. the fields E and B are observables. we recall.A« we can construct only the following scalar products with this invariance: (A(°\A(°)). for instance. which leads to an outgoing wave which we give the label "1". (9. (AW. the vector potential A.8). Observable quantities can only be those which are invariant under translations and rotations. the direction of polarization. not. a calcite crystal or a film of nitrocellulose.e.7) We consider now measurements with polarized light.e. p.2 *i*4 0) (98) with coefficients F^. (9. represented by a filter "F". i. . / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2.A(°)). as e. The intensity 7 of a light wave with amplitude coefficients A\' . however. AW = £ fc=l. From the vectors A(°). Then their connection is given by Eq. with k .A i = 0 = k .9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.t in which k and ui remain unchanged. (A(°). See for instance A. ie.g. In classical electrodynamics. the intensity of a wave. Here we are interested in observable properties of light waves — like. I=<l2x2>:=E40)*l*40)i.A2 is the sum of the moduli.A 2 .A«).e. (AW.AW). i.2) the vector potential A is completely determined by its two (say (x. (9. Rae [234].11) 'This means we consider the polarization. A2. 17.k (910) We give the initial wave the label "0" and subject this to an experiment.y)) components A i . 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. (9.9. determined with the help of a polarizer.
A*FikAk. pki = AkA*.17) .k (9.13). (9.k l f c 4 ° \ (9.A( 1 )) = ^ ^ F i.e.13) the quantity F must be an hermitian matrix and thus representative of an observable.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. (9.k i (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments. if (F) is an observable quantity.12) i. (9.15) in order to rewrite Eq. i. (9. (9.16) i.e. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. We write the observable quantity as (F):=Y. so that (F) can be looked at as the expectation value of an observable F.164 CHAPTER 9. Thus (F) = J2FikPki = ^(Fp)a. We use definition (9. the connection between the initial polarization and the final polarization. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. (9.A*kF£A* = E and hence F = Fl A p i tkAk. i. Thus. i. (A(°). To this end we observe that as a consequence of Eq. (F)=Tr(Fp).15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). Then according to Eq.13): j2^FikAk = Y^AiF*kAi = Y.e. it must be real and hence (F) = <F>*.
21b) This state is still a pure state. is given by a single polarized wave. A2 = b as </*) . <»^> (9. that of a pure state. . it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions.19). 165 (9. b = 1 we obtain the expressions (9. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip.e. (9. we have (9 19) ' In this case there is no preferential direction. The density matrix for the pure state (9. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .15) with A\ = a. is in general unpolarized. i. i.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). In the present case of a polarized beam of light we have two possible. "incoherently. i.d "»=(o ! ) " In the case of an admixture with equal portions of 50%. This admixture is described by introducing matrices p.18) The vector A describes a particular ray of light.(AAt) = ( £ £ ) .e.( J o ) . (9. For a 45° polarized wave we have 1 . like that of a bulb. The socalled pure case.21a) follows from Eq. But light. 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. b = 0 and o = 0.9.e.* For a lightwave travelling in the direction of z.22) For a — 1.
The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . (9.24) The following two 50% admixtures yield the same effect: 1 1 (\ 0\ P = 2P*+2Py=[Q 1 P = 2^ 4 5 ° + 2P135° i J.™ V 0 n kmAm / ^^jPim^mk' . as a consequence of absorption). l l\ " i °\ ) ' ^925^ 9. (9. 1 = . b = —=. Pi35° = \ \ 2 ? )• (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. Equation (9. (926) where a is a parameter whose variation describes the continuous variation of the polarization from its initial state labeled with "0" and hence parameter value OQ (analogous to t or j3 in our earlier considerations of the density matrix) to the final state with label "1".15) Pik — A i A k .Z^ 3.3 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. Then according to Eq.23) 1 . We set in analogy to Eq. (9. in which the intensity I = Tr(p) is not conserved (e.g. We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F.8) A(1)=E^40)> k R = R(<*).166 and CHAPTER 9.
9.29) where (i. In the case of Eq. (9.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.e.31). here this contains the dependence on a.t) in Chapter 2 in what we described as the Schrodinger picture there. The corresponding condition here is that no absorption of the wave takes place. i. and p(0' remains unchanged. in the state functional) in analogy to the time dependence of W(p. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).33) i.30) Ti{FRp{®I$) = Tr(F. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size.32) (9. In the case of Eq. (9.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. in other words without creation or annihilation of systems. the dependence on a is contained in F' = R^(a)FR{a). (9.34) .pW).e. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations.31) (9.q. (9. 9. the observable F remaining unchanged.28) (9. (9. that the initial intensity Jo is equal to the final state intensity I\. i. (9. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .e.e. the observable is transformed. on the other hand.4 The Liouville Equation 167 pW = RpMRl (9.
35) Thus the matrix R has to be unitary.37) (9.e.p(°\ao)]da.36) (9. (9. (9.B}:=i[A.e.2) has to be taken into account. Eq.46) suggests to define Poisson brackets of matrices by the correspondence {A. = Tr(R^Rp^).168 i.M . j£ = i[H.F}. In view of Eq. CHAPTER 9. (9. (9. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1. .21) and (2. the matrix M is antiHermitian: Mf = .40) can be considered as quasiequations of motion.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. We set therefore M^:=iH. (9.p]. Comparison of these equations with Eqs. i.40) Equations (9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H. (9. (9.41) We see therefore that the 2 x 2 matrices entering the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics in Chapter 2. (2. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.B].39) If the dependence on a is not contained in p^1' but in F' (cf.39) and (9. 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.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. we establish their Heisenberg equations of motion. 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). Thus in particular we establish the uncertainty relation for observables in general. (10. To insure that the 169 .1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real.Chapter 10 Quantum Theory: The General Formalism 10. the Schrodinger. and consider timedependent perturbation theory.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. 10.
and assume that A = A(p. Such a correspondence with classical mechanics is not always possible.q).Pj] = 0.3 represent the three Cartesian coordinate directions of x.2) (F(A)) = {u\u) <"lf. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator.* a conspicuous exception being for instance a spin system. E. we must have (u\u) = 1.Qj] = 0. we assume this now. classically qiPi = PiQi. Pi • Pi.2.B] = ih.2. We can describe a system as one with a classical analogue if the Hamilton operator of the quantum system is obtained from the Hamilton function of classical mechanics by the correspondence •^i • Qi. Naturally one can also write (10.y.4) *Even in those cases where this is possible the operator correspondence may not be unique.Pi are called fundamental observables. which clearly does not hold in the case of operators.170 CHAPTER 10. (10.g. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. [qi.1 U n c e r t a i n t y relation for observables A. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. 10. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi. (10. q). however. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. where i = 1. . Here. The phase space coordinates qi.Pj] = ihSlj. \pi.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p. y. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. such as for instance of the angular momentum operators Li.
12) where \u) 6 "K is a ketvector representing the state of the system.8) and. Applying to the expression (10.2A(A)}  {A)2)1'2 (10. (A{A))(B{B)){B{B))(A(A)) (10. Since A = A— (A). Thus AA = AA = ( i 2 ) V 2 . Eqs.2A(A) + (A)2.7b) = = [A.7a) (B) = 0.(A)2 = (AA)2.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.2(A)2 . (10. using Eq. (10.10. B}= ih.7b)..0)1/2 = AA. so that (A) = (A) .6) B:=B(B)..9a).10) .9) ((A2) + (A)2 .. (4. Ai = = = ((A2) .5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A).{A)2)1'2 {{A2) . Then [A.11) (A2) = (A2) . (10. AB defined by the mean square deviation AC=((C2)(C)2)1/2.B. (10.2 States and Observables then their uncertainties AA.. 111 (10. (3. (10.12) the Schwarz inequality (cf. (10.B\u) G IK as vectors) we obtain (for an application see Example 10. = \\A\u)\\2\\B\u)\\2 (10.(A)2)1'2 (1 °= a) ((A2 + (A)2 .5)) (for A\u). C = A.B] = ABBA > ^h. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). (10.13) AB = AB = (B2)1/2.(A) = 0. we have A2 = A2 .
and hence for condition (b) 0 = = tSee Sec. We can therefore rewrite Eq.18) >R2 + I2. (10.16) I = h. R= (AB + BA). (10. AAAB>I=h. (10. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq.17) as had to be shown. (10. since A and B are hermitian. . The condition (a) is satisfied when. The equaltosign applies in the case when (a) it applies in the Schwarz inequality.15) as (u\AB\u) = R + iI. and (b) when (AB + BA) = 0. and with Eq.13) (AA)2{AB)2 i.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. (10.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. 4. Zi (10. Zi Zi (10.3.e. (10. (AB + BA) This means (AB + BA) is real. Zi + BA)\ + ^ih.14) Hence (u\AB\u) = /^(AB where. From this we obtain (uiJBu) 2 = i22 + / 2 .172 CHAPTER 10. A\u)=cB\u). for a complex constant c. (10.4).
Lj] = (LiLj + LjLi) + ih£ijkLk.10. 9ftc = 0. (ALZ)2(ALX)2 > \h2 = (Ly)2. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2.20) or. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. (Lx±iLy) ^ 0.e.3 OneDimensional Systems 173 Thus c = — c*. . + LjLi\u). ALy = (u\L±\u) ^ 0. (10. Ly are not "sharp". implying {Lx)=0. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. (Ly)2=0.i relation holds: (ALi) 2 (AL. 9 = h{u\eiikLk\u). i. such that Lz is "sharp".z. m\lm)). for which one component of L is "sharp"? Solution: Applying Eq.) 2 > = x. (Ly)=0. h2(Lz)2.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. in view of Eq. the following uncertainty h2{Lk)2. i. What does the relation imply for states u) = \lm). (A{A))\u)=i(Zc)(B(B))\u).7a). = 0. The observables like angular momentum are functions of p and q. i. = 5R2 + G 2 > 9 2 = hi2(Lk)2. Separating the product into symmetric and antisymmetric parts. (10.e. q) = ih. For these we have in the onedimensional case the relation [p. (Lz)2^0. (ALy)2{ALz)2 > ~h2(Lx)2.y.e. For states \u). {u\LiLj\u) = R + iQ.13) to operators Li — Li. i. Example 10. Thus in this case Lx. ALX 10.e.
We have (cf. q]q + # > v] = ~2ihq and \p. (10. they do not have a common system of eigenvectors.C\. i.p)] = . Q2] = [p.p] = ih we obtain therefore [q.A(q.q). i. (3.21a) and (10.i h ^ ^ .A(q.pn]=ih^(pn). Then (see below) we can derive for A the following commutator equations [q. Eq.p}p + p[q. .174 CHAPTER 10. We obtain these equations as follows.p) can be expanded as a power series in q and p. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).21b) (10. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. This is why the operator p requires a representation in the space of eigenvectors of q.23) = ihg(q2) (10. A — A(p. Next we have a closer look at the operator q.p2] = [q.p)]=ih^^and \p. (10. and (c) the spectrum is not degenerate.22) d ih—(p2).21a) Equations (10.B]C + B[A. Similarly we obtain \p.e. (b) its spectrum is necessarily continuous. as remarked earlier.21b) now follow with the assumption that A(q.BC] = [A. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function).e.qn] = ih^(qn). Using [q.34d)) [A.p] = ^ihp ~ Proceeding in this way we obtain [q.
e. Replacing in Eq. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10.10.25) It is shown in Fig. an operator. we obtain [q. i.27) (10.f^ ^ J—oo dke ikx —ikx' (10. and a a cnumber. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization).1 how the delta function arises in the limit K —> oo. > We begin with (a). .26) it Here p is an observable.21a) A by U. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.24) lim SK(x — x') — S(x — x'). 10. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±. 10.28) i.3 OneDimensional Systems 6K(x) 175 Fig. = eipa/h = Uia)'1 = U{a). U] = ih^= aU. U is unitary.1 The delta function for K — oo.e. (10. (10.
31) . q) can be constructed representing a dynamical variable. or qU(a)\x') = (x' + a)U(a)\x'). cf. 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. Thus all these values belong to the spectrum of the operator q.1 T h e translation operator U(a) We saw above. b a cnumber). (10.q)U^U\ilj). Eq. that with U(a) = e~ipa/h = U. WU = 1. we can see that also p possesses only a continuous spectrum (from —00 to 00). the spectrum is not degenerate. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx').3. 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. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.29).00). q any other observable F(p. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). (10. Evidently every eigenvalue has only one eigenvector. 10. (10. CHAPTER 10. ^Observe that {iji\A{p.29) This means: U\x) is eigenvector of q with eigenvalue x + a. or qU\x) = U{q + a)\x) = (x + a)U\x).176 i.e. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a).e.30) (10.q)\i>) = (i>\U^UA(p. which means that the spectrum of this operator is a continuum.00). i. Since this holds for any arbitrary value of a in (—00.
/ (10.x') = = i.32) Since U is unitary. i. (x'\p\x"). We choose the phase such that U(a)\0) = \a).33) (10.x" .e. \c{a. we have q\x' + a) = (x' + a)\x' + a). The operator is therefore called translation operator. i.x") U{x'\p\x" % pz.e) ~ 5{x' .31) and (10.38) h 6(x' .39) .x')\ = l. offdiagonal elelemts of the {x}representation of the momentum operator.36) i.x")c(a. (10. i. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .34) {x"\U\a)U{a)\x') c*(a. In the expression of Eq. U(a)\x') = c\x' + a).e. It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). (10.e.e) (x \p\x ) = .5(x> .x')5(x" = (x" + a\c*c\x' + a) x'). (10.35) (10.x" .hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). the operator U(a) acts to shift the value x' by the amount a.10. since q\x') — x'\x'). Comparing Eqs.x") .x" .37) With the help of the latter expression we can calculate. it follows that 5{x" .e. (10. for instance. \x') = U(x')\0).37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . (10.e.a).32) we see that \x' + a)ocU(a)\x'). (10. (10.3 OneDimensional Systems 177 Moreover.
.x") = (x'\p\x")*.to) exists which is such that^ m)) = u(t. isolated from any measuring apparatus which would disturb the system and hence its behaviour in the course of time) and let \tp(t)) be its state vector at a later time t > to (with no interference in between). 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)). Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E. See Eqs.t0) = HU(t. In this case the classical Hamilton function does not depend explicitly on t.40) 10.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i.to) is called the time development operator. . Quantum Theory: The General Formalism We mention in passing that. (10.e. i.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian. to) with respect to t we obtain ihjtU(t.to)m0)).43) (10. The operator U(t.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . Differentiating U(t. (10.e. H\uE(to))=E\uE(to)).27) and (5.L fdpeipx. i.t0).e. a conservative system.4i) We assume first of all.° ° d(p) (ip) e"« = S'(x). since 5(x) = . 6'{x) = !. so that U{t. it follows that 5'(x) is an odd function of x.178 CHAPTER 10.35). (5. This is in conformity with our earlier postulate in Sec. 5. (io.4. 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.f dpipeipx.44) (10.45) (10.tQ) = eiH(tt<Mh.
t0) (10.t0)]1 Let us set for small values of e: U(t + e. From Eq. Ufa.z / dt'H(t')U(t'.10. = U(t0. (10.t) = l.45) has very general validity.*l)^(*l)> = I7(*2. But obviously (cf.t). (10. (10. (10. Eq.t) and therefore [U(t. The operator H is d efined by this relation.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.to) (differentiating we obtain immediately Eq.tiM*i. (10.to) = 1. (10. t0) = E/(t 2 .*l)^(*l.48)). (10.49) .t)l]U(t. i.t0)\^(t0)). Then with U(t + e.t 0 )^(*0)> U(t2. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.47) e.t0)U(t0.t0) [U(t + e.e.t0) or ih—U{t.44)) we have U(t.48) with U(to.to).48): [/(Mo) = 1 .t) = l^eH{t). Evidently U(t.41) we obtain ^(t 2 )> = = J7(t2. t0) = H(t)U{t.t)U(t.4 Equations of Motion 179 We now demonstrate that Eq. Hence U(t.to) can be expressed as an integral as the solution of Eq.t) = 1.
5o) ihjtm)) = Hm)). (io. it follows that U(t + dt.53) . >(*)> = i.t0)\ii>s{to)) (10. 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. Since the only measurable quantities are moduli of scalar products. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.30). the Schrodinger equation ju(t. (10. Then \Mt)) = u(t.(£ + dt)) = U(t + dt. The operator U(t. 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.t0) m0)).41).t) = l % ^Hdt) \il>(t)) Hdt (10.180 CHAPTER 10. (10. and these remain unchanged under unitary transformations. requires the identification H(t) = H= Hamilton operator. the instant at which a measurement is performed on the system.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^).e. The eigenvectors of these observables are also constant in time. t)\t/>(t)) =U~ and since H is hermitian.e. Eq. Another such description is the Heisenberg picture. i. (5. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. Comparison with the equation postulated earlier. the predictions of different descriptions are identical. 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. Now let \tp(t)) be the state of the sytem at time t. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.51) Since </.
54) Observables transform correspondingly with a similarity transformation. so that a matrix element (i.With explicit time dependence. as a rule. (10. we consider only cases without explicit time dependence of As. in order to obtain the Heisenberg picture (description).58) HHAH (U*ASU)(U^HU) [AH. For As(q.(tf HU)(rfASU) (10. As(q.57) and rf[As.55) Thus AH is explicitly timedependent. It does not always have to be the case that As does not contain an explicit time dependence. t0). (10.HH]. {AH. .10. (10. Differentiating Eq. (10. i. Thus.e.t0)\ips(t)) = \ips(to)) = constant in time. for instance in cases of nonequilibrium or if part of a system disappears through absorption. we have to take into account the contribution dAs/dt. = so that ihjtAH or ihjtAH = = [AH.e.59) This equation is called Heisenberg equation of motion. 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. even if As is not timedependent.55) and using Eq. subject to a continuous change. if ^ = 0. However the associated observables are timedependent.p. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10. (10.48).HH]. In the Heisenberg picture it is HH = U]HU. but here. Here H is the Hamilton operator in the Schrodinger picture. i.i).e. 181 (10.4 Equations of Motion and \^H) = U\t.HH]+ihU^U.H]U + iW]^U.56) U^[As.H]U = = = U]ASHU AHHH — U^HASU . t0)AsU(t.p) in the Schrodinger picture dAs/dt = 0. a scalar product) remains unchanged: AH(t) = U\t.
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. Example 10. An observable CH is called a conserved quantity. [CH. In many cases the Schrodinger picture is more amenable for explicit calculations. We end with a word of caution. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms. We also observe that the momentum pi is conserved when \pi. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. Finally we add a comment concerning conserved quantities. provided the Poisson brackets in the latter are replaced by commutators.13) to derive the uncertainty relation for energy and time. 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.Pi we have in the Heisenberg picture (replacing in the above AH by qi. since in the transition from one to the other the commutator remains unchanged. In our treatment of representations.21a) and (10. (10.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. Orfanopoulos [224]. we used in essence unitary transformations of matrices. In particular.182 CHAPTER 10. Solution: We have for an observable A (A) = (i>H\AH\TpH).21b). i. (10.e. in the case of the fundamental dynamical variables qi. For an extensive discussion see B. In the treatment of the pictures we used unitary transformations of operators (and vectors).HH}=0. if ffCH = 0. A. We see that formally one obtains Hamilton's equations of classical mechanics.61) Thus conserved quantities commute with the Hamilton operator. This holds also in the case of the Schrodinger picture.Pi): and where the expressions on the right follow from Eqs.H] = 0. .
7a).64) = 0. i. (10.15): (AAf(AH)2 Using Eq. (10.62) The Hamiltonian H does not depend on time. (10. since \IPH) is timeindependent. AW = {A(A»V>. in a different form: The eigenvalues of an observable are called "good" if [A. if we set AH = AE we obtain TAAE (10. i.10. of A\<t>) = a<£)) are then called "good quantum numbers".65) and rA AA d{A)/dt' . (10. > (10. H] = 0. then if a measurement is made at time t'. If r is the smallest such characteristic interval of time.66) This means.62) and dAs/dt > \WAH\i. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) . a position or momentum observation is independent of t. dt in.63) so that with Eqs.\HAM\2) \(nA. 183 dV ' \ dt >"•"»» + ( £ ) • (10.HH]) Hence we obtain by ih(A).13) and (10.H}\f)\2. we can replace ([AH. and TA O C (10.68) To put it The eigenvalues a of A (i. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t.Such a characteristic time interval can be defined for any dynamical variable. (10.e.)\2 ( = \{i.e. . 2 d{A)/dt " and. This means that for a physical variable A(q.e.67) 1 d(A)/dt is oo and AE = 0. H\tl>) = {H(H))\f).e.p) we then have 0=^M) = U[A. this is practically the same as at time t for times t with \t — t'\ < T.H]) = 0. (10. (A). i.4 Equations of Motion Then. Let \ip) = \"4>H) represent the state of the system. We construct the following vectors using Eq. the "centre of mass" of the statistical distribution of A.2 > h. is displaced by A A in the interval A T = TA. at ^±AH>h.
70) we can determine the function fEo{E) integration with respect to t. i ) = J dE'^El^)eiEltlhfEo{E').5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. states with sharp energy (AE = 0).tyEt/hdt = = = / J—oo dt J dEJrl>E.)exp(iEt/h)). But the number of radioactive (thus excited) nuclei decreases exponentially. the wave function ipEo(x. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime. (10.70) we obtain: oo / oo TPEO (x. Quantum Theory: The General Formalism 10.69): oo poo p / oo *PEo(x.(x)fEo(E')e«EE'Wh 2irh f dE'^E.71) From Eq. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. the wave function ^ ( x .E1) 2irhfEo{E)4>E(x). First we obtain from Eq. /oo (10. i. as is observed.H ) ( £ . Recall for instance radioactive decay. (10. This means that in the case of decaying particles. Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei.70) by From Eqs. This means.e.t) — ipE(x. (10.184 CHAPTER 10.69) and (10. (x)/ E o (E')5(E . then fE{E') = S(EE') and integration with respect to E' yields again t()E(x.72) = V'EoW 7/2h + 0i(EE )/h  oo . The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 .69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. (10.Eo(x)eiEot/he'^2h9(t). for instance. (10. have around EQ the form ipEofct) = i.. the particle density diminishes in the course of time (t > 0). On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x .
Considered as a function of E. £/ = £/W(Mo)^'(Mo).75) (10. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. We saw previously that if the state of a system is known at time to. . 10. where U is the solution of Eq. (10. ihU(t. We obtained the result (10. They are called "resonance states".e.74) Thus the solution of this equation is the main problem.45) can be solved only approximately (in actual fact. The state with lifetime r — h/'y is not a discrete state. i.76) **For specific applications see e. whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use.e. and is called "interaction picture" or "Dime picture".t0). (10. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. to) be an approximate but unitary solution of this equation. that** EQ. every unitary transformation of states and operators defines a possible "picture"). 7 > 0.e. (C/(0) unitary). i. Eqs.toM{to)).to) = HU(t.e. i.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). (10. Let U^°\t. (14. ^(*o))> then \xp(t)) for t > t0 follows from Eq. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum.41).g. of course.10.45). (10. 185 i. W)) = U(t. This description contains effectively parts of the other two pictures and is particularly useful when Eq.105) and (20.e. i.11). (10.
83) ih^U^ ishermitian. (10.79) ^'~°i. H(°Ht) = ih^uW. = HU<®U'.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. . In addition we have (10.e. £/(°)£/(°)t = 1. with Hamiltonian H = H^ + H'. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). [/' has only a weak ^dependence (i.82) (10.78) (10. (10.e. varies only slowly with t).h^U> = at ( io. (10. with Eq. i.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.186 Then CHAPTER 10.e. Quantum Theory: The General Formalism ih^(uWu') i.e.81) But [/(°) is unitary.e. Now let H^(t). i.e. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. i.t0) 0 = l.e.
(10. We also set H^U^H'UW. U^H'U^U' (10. and obtain from Eq.86) at = U^HU^U'. (10. (10. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10.85) multiplied by U') on the right of Eq. (10. (10.U^HU^U' H'fi'.e.90) = U^H'\^s(t)). H = H{®{t) + H'. (10.84) U^ where H' is also hermitian. (10. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. (10. (10. i.85) = where we used Eq.87) = We define correspondingly as interaction picture state iM*)> = t/ (0).78) at (10 85) (10. U<®\Mt)) = \Mt)). We multiply this relation from the left by and from the right by U^ and obtain umHU(o) = c/wt (^WtjW) + umH>Tj(°) i n u ^ ^ + U^H'U^.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.10. (10. lfe(*)}.78).89) where ^s(i)) and As are state and observable in the Schrodinger picture.92) .80).91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)).
ot i. In order to obtain the equation of motion of an interaction picture observable Ai. dAs/dt=0) .e. near zero and \tpi(t)) is almost constant in time. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). (10. (with ihjtAI(t) = [AI. (10. cf. A^t) = U^AsUi0).93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. (10.60a) or (10. a perturbation so that H ~ H(°)). (10.90) and replacing the left hand side of this equation by the right hand side of Eq. (10. comments after Eq.e. ^ (0) l^(*)>.89). (10. H'j = UWH'UW.o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.188 CHAPTER 10. we have i.55)) at + imm<Ms_u(o) at dt _umH(. Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq.e.94) _Hf)Al 1 + lhdAl+AlHf).92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .89)^0. so that with Eq. i. (10. the vector varies with time (different from the Heisenberg picture). Since H'j is assumed to be small (i.e. i.94) Then we have (normally with dAs/dt dt (io=8o) — 0.H\0)}.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10.e.e.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)). i. (10. we differentiate with respect to time the operator defined by Eq.60b) with iff0) instead of H.
i0) = e<Ho(tto)/ft.99) (1098) Instead of the original Schrodinger equation we now solve Eq.t0)H'uW(t. (10. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.98).7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture.e. We obtain the operator U^ from Eq. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. {ipn\<Pm) = Snm. (10. (10. H' = H'(t). (10.10.97) The equation to solve is Eq.e.80) as E/(°)(t.93). We assume again that the spectrum and the eigenvectors of Ho are known.87) and (10. We have the Hamiltonian H = H0 + H'. H0\ipn) = En\(fn). ^ \<Pn){Vn\ = 1n (10. i.100) (10. where according to (10. Since .96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion. ih±\Mt))=Hi{t)\Mt)).7 TimeDependent Perturbation Theory 189 10. i.e.94) Hi = uW(t.t0). Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')). (10.102) We assume that before the perturbation H' is switched on at time to. i.
(10. Hence we set ^> = £°«(*)l4 0 ) >n.103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). 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'.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10. (10.102) and truncate the series after the first perturbation contribution.104 ) Here we insert Eq. i.e. (10.107) Jto With this we obtain from Eq. (10.^ V ™ ) . (10. (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.106) We insert this expression into Eq.103) for m — n and use Eq.108) .105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. (10. We express this state \tps) as a superposition of the states \ips ) n of #o with coefficients which as a result of the timedependent perturbation depend on time. Quantum Theory: The General Formalism HQ is independent of time. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 .190 CHAPTER 10. we can write l 4 0 ) ) m = elHot/hWm) =e . the time dependence of \ips } can be separated as for stationary states.
This is the case.10.Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. for instance. if in the scattering of a particle off some target its momentum p changes to p'.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n .109) 10.111) Different from Eq.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . A further example is provided by adecay. < KO?t . or if by emission of a photon with continuously variable momentum an atom passes from one state into another. where the momenta of the aparticles form a continuum. » — < (10.110) With this we obtain from Eq. (10.Em)t} \{tfn\H'\ipv h 2h(En .En (10.109) the transition probability Wmn(t) = jp Jo . but that otherwise the perturbation is timeindependent. (10. (10.112) for a^0. t o = 0.8 Transitions into the Continuum 191 Hence the probability for the transition of the system from state m into the state n ^ m is K(t)\2 = I fdt'e^. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. (10. We assume that the perturbation is switched on at some time to.Em)t' (<Pn\H'(t')\pm) = Wnm(t). Thus we set H'(t) = H'0(t).
113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.io. 10. > 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. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum. though sufficiently small. and no other state would be nearby. Hence lim St(a) = 5(a).Em)t} r i 7 TTT.2 The delta function for £ — oo. .— = o (En 2h En 2h5{En .E7m)t} = (10. 10. if the energies belonged to the discrete spectrum. we would have En — Em ^ 0.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. so t h a t Wnm would be zero. Otherwise. Fig.2.192 CHAPTER 10. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .Em).
75. Fermi gave this formula the name "golden rule" .4)./ t>oc TZ J_00 dx . (10. Fermi [92]. F Jo . We evaluate the integral 0(0) = 1. if p(En) varies only weakly with En. Then. (10. Pauli [227].* In view of the usefulness of Eq.113) Use the method of test functions to verify in the sense of distribution theory the result (10. 142.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. for instance from a potential. _ 2irh AE > > Se.8 Transitions into the Continuum 193 as equal. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t.3: Application of test functions to formula (10. (10. In the case of the functional (10. in the article of W. If we are concerned with pure scattering. f See E. 266. i.r Z x 1 ^ ) = 4>{Q).W .10.e. (10.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . Schwabl [246].116) The transition rate V is the transition probability per unit time.117) provides the outgoing particle current (see Example 10. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). so that initial and final states belong to the continuum. pp. Example 10. then Eq. Let p(En)dEn be the number of states in the interval dEn.113). p.117) were derived by Pauli in 1928.117) in many applications. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.115) and (10.113) this is* hm . sin 2 (tx) ax i—( * According to F. .
. . 2 . M. . V +(ab)2 a _x 2p6 7 m —^ '. L \3k2dkdUn It follows that T.e " 1 * 1 = 2 lim — .. iknx . = 0 .„I. p = 1 we obtain I = . 77^ + . 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.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . x . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) .. S. .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn .947. . Then ^ . 491.tan p 2 + a 2 .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). . The ingoing particle flux is therefore v / L 3 . with n . ^o n .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n .? 2TT / L \3k2dkdUn 1 /V(x).194 CHAPTER 10. .4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. 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 .<p). § We have r Jo \p\x dx sin ax sin bx  V. Ryzhik. Jin § m = See I. * ^ = . 2 T / ^ M. 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. Gradshteyn and I. dEk = h2kdk//j. = ( — ) dk= 2ir (—  2irJ k2dkdfl..a 2 ' i that for a = b = t .. formula 3./ cfc a n p 2 +fc 2 .( ^ V V • < ih 2 _ i k n . Quantum Theory: The General Formalism with the help of Tables of Integrals. Then p(Ek)dEk Since Ek — h2k2/2[i.k. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. E x a m p l e 10. and is obtained as ih. Hence the result is 0(0) as expected. and hence . . 1 . E = h2k2/2fi. . . p. with v the velocity of the particles. .
Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.^ We start from the Hamiltonian H = HQ + H'. by which we mean without leaving the Schrodinger picture. t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn). 1 rfx^°)*(x)^)(x) = 8mn./continuum = <San(£)</4°)(x.t)+ / n a{E. (10. For a possible application of this result.^°>(E. we present another derivation using the usual method of timedependent perturbation theory.t) = ^On(t)^(x. (10. 148. We expand the wave function V( x .121) . pp. see the discussion after Eq. Thus we set ^(x.122a) (10. 193. (10.122b) S e e also L.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. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>).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.69).10. (21. specifically in the case of the Coulomb potential. 10.x)^(E'.t).9 General TimeDependent Method For a better understanding of the "golden rule" derived above.120) In the case of Eq. . Schiff [243].x) = S(EE').t)^°\E^t)dE (10. (10.
t)dx. (10.„„.122a) and (10.x.128) . . (10.196 CHAPTER 10. dnptMrnW = { ^ _ En) (10. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).123) We now insert these expressions into Eq. (10..129) . (10.iJ5 »*/Vn(x). (10.125) where an(t) — dan(t)/dt.. (10.126) + H')<pn^)eiE^h}. .127) i n c a g e Qf c o n t i n u o u g £.t)rJ><®(E'.x. since f rJ>(°>(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.t) = eiEtlhy(E.e.127) we obtain from Eq. = e^EE'^h5(E Thus with Eq.x). into dip (10. f 6kn in case of discrete tp's. We multiply this equation from the left by </?£(x) and integrate over all space.E') = 5{E .124) and obtain ihSdn(t)ipn^)eiEnt/h + San{t)En<pn(x)eiEnt'h = San(t)(HQ + H')vn{*)eiEntlh = San(t){En + H')vn(x)eiEnt/h..E'). V>(x.122b).. Next we use the orthogonality relations .. The second term on the left of this equation and the first term on the right cancel out. Observe that these conditions do not contradict Eqs.x. .i) = e. t) = San(t)<pn(x)eiE^h. . Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. / *. ^0)(E. (10. i.„.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10..120).
Comparing coefficients of the same powers of A on both sides. i. We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. (10.10.130) This result should be compared with the Schrodinger equation (10. (10.130) we develop a perturbation theory. let's say am / 0. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.131) and insert this into Eq. (10.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] .9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h.120). i. In the first place we replace in Eq. In order to solve Eq. 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. SH'kn(t)e^~E^ha^(t). (10. to perturbation theory in the lowest order.e. determine the state of the system at time t — 0. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• .e. and we set ak°}=5km or ak0) = 8(EkEm). Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10.132) These equations can be integrated successively. Thus first we have ak0) = const. (10.130) along with the parameter A in front of H'kn. (10.130) H'kn by \H'kn and at the end we allow A to tend to 1. We restrict ourselves here to the two lowest order equations. since it is equivalent to the latter. SH'kn(ty^E^l^\t) (10.134) .133) The numbers ak = const. 0. they are fixed by the initial conditions. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.
(10. ipm) into the state k.136) Comparing Eqs.136) we see that both "amplitudes" are Fourier transforms and look similar.ml2sin2{^frn (En . so that ihakl\t = oo) = 0. Integrating the second of the equations (10. whereas ak is a Fourier transform in time (a distinction which is easy to remember!). an expression of the following type AS?) = " 2 ^ 2 / e j ( k .135) assumes a particularly simple form if — as we considered previously — the perturbation H' is taken to be timeindependent except that it is switched on at a particular time t = 0 and is again switched off at some later time t > 0. Equation (10. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. i. In this case we obtain from Eq.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. ) ' X V(x')dx'. The amplitude / B o r n is a Fourier transform in spatial coordinates.135) We put the additional constant of integration equal to zero. What is the physical significance of Eq.198 CHAPTER 10. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum.e. (10. ip) in the context of timeindependent perturbation theory.e.k .135)? The quantity ak\t) is the amplitude which determines the probability that as a result of the perturbation ^H'km the system makes a transition from the initial state m (i. (10. (10. i.135) and (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.Emy in agreement with Eq.111). ifm{kl) = f J —oo (10. (10. H'^y^"^^/hd£.e.137) . (10.132).
2. 199 (1L2) .i be respectively the momentum.y. such as the electromagnetic field (the quantization in these contexts is often called "second quantization".Chapter 11 The Coulomb Interaction 11.m.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. the position coordinate and the mass of particle i with i — 1. The Hamilton operator is then given by w= jt + ji_^M. r=  r . Let r : = (x. Z<i = Z) y(r) = _ M ^ .Ti. 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. ( 1L1 ) We let pi.2 Separation of Variables. 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).1 Introductory Remarks The quasiparticle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields. 11. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.
z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. where m 0 = mim 2 . The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . . Z) and r = (x.3) for the individual position coordinates we obtain ri=R+ •^ —r. 2Pip2 m\m2 pip2 mi + m 2 . (11. y. mi + m 2 iTi'2 r2 = R ^ —r. Solving Eqs. (H4) \mi m. P = pi + p 2 . /'Pi p = m0 P2 A v. mi + m 2 va\ (11. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2.4). Pi 2 2mi(mi + m2) and P2 2(mi + m 2 ) Pl"»l + pjmi 2m 2 (mi + m 2 ) (P1+P2) 2 2(mi + m 2 ) P2 f f l 2 Pi P2 ^ 1 1 7 ^ 2mi(mi + m 2 ) 2m2(mi+m2) mi + m 2 ' . r} representation they have the differential operator forms Setting R = (X.200 CHAPTER 11. p are such that as operators in the position space representation or {R.U7&.5) The momenta P . so that as expected we obtain the expression for p in Eq. (11.4) P = p i + p 2 . so that as expected according to Eq. (11. Y.2/ mi+m2 The mass mo is usually described as the reduced mass. r = rir2 mi + m 2 (11. Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc.3) and centre of mass momentum P and the relative velocity p/mo by setting r» .
2(mi + m<i) 2mo 2m\ 2m.2 It follows that p2 2 P2 n2 n2 n2 (H. .il) ""Op' i.Ri.10) we set H = Hcm + H.13) * and = M ' 1V1 P = 0. 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). (i.+ P. with M := mi + vn^R = ^ P. (11. we mean n = rx s x. r = —.z).pj]=iMij.Pj). pj.= ^.pj.16) Although we write ri. etc.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. (11. p = W. for which we postulate the commutator relations* [ri. (1114) m0 We observe that the equations of motion in R. (11. (11. r) is a canonical transformation.11.y.pj. p on the other are completely separated. (11.2 Separation of Variables. Ri.12) (11.+ ^2. [Ri.j = x. Angular so that Momentum 201 ^ ^ . Pj (which in the following we write again ri. The canonical quantization must always be performed in Cartesian coordinates.8) and so W =777 v + 7 T .Ri.r2) to (R. Hence Hamilton's equations apply and we obtain.Pj} = ih6ij. Pj over to operators ?i.+ ^(r) 2(mi + m 2 ) 2m 0 (H. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.e.10) One can convince oneself that the transformation from (ri. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. P on the one hand and those in r.15) In accordance with Eq. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii.
22) If. 0 < r < oo.24) . (11. V(r) = V(r).18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . 2mo (11.20) r2MAR_ $(R) = ER*(R). —n < tp < ir. (11.r). as in the case of the Coulomb potential. r ) = J E t o t a i*(R. i.Q 2M' + V(r). to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R.23) y — r sin 9 sin tp. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11.r) = *(R)V(r).17) (11. (11. (11.21a) L*+v{r)] ip(r) = Ei/j(r) (11. Thus the twoparticle problem is practically reduced to an effective oneparticle problem. = rcos9.202 CHAPTER 11. ip: x z = r sin 9 cos ip.21b) + E. it is reasonable to go to spherical coordinates r. where Hcm$(R) = and Hip(r) = with ^total — ER (11. 0 < 9 < ir. The Coulomb Interaction nc H = 2m.e. 9. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .
2. (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. where (r x p)j = ^eijkrjpk._0di})\ x We now demonstrate that 1.k a clockwise permutation of 1. i( i. k = 1.3.2 Separation of Variables.27) We can now write (see verification below) p V = h2Ar^ where 2.31) . (l + .1 it is shown that pr is hermitian provided lim n/>(r) = 0. (11.32) (11.2.3.11. is the operator of the relative angular momentum. i—>0 (1L25) (11. j . n ( . (11.i ' . i. 2 1 9 / 1 9 . the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. (11. (11. We have I2 = (r x p) • (r x p).2.pr]=ih.12 — I2.3. _ h2 \ .28) .j.A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. . Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«.29) .5r = .26) However.j. otherwise. = 0.)In Example 11. if i. k an anticlockwise permutation of 1.33) i. with tijk — + 1 . = —1. = (p2 + ^ V (r + 0).
(11.ihr • p ) .44) . Eq. (11. * ' S Since r =xz we have (cf.(r • p ) 2 + 3ih(r • p) . The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p . (11.40) (11.38) + y* + z .ihr • p r2p2(rp)2 < + ih(rp).41) o = r pr.35) j.k Using Eq.34) W i t h this we obtain — keeping in mind the ordering of r and p — I 2 = 5 Z eijkrjPkeij'k>rj>Pk' = ^{rjPkrjPk i.42) (11.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .ih5jk)pk (11. (11.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11.15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.j. This means that l2 = so that 9 9 1 (11.37) . r— = r • — or or (11. (11.43) r2(p2p2).ih)pr and x 2 2 9 9 2 2 Q O (11. (11.27) we obtain r{prr)pr = r(rpr . ih6jkrjpk (11.39) + ih).k ~ rjpkrkPj) (11.ih) — rpr(rpr With Eq.
Pri/Oi V0. h f l d —(r<l>*(r.26).<p))ril>(r.28) and (11.[ i J With partial integration with respect to r this becomes (<l>. S o l u t i o n : Set d£2s = sin. 9 € [0.6d0dip. prip) = = I <j>{r.Pril>) J / r= dQs rcj>(r.— (riP(r.e..45) oe Thus in Eq. 11. ip 6 X>Pr C 7i.<p)*?(ri.e. H 1 2mo pi + K) +V(r).<p) (11. ¥ p)r 2 drdn s = (p r 0.2. <p)*rip(r} 0. 9. 6.(r.6l.47) Example 11.1 Separation of variables We deduce from the expressions of H and l 2 that [H. 0.e'iP^} .Q ' 2mor2 with the condition (11. <p) d °° . <p)* . pr = . r—>0 il>(r.48) .ip) = (<£.<P) = Etl>(r. <p)*Pri>(r. /'(r. 6.e.6. (11.30) I2 is to be identified with the square of the angular momentum operator.7r].30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11.d. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.2TT]. Angular Momentum 205 Comparison of Eqs. i.[ 4>(r.e.l2} = 0.2 Separation of Variables.! /').e.e lim rip(r) = 0. or J . <p))r2drdns % J r or r<t>(r.<p G [0.. reR3. r := r : lim rip{r) = o i . i.<p))drdns.46) (11. Then {<j>.11. (11.d_ r dr the operator pr is hermitian.<p)drdna " r r=0 J \ll§r~(r4'(r. 8. (11.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3). (pr4>. <p)r2dfls = .
These operators here play a role analogous to that of the quasiparticle operators .q2 = V.53) Either of these operators is the hermitian conjugate of the other. However. H and l 2 possess a common system of basis eigenvectors.ly]=ihlz. 11.e.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .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. [lyA = 0. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. ly] = [yPz ~ zpy. Example 11. In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.49) By cyclic permutation of x.lz (or l . we have [h. i.2: Verify that [h.lx] = ihly. Since 1 = r x p .52) This means that l .pz]x = ih(xpy . ?3 = z).e. i. (11. zpx] + [zpy.e. y. 2 . 3 .l2} and correspondingly [lx.xpz] = [ypz. xpz] = y\pz.l2] = 0.206 CHAPTER 11. The Coulomb Interaction i.51) (11. [lz. so that [lz. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously. z]px + py[z.ypx) = ihlz. (11. We therefore first search for a complete system of eigenfunctions of l 2 .lx or l .ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. (11. /z> *xJ = l <i\}ylx + tx'j/Jj [h. for any two of the components there is no common complete system of eigenfunctions. 2 2 2 = 0.50) Thus the components of 1 do not commute pairwise. zpx . [ly. g i = x. (11. z we have altogether [lx. they are incompatible variables.lz]=ihlx.
60) Similarly for l+ with the help of Eq.49) we obtain the following relations: [iz.58) {l0 : max. assuming a finite dimensional representation space. (11.52): 0 = [ l 2 . Similarly we have lg{l)2\l) and more generally lz(l. l\l) is eigenvector of lz with eigenvalue (IQ — l)h. (11. i + ] = [l2. [ig.51). With the help of Eq.i] = m.+ ll.61) But lo is by definition the largest eigenvalue of lz.e. .r) (11 = {l02)h{l)2\l) (11. (11.3 Representation of Rotation Group 207 A. Therefore (lo + l)h cannot be an eigenvalue of lz. (11.11.r) := (Z_)rZ) so that lz\l .Y\l) = (lQr)h(lY\l). in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. y .h~lz(1156) (11. (11.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. We write or define: / . since lz is hermitian. [i+M = 2%iz.l)hl\l). A^ in the case of the linear harmonic oscillator.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real.54): lzl\l) = {llz . (11.54) l 2 = \{l+l~ + 11+) + ll = l+l.59) = 9) h(l0 r)\lr).55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. (11. Hence we must have Z+I0=0. (11.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). (11. (n.Z_] = [ l 2 .lh)\l) = {l0 . (11. (11. and with Eq.i+] = m+.54) On the other hand we obtain from Eqs. In the present context the operators are called shift operators for reasons which will be seen below.).57) i.
Z_] = 0.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11. i. i.y\i) l)h2\lr).n)}h2\ln).e.67) i (11 60) = But according to Eqs.e. (11. l\l .+ 11+) + l2z\l) = i+i.. (11..l2\l) = l0(l0 + l)h2l\l).63) + l ) ^ 2 .60) and (11.54) (n = 57) (11.e. (11. i.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.56) CHAPTER 11. (11. \l).n) (1 = 60) {l)n+1\l) = 0.64) i. [12. (cf.. (11. IQ(IQ 1 0 (11.\lr) terminates at some number (let us say) r = n (n a positive integer).. (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.n).55)) we have 12Z_Z> = l.From Eq. (11.+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i).65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .h)\ln) {(lQnf(lQ.208 Thus for l 2 we obtain: V\l) ' = 11.\ll). Eq.66) It then follows from the second of relations (11. l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 . (11.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.e. The Coulomb Interaction (l+l.68) .
.n)} = (l0 .l.l_. \l — n) into one other. 17. (11. m + l)h.m). § See e.1).m±l\l±\l..2.. R... Satchler [38]. According to Eqs. *D.m\lz\l.g.m)... lo = l0n. The nonvanishing matrix elements of 1 = (l+.69) Prom this it follows that IQ must be either half integral or integral. The dimension of the representation space is therefore n + 1 = 21Q + 1.Z 0 ). .{lo .71) can be established in analogy to the method used in the case of the harmonic oscillator. (11. we skip this here.m) = mh\l.  U o . lz) are then: (l.m) = = = = y/lo(lo + 1) — m(m +l)\l. i. 24. y/(l0±m + l)(l0^m)h.e.70) According to Eq. where lz\l. The basis vectors \l). lQ = . The operators l+. In the following we consider primarily the case of integral values of IQ.m) l\l.1)..m + 1)(Z0 + m)\l. \l — n) are eigenvectors of lz with eigenvalues loh..67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . p.m) = = mh.m) : Mo). A. We therefore write the 2/o + 1 orthonormal vectors: \l.m) (l.70) they satisfy the eigenvalue equations (with m = l0Jo l.l)h. II. —loh. We write these* for integral values of IQ: \l.. (11..n){l0 .m)\l. p. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator. The phases are chosen such that§ l+\l...m).m / + l)h .lz transform the vectors \l).3 Representation of Rotation Group so that with Eq. lz\l.11. (11.m y/lo(lo + 1) — (m — l)m\l..65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. As in the case of the harmonic oscillator we introduce normalized basis vectors. (IQ — l)h.m) = IQ(IO + l)h2\l.l0): l2Z.m) — mh\l.n) 2 . Brink and G.n . Messiah [195].m) with — lo<m<lo2 (11. (11. The normalization factor in Eq....m . M. y (lo .63) and (11. Vol.l)h \/(lo + m + l)(l0 .
72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. The Coulomb Interaction Z+Mo> = 0. . as remarked earlier.<p). $ oc eim*. where IQ = 0.yp.. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ .74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).210 Then. mhY£(9. i. l_\l. as above. as this is sometimes described..With the ansatz Y(9.75) the variables contained in the expression (11. (11.. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). /o.e. (11. .76) d2 (11. .<p) = i0{io + lzY£(0.77) $((p) = m2<b(y). —lo + 1 .<p) = i)n2Y^{e^). ip) = e(9)$((p). In these coordinates we obtain: h l± so that h2 = xpv . CHAPTER 11. (11. . for the simple "rotor".l0)=0.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. with i2Y£{e. We . 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. We now perform the analogous procedure for the angular momentum or. 9.. and m = —IQ. d2 sin2 9 dp2 _' (11.73) 1 d d sm0 06 \Sm9dd) + (11.1.2.4 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. ip.
4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.\ = W .n>).=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. m > 0.76) and (11.82) . This is the case if one demands that l 2 and lz possess a common complete.ypx = tfrK(11. These functions are defined in such a way that they are normalized to unity on the unit sphere.V>)Yir'(8.81) V l07T V47T y? = Y? = One should note that the functions Y™ are square integrable only for the given integral values of m and I. (11. The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. is given by (2Z + l ) ( Z .78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.r o ) ! YT{e. orthonormalized system of eigenfunctions.(5cos 03cos0). (11.11. Uniqueness implies immediately that m has to be integral.(p) are called spherical harmonics.<p) Jo = 6mm'5w. (11. so that im(ip±2ir) The functions Ylm(8. and that their phases are such that Yt0(0.0) is real and positive. (11.i».79) . We mention in passing that in spherical coordinates d lz = xpy .<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.^ = *(n . y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.* ? ^ .t»'\Y7n((l'. (11. / 3 / 5 J — casO.80) y? Furthermore Y°l = V^fcosO).
i. is also in use.x2)Pf .89) Inserting this expansion into Eq. (11. Consider the equation for Pi(x). f. i. of Ytm.. 22 (11. i. exp(imip). i.e.h.2. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed.e.m = 0. If we did not know yet that I = IQ = an integer > 0.87) we obtain the equation (1 .1(1 + l)]xK+i = 0.d. (1 . in the replacement ip —> if + 2K.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .x2)P[' .Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) . The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 .. oo (11. For the solution we use the ansatz of a power series.1. respectively..90) .e.88) y = ^aiXi+K.p. (11.85) (11. i=0 (re lowest power > 0).x2)y" . we would now have to search for those values of I for which the solutions are square integrable.212 CHAPTER 11.2xy' + 1(1 + \)y = 0.m(m + l ) ] i f = 0..x2)y" . It is instructive to pursue the appropriate arguments.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.2(m + l ) x i f + [1(1 + 1) . (11..g. Instead of the values I = 0. (11..e. the designation s. (11.86) y = 0.l)xK+i~2 i .2xP[ + 1(1 + l)Pi = 0. Setting y = (lx2)m>2Pr(x) m .88) we obtain for every value of x ^2 ai(K + 0( K + * .
Eq. > 1.*" 11.Qr2 with (cf. M.94) from the left by r) h2 d2 h2 2 + H! + \)IL. .94) ^For further details see H. This is precisely the case.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11. (11.1. first line of Eq. ai(re + !)« = 0. these are of little interest to us here. (11. Murphy [190].11.91) from which we deduce that K = 0. (11. p. For \x\ < 1 we can obtain from this the coefficients of a convergent series. 93.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.E Xl(r) =0 (11. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) .96) + V(r) . Margenau and G.92) This is a recurrence relation.26)) yi(r = 0) = 0.93) where Xl(r) is the solution of the radial equation (cf.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0. l(l + l)h2 _2mo + 2m.1(1 + 1)]. (11.78) to be square integrable. (11.5 R a d i a l E q u a t i o n for Hydrogenlike A t o m s The common set of eigenfunctions of the operators H.46)) p. they have to be polynomials (recall the orthogonalization procedure of E. We have therefore (multiplying Eq. We see that this happens precisely when I is a positive or negative integer or zero. However. (11. . when the series terminates after a finite number of terms.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. Eq.2 + 2mo dr 2mor V{T)E Viir) = 0 (11. but also those for a different series for a. l 2 and lz are the solutions of the Schrodinger equation of the form (11. Schmidt). For the solution in accordance with (11.
and others. Z\ = Z2 = 1 in Eq. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.e.1.214 with CHAPTER 11. He + . The equation also has an irregular solution. We now consider the hydrogen atom with V(r) — —e2/r (i.e.96).99) e ip(r) = Eip(r).102) . (11. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. oo) of a onedimensional space. Obviously the irregular solution has to be rejected.1)+1(1+ 1) = 0.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. we obtain the socalled "indicial equation!'' s(s . (11.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. 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. r m0 = meMp me + Mv (11. since it satisfies neither the condition yi(0) = 0. Equating the coefficient of 1/r2 to zero. i. and substitute this into Eq. nor the condition of normalizability for l ^ 0. L i + + . for instance.100) with (11. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ). This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. The Coulomb Interaction (11. an analogous consideration applies to hydrogenlike atoms like. s = 1 + 1.1)). In the first place we investigate the behaviour of yi in the neighbourhood of the origin.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr. which behaves like 1/r1 in approaching r = 0. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. (11.
529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11. hypergeometric (11. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT.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.103) °y moe2 ( = 0.11. (11. (11.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). Oberhettinger [181]. We can see from Eq.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.108) This equation has the form of a confluent x$" + (6 . p.107) equation.e. W. (11. i.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. for E < 0 they > decrease exponentially. Magnus and F.v) vi(x) = 0.g.+if>x/2 vi(x).105) we obtain the equation for V[(x).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. T This equation is known as Whittaker's equation.e. (11. *The factor 47reo appears in SI units. i.104) W i t h these substitutions we can rewrite Eq.6 11.e.a * = 0. (11. Inserting the ansatz into Eq. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 . It remains to determine the function vi(x) for a complete determination of the solution.105) 11. see e.6. .x ) $ ' . (11. (11. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. We set therefore Vi X J.88.
i. . called dupZication formula. since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. meromorphic function. (11.1/2)!. it is convenient to use only the factorial notation. (4) r ( i ) = V5F. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. In general the function $ is an infinite series and behaves at x — oo like » x{l+l+u)ex_ Obviously such a series is useless in our case. The function T(z) is defined by the integral /co T(z) := / eHzldt.z(zl)\ = z\. (11. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z.216 CHAPTER 11. particularly pure mathematicians. (z)!(z .111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (2) r ( z ) r ( l . However.z) = ^f^. with many of these around in some calculation.J. the factorial for an arbitrary argument.106). (5)r(n + l ) = n ! . ^ ( 2 ^ ) ! = 2 2z z!(z .112) The gamma function is a nonvanishing analytic.(0_i)!(6 + n l)!n!' (11. (11. and hence the *Some writers. and for both cases. of its argument z with simple poles at z — —n and residues there given by (—l)"/n! as may be deduced from the property (2) above. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .e. more precisely.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function.! ) ! = imlFI' c a l led inversion or reflection formula.
1.» • » • » .. every n. This is achieved. = n = Z + l + n'. for z.101) to (11.1) + n = n(n ..113) This is a quantization condition. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin).v) T(l + lv) _ ~ K sin{7rQ/ . Thus the infinite series must break off somewhere.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ .114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/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. i. (11. and so to the number of different values of m.104)) 1 n =v = K.11.1) + n = n2. .e.e.V))T(v . d = 1. which are also described as its nodes. Eqs.. 2 . if the series terminates for certain values of u. The quantum number n' is called radial quantum number. which is thus seen to be a consequence of demanding the square integrability of the wave function. (11. . . .. in our case: a = 0. 1=0 (11. 1 .2. For every value En. With (cf. i. i.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. The degeneracy of the levels En is therefore of degree* n—1 1=0 n—1 V(2Z + l) = 2 V Z + n = 2^(n . (11.110) as follows: T(l + 1 + p . the orbital or azimuthal quantum number I can assume the values 0.I) r(vl p)sm{n(i/ I . One defines as the principal quantum number the number n.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. n' = 0 .e. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. . must be a polynomial.
4. / = 1. hence s).1 = 1 to n = 2. and from n. . relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. from n. I = 0. ... 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. . 2 . where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. 0.1 = 0 (called principal series.§ One also writes n2l+1l.1 = 1 (called diffuse series. One describes as Balmer series the final state n = 2. 11. 11.. (b) The spin of the electron was not taken into account. (c) The nucleus was treated as pointlike and not as something with structure.. hence d). The spectrum of the hydrogen atom thus has the form shown schematically in Fig.. I = 0 to n = 2.'.d. from an initial state n > 2.p. The following spectroscopic description of states is customary: nl —> ns. np.1 Hydrogen atom energy levels. nd.1.. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.l = 2 to n = 2.. and hence the fine structure of the energy levels is excluded.'. nf . . respectively.218 CHAPTER 11.1. . stand for I = where n denotes the principal quantum number and s.1 = 0 (called sharp series.\ I = 0 or 1. ' O n e describes as Lyman series the final state with n = 1. ne. from an initial state n > 2. hence p)..
. S.117) The function *Lk(z) is a polynomial of degree r. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. Vol. however.115) ' =0 (ri . (11. With the help of Eq. formula 7. for instance. 3 .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. For clarity and later use we denote the function used there by its normal form L^. 11.r f . . Oberhettinger [181]. • fc. k + l.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. W. • ( .2 (11. Ryzhik [122]. • (11.414(3). Appendix B. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). in fact. We distinguish between different definitions in use. . so that one always has to check the definition. p. (11. Messiah [195]. We mention here three different definitions used in the literature. p.114b) yields the binding energy of the states. 844. Unfortunately the definition mostly used in mathematical literature differs. Gradshteyn and I. see. (u. 84.x).11.z) = (r + k)\ Lkr(z).= t .6. I. several different definitions are in use. it is given by^ *Lkr(z) = ^ ^ f 1 *ir.118a) where Lk. (11.1.2. l . Apart from factors.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'.{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z).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.p)\(2l + 1 + p)\ The spectrum (11.r = 0 . M.2l { + 2. the polynomial contained in y\ of Eq.115) is effectively the associated Laguerre polynomial normally written Lk(z). Each can be recognized by reference to the generating function. "1'f„+11" P] „. a + 2:*) = t v . 2 . Unfortunately. . and I. Magnus and F. .
Example 11. N.122) "See A. Ll(z) = 33z + ±z2. It is said that this is the definition preferred in applied mathematics.. L\(z) = 2z. . **I.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. Vol. N. p.3): e2t/(lt) (1 . r r Z r=0 v tr k k (11. I. Sneddon [255].„\*Kk. . ^ F o r this definition and the following equations we refer to I.. and Ll(z) = l. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1.„M„ z"*L.** The associated Laguerre polynomial here. pp.d z—^z + {k + l~z)—+r Lkr{z) = 0. There at some points n + 1 is misprinted n + 1.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. is defined by^ dk (—'\\k(r^\2 (11.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk.121) In Example 11. (11.2.220 CHAPTER 11. Messiah [195]. which we write ^Lk. Appendix B. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. L1(z) = l . L2(z) = l2z + z2.164. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . . 163.1.z . Sneddon [255]. 162 . 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.
124a) (i*) pi The orthonormality condition is (cf.4z + .. (11.8): (11.121).107) implies k = 11 + 1 and r = n + I..120).z . ^L\(z) = .119): ty"(t) + (a + 1 . (11. and *L\(z) = . N d rk 0.4 + 2z. Then.124c) Example 11.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). Jfpd>2s{p))eptdp. '"''Comparison with Eq. *L\(z) = . The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp. and partial integration.118a) and (11.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n . ^L2(z) = 2 .1 . i L1(z) = l . and obtain with this the normalization condition (11. (11..3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.124b) Comparing Eqs.123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. Example 11.120)) it)k& zt/{lt) ~*P p=k fc (11. Solution: We consider the following equation of the form of Eq. The limits will be determined later. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.11. (11. .119) derive the generating function of associated Laguerre polynomials (11.1 8 + 18z 3z2. (11. (11..
n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). j p = (P 0 "V?. (2) from —1 to oo. The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. since for n > b : (b — n)\ —• oo. 2TT_ Jq=0qb+l(lq)a+l 1 . Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} ./ ( l — <?) —— = G(t). s bK> (lq)°+i (11.q) and P~ t ( .( a + l ) p s ( p ) + 6s(p) dp 0.] from the above equation: C(p) := p(p + l)s(p)ept = 0. p and one has (as polynomial of degree 6. which removes [. this is the rejected irregular solution. .222 CHAPTER 11. (3) around p = — 1 if a + b is a negative integer.p + 1 = 1/(1 .) 2 J J p 6 + i ( l . (11. and (4) around p = 0 if b > 0 and integral. We demand (to be considered later for integration contours) the following condition." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. Case (2) implies the exponentially increasing solution ~ e + t .p ) o + i 7 t = 0 . Thus in the case of Eq. It follows that 1 d i I _i. 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. the only possibility is the last. (11. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo.. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* . Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. 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.. Then dp = dq/(lq)2.119) with a and b as there. agreeing with Eq. Hence the expression in square brackets vanishes.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.a = 21 + 2.^ .118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. case (3) similarly. with requirement of a finite solution. t > 0.p ) « + i q c + i ( i .
11. Then Nni is a dimensionless normalization constant and (note that we use the associated Laguerre polynomials as in the book of Sneddon [255]) Fnl(x) = xlex^LZ. and 0 if cj=b. Eq. 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. (11.—i. where Fnl(x) = yi(x) (11. Jo (11.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11. is introduced for dimensional reasons. (11.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .112). The wave function of the state (n. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . .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. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.125a) and the factor a~ 3 ' 2 with a = Bohr radius. so that I = (c + a)! 1 f .?).127) .(x).6. I. 11.111). Clearly only the second form is sensible.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En.
a (r)is = ~a.2.3fc + 2).7.2Z . The Coulomb Interaction One finds in particular (see also Example 11. 11.129) n2a' Analogously we have roo j = p2 I r n \ 2J ° J0 'V^yr nl _ na J^° xF^(x)x2dx _ na f^SJn+K 3 ) \na) na {6(n + Q2 .(l) n+l. + .n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .3(2Z + 1) + 2} 2 {2(n + 0 .8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .1 + 1) (11. (11.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).v (6n2  6nk + A:2 + 6n . For the ground state (n = 1. = 0) we have ) r /u =. This behaviour is indicated in Fig.. (11.224 CHAPTER 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 + /„. thus on the average the electron is the farther away from the proton or nucleus the larger n is.! ] + (21 + l ) 2 .130) We see that the mean value of r is the larger the larger n is.(2Z + 1) + 1} = ^[Zn21(1 + 1)].131) . and for the explicit derivation of the second case below Example 11. 2 (11.6(n + Z)[(2Z + ! ) .
S. Problem 5.132) (11. / ! J — .( n . 173/174. i.133) for large r. or (r ) ~ n a 2 2 ^ [ 3 n 2 . This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o.J n 1s 2s 3s *r Fig.2 I 3/2 Kl r 225 \2 1 R„.134) *These are taken from I. v ^ V2n + 1 (11. we have and tp becomes particularly simple. Sneddon [255].11.j .6 The Discrete Spectrum of the Coulomb Potential r.20.(r) = x W 2 n + 1 =. pp. 11. In the case in which I assumes its maximal value n — 1..2 Hydrogen atom wave functions.^ We obtain therefore for the quadratic deviation (r) .e. with n +1 replaced by n and 21 + 1 by k. = l a nl n N'nll rFnl.l ) n ] =  [ 2 n 2 + n] an[ n + (11. \f{r ) ~ n a. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .2 1 A r = V (^2) . In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton .
i. the states Ytl are predominantly concentrated in the xyplane). determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". 6.4: 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).135) This expression becomes very small for large values of n. Example 11. However. for these values of I it is not distributed with uniform probability over the spherical surface. (c) The nucleus was treated as pointlike and not as something with structure. in which periods are obtained from the BohrSommerfeldWilson quantization condition. = ngh. . the electron remains practically localized within a spherical surface of radius (r).— > where ampler is the length of the semimajor axis of the elliptic orbit. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded.3. pJ = —— and j> Pidqi = nth. Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. \/v vT 7 v^ See also Example 14. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. p2 = p2 + — pg • eV. and separating this in cylindrical coordinates r. but rather like classically in a plane (i. Thus. Since 6 is a cyclic coordinate. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.e. This means. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). we consider in Example 11. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler .e. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. In order to obtain an impression of relativistic corrections in a simple context. i = 0.r. we have pg = const.4 the relativistic equation in cylindrical coordinates. The Coulomb Interaction or TT = ^ = (11. r.226 CHAPTER 11.6.
.6.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. 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 ip = 0. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.136a) (11. 2 2 Therefore the energy is no longer a function of n = nr + ng. ( v c + i ) .2. it is possible to separate the Schrodinger equation in other coordinates. <p = <p. 11. as also indicated in Fig. r\ = r ( l — cos#) = r — z. y=y/%qsanp.(p . Whenever there is degeneracy.V .9. (11.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. where the constants are identified by comparison. The deeper reason for this are symmetry properties of the problem. ip are the spherical coordinates. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . where r = ^/x1 + y2 + z2. z=^rj).11.3. which are related to Cartesian coordinates in the following way. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».T). y — 0: V ^ c o s tp. Schiff [243]. 11. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'.136b) £ = r ( l + cos#) = r + z. (p). the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. which correspond to a new choice of polar axis.* See L. p. . 86.
228 CHAPTER 11. 1 . and r\ = const.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. a separation constant. . The Coulomb Interaction Fig.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. We rewrite the first of these equations with the help of the relation M (w ^  m + e) . (11. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike.3 Parabolas £ = const. with zaxis as axis of rotation. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . for uniqueness of the wave function.136b) the Schrodinger equation of relative motion (11. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o.22) becomes — as we show in Example 11. . 2 . . = 0. 11. so t h a t the variables can be separated. In terms of the coordinates (11. there is really only one basic equation to solve here.
. . .106) and (11.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11.105)) Hence we write I = (m — l ) / 2 . 2 .e. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . . . .± ( m + 1) = 0 .o. We leave the calculation of the degree of degeneracy to Example 11. Adding now / . E x a m p l e 11. p = /?£ and cr = /^r?.125b) (dividing by p 1 / 2 ) £ oc e .' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a). we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". . i.6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. ds2 = dx2 + dy2 + dz2. In the spherical polar case we had (cf. . .i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . n 2 = n" . Eq. 2 . (11. 1 . . . (11.6. To obtain a closer analogy with the separation in the spherical polar case. from Cartesian to parabolic coordinates. " . as well as the Laplacian A = V and the Schrodinger equation. m0Z1Z2e2 ^ 2 = n (say). T h e n by comparison with the spherical polar separation the answer is as in Eqs.114b). 1 .11.
dx = d{ y £77 cos ip) = . Eq. . We consider first the case m = 0. we have to find the number of values of the quantum numbers n\.tx2 .e. W. We use the method of selector variables. 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.[df + dr]2 . 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. see Eq. N2 where g^dS. (11.1) ~ .e. last chapter.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 .2)d} = (n. . Similarly we proceed with dx and dy. i.l)a + (n . " . etc.1 in this expansion is seen to be n.. all cross terms cancel in the sum of 7 the squares.136a). Thus dz2 = . Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). </. (11. In the case m ^ O w e have n\ + 77. are elements of length. + l 4 {^^)drl f i + v \ .l ) ( n .2 = n — 1.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. as t h e two signs of t h e exponential indicate. Solution: For the wave functions ip = £(£)J\f(r])&(ip). . 712 and m which give n. n 2 (1UJ)2 The coefficient of a . The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. / — cos ipdl. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. and so dz = (d£ — drj)/2.114c). z = (£ — rf)/2.g<p 9 \ . when m = 0 = 77.E E n l „ . The Coulomb Interaction Solution: We begin with z of Eq. 2 _ = (9id02 + (gvdvy + /„ . 2 + ^dV . (11. 2 2V? V^ One now performs the square of this and obtains similarly dy2.2)d/2.2) and summing over m. > .2d? dr)}. 0 there are two. {g^Y. Here we have (see preceding text) 711+77. Beware! For m = 0 there is only one mstate. Magnus and F. Oberhettinger [181].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. i. Thus we want to obtain the coefficient of wn~1 in £2 .e. and attach a factor to to each term introduced in the wave function. Thus. n the double sum ]T]2 is TI — m. = e±im^e("+^'2{par'2L^{p)L^ (a). We then sum all possibilities and select the coefficient of ojn~1 in the result. (n _ !) n = n2 §See e. \— < / — cos ipdr] — \/£r] sin ipdtp.e. / J„^2 . for m / .g. i. for instance..230 CHAPTER 11.2 .136a).* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . Observing that the highest possible value of m is n — 1 (i.
9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in. = j (2ni+m+l).™(p)] 2 By contour integration (see Examples 11.11.128) if one makes there the substitutions n —> n\ (or 712) + m. with Z\ —> Ze. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function).127) and setting n = r + k. with Eq. With the help of the wave functions we derived above.r. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . .•* n (l) = / Jo It follows that . with RayleighSchrodinger perturbation theory).a). 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. eF{z). with the connection » (11. In fact.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz.124c). . fc — m. these expressions agree with those of Eq. where F is the electric field. {p — a) — = dxe~*xkvLkn (z)] 2 .(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. (11. . with z = r cos 9. we can relate the expressions Iq to those of Eq. 8. which will then give the change in energy of that state to the first order (i. Sec. and determine the average value of v for a given state.—r~n = 3(mn 2 ). I. if we expressed the perturbation in spherical polar coordinates. Z2 —> e.) = ^eF(p . L ni\ ny.e.6 The Discrete Spectrum of the Coulomb Potential Example 11. However.g. . s 0/ . (11..1 (cf. .drid<fi oc (p + o)dpda.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—.e.e. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. we would have to use degenerate perturbation theory and obtain the same result.128).6). Thus e.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.8 and 11. Ikr+k. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. 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.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. i. i. (11.
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 .( 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)!. In Example 11.p ) r + 2 [(lp)(lg)]"l+1(lpg).x ( n + m)! ( n .^~T^——^ ) = coefficient of (pn .a. tt„:_ (1 . _ „ . 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 .l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . the integral / is the coefficient of pnqn in the expression _ / . The Coulomb Interaction E x a m p l e 11.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) .3: /"OO I := Jo et[L^{t)]2tm+1dt.9 )« Solution: Using the generating function for both associated Laguerre polynomials. .g )(l.3). (n + r .b.1)! n\ This verifies the expression I™t in Example 11.124d) of the alternative integral representation obtained in Example 11. dt = ds/a.232 CHAPTER 11.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l . (11.9 we consider integrals with an arbitrary power of t.n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 .1 ) ! (n .pK — — —L . (11. s = at..9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq. E x a m p l e 11.3 by one power of t.p ) ( l .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.8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11. the integral arising here becomes with Eq.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 .128).9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.p ) ( l . and is a special case of Example 11.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11. c positive integers.
c + b — 2r = l (i.^ ( l .(*) from the coefficient of a1: />00 233 1(a) = / Jo dte. we have 2r = 6 + c = even. Using the contour integral representation (11.e. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . c + b even) and/or j / = 0.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£. (lp)(lq) + a(l q)(l . . where (a+**£)! c j n Jo = .(i) = i ! ( .w(r) = and / a ) = r\(P ~ r .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 «.l ) * x coefficient of a* in the expansion of 7 ( a ) . 7£.r ) ! a ' . so that = (q+fc±£) . c + b odd). we have to have y + (c + b — 2r) = 1. i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .11.124d) of an associated Laguerre polynomial.p) .e. c + 6 — 2r = 0 (i.™( r > (1 + a)P r £^0y^w Jc+6_2r = .p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . ° ^ + a ( l .?(t)t a e.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 .a t [Lg(i). i.Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.q)(l .L. (a) j / = 1. = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. .p ) ] .e. i.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 (11.1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.f ].a _ 1 = [1 + a ( l . — (cr)!r!(6r)! Kv.111) of the gamma or factorial function.e. In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.
Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.7 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. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. b\ V cited in Example 11. These results are seen to agree with the expressions 1^. . + (!±<)«. In view of the slow falloff of the potential at infinity. Since such a rigorous treatment is beyond the scope of this text. . cannot really — and this means in a rigorous sense — be applied in this case.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. the standard principles of nonrelativistic scattering theory. however parallel to the Coulomb case. is the work of J. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions.^)(.7. 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. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. Taylor [267]. also with regard to earlier literature.)!' f^)(.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11.1%^ 11.~* Hence (recall that 7£ c (i) = i!(—l) l x coefficient of a% in / ( a ) ) (a+^)!(/3a) r r in (£f^)!(^)!(^±£)! t J = ( a + b)\((3a) > 6! _ (a + ft)! = [a + lb + l ) a . R. t We have the Schrodinger equation of relative motion of particles with charges Z\e. ^A very readable source to consult.
b. .145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr).108) and (11. A = const.138) becomes 235 ^A + k2 .136b). (11.z).7 The Continuous Spectrum of the Coulomb Potential Then Eq.142) This equation is of the form of a hypergeometric equation.e.144) *(a.140) and using (11.143) and v = ikrj. for instance. „ az 1. Magnus and F. (11..+ ^  + .z) + W2(a. (11. for r — oo) • ipr = Aeikzf(r ..z).^ W ) = 0.146) See.87 under Kummer functions.e.109) we can therefore write the solution ipr = Aeikz$(i>y. what is the > behaviour of 3>(a.b. (11.z) = Wi(a. (11.z)).. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). . Our question is now: How does ipr behave for r — oo. i. According to Eqs. Oberhettinger [181]. that we encountered previously. a(a +1) z2 ] (11. z = rcos<9. where ^.11. W.140) The simple Coulomb potential permits particularly simple solutions. 86 . z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.141) Inserting this expression into Eq. (11. pp.. (11. . (11.M) = l + . b. with the asymptotic expansions (11.b. i. v = ik(r .
with 2. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(.a) (z) na)"~ r(la) T(ba) n\ ' (11. Without t h e logarithmic phase factors. The Coulomb Interaction W2(a. = r cos 0.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.154) .(11.z) + la)T(n + b.e. (11.l.kz+l^ktrz)) C .i 7 ) (11.ik(rz))} =^ + ^d.151) with Eq.e.152) Since z = r c o s # .^l*7 l + O (11.152) 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) .153) The asymptotic solution represents t h e stationary scattering state (E = const.z)T(l . if tbr were „ikr ibr ex el"z + f{6) r (11.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].148) .l. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr. (11. a comparison of Eq.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n.b.236 n\Zzab CHAPTER ^T(n '£ fo 11. i. (11.149) (11.7 ln*(rz)) k(r .150) ^(rz)]1"^ 1 + 0 T(«7) i. J\") ci(kr~/\n2kr) (11. > A = Aeikz[W1(i1.
The quantity /(#) 2 would then be a measure for the scattering in the direction 9. the effective range of the Coulomb potential is infinite.155) Since the gradient is proportional to the configuration space representation of the momentum. As a consequence of the masslessness of the photon. The problem of the logarithmic phase.Q hk mo v0. we see that this expression is the exact analogue of a current density in classical considerations. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. 2im.7. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. (11.7 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. 11..1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. that we encounter here is a characteristic of the Coulomb potential. also called phase anomaly. 11. (11.156) . do dQ K 12 K Fig.4 Variation of the observed differential cross section with 0.n .11. z — — oo. the current density is * h . In other words. We define as current density the quantity h J := 2imo [^(v^)v(v^n. For the incoming wave ipi = exp(ikz).
152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.e. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). for 0 —> 0. the square 7 2 ). is called scattering amplitude. (11.154) we chose the prefactor of the incoming wave in (11.157) 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.153): 4fc2 s i r  The expression /(#). That this formula retains its validity here in quantum mechanics is something of a coincidence. i.e. thereby screening itself off from the surroundings. In the context of quantum electrodynamics one refers to "vacuum polarization" and relates this to the idea that a charged particle like the electron polarizes the otherwise neutral vacuum by attracting virtual charges of opposite polarity and repelling virtual charges of the same polarity. We recognize the result (11. in the current this would in any case contribute only something of order 1/r. (11. Correspondingly the other part of (11. We see that a depends only on the modulus of the potential (i. which yields the scattering cross section.159) as the Rutherford formula which can also be derived purely classically. This implies that scattering takes place for the attractive as well as the repulsive potential. In Eq.152) as 1. 'screened ZlZ<2e2 cr/r0 ° . since in actual fact the Coulomb potential occurs only in a screened form. In reality this divergence does not occur. We make a few more important observations. We obtain the total scattering cross section by integration: „2 — /£<" = £/777*V sin 4 « 1 »«°) We observe that this expression diverges in the forward direction. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V .238 CHAPTER 11.
and which provides the uncertainties arising in quantum mechanics.165) ^ ( x ) = £ fe Vk(x). whose maximum moves with the particle velocity. 2mo V + F(x) k (11.i i f (''°>/Vk(x. 2 (11162) where Vk(x) = ^ ( x .^ ^ % ( x ) . i. t 0 ) . 11.e. t) = U(t. of continuum wave functions. / dk 7^340(k)4(xxo. t 0 )^k(x. the wave packet in the absence of a scattering potential. (11163) (11.4.t).11.166) Let the incoming free wave packet at time t > t$.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. of the equation r fi. 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. (11.yJ ^ j A k o W 727)3/2 e 0. The particle velocity v is v = — .161) where Ak0(k) ~ A(k — ko) and ^ k (x.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. which in the case of free motion (zero potential) are simply plane waves.t). t 0 ). m0 m0 and V.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.e. v0 = .2 fr2 .t) = e . t0) = e.e.164) Ek~Eko + {kk0)v0h. 11. The wave packet is a superposition of waves ip(x. (11.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . i. (k). _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . and if H is timeindependent one has the stationary wave function ^(x. be given by ^ f dk AL. i.
^k(xxo) ^k(x) ~ = e. Eq.167) we have ^k0(xx0.165) ko • (k .167) by replacing the plane wave exp(zkx) by this scattered wave.k0) ~ k0 + and the expansion of E^ in Eq.e.166). k = kez.<p) + Ol Akr l (11. The Coulomb Interaction In Example 11.171) The phase of fk(0. (11.x 0 ). (11.172) h h k0 {rv0(tt0)} (11.xo.154)) that for x — oo. With Eqs. (2TT)3/2 eik* + —fv(P.e.10 we show that the solution V'k(x) of Eq.169) The wave packet in the presence of the potential is obtained from the free form (11.173) . (11. ip) could be carried along but we ignore it.161) and (11.163) we obtain </>ko(x .x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11.ip) is the scattering amplitude with respect to the incoming plane wave.^ ( * .t o ) / V k ( x .x °Vk(x).ik . From Eqs. Then with k = ^(ko + k . (11. i.k 0 ) 2 ~ Jk20 + 2k0 • (k .240 CHAPTER 11.t) = J ^ A k o ( k ) e .iB *(**°)/ R °'tj + 1 pikr e * . can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf. i.169) and (11.170) A ko (k) e . (11. the stationary scattering wave. this solution > has the following asymptotic behaviour in which f(6.i) dk (2TT (11.ko) k0 (11.
and determine the Green's function G(r.r ' ) . T h e scattering amplitude is t h e same as t h a t given by Eq. . t) f. E x a m p l e 11. r ' ) = 47r<5(r .10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. U(T) = ^ V ( r ) .x 0 . Solution: We define the Green's function G ( r . *.*0ko (x .8 Scattering of a Wave Packet so that with Eqs.r')U(r')ip{r')dr' /4T7 (as we saw earlier) . however.167). 11. Adding a suitably normalized solution of the free equation. (A + fc2)t/. (11.r')V(rW).174) r /k0(^.(«)/ ipko(x . the solution can be written iMr) : (2TT)3/2 ^dr'G(r. t) . i. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^. '.e.11.to).^y /k0^. (11.z Fig.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r.5.y)eiko'Xoe'[fcor£.5 Scattering of a wave packet with scattering centre O. Here ^ (x' — xo.r')l/(r>fc(r').154). with the energy of the maximum momentum of t h e wave packet.171). r ' ) by the equation (A + fc2)G(r. (11.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.to) has t h e form of t h e wave packet in radial direction as indicated in Fig.fco(tto)/ft]^0o)(x/xo. A particular solution of the Schrodinger equation is — J G(r.r').x 0 . 11.
which is such that the contribution along the infinite semicircle vanishes. dk' ik'*r 2TT2 fc'2 . we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k).fc ) oo dfc' 1 d nik r 7rr dr / .Rek' kfc/2K Fig. 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. Hence we write G(r)G+(r). . Imk . one obtains another Green's function given by G_(r)= r eikr. These are the socalled outgoing and ingoing Green's functions.=k+ie/2k 1 d f.6 The contour C+ in the upper half of the k'p\ane. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . The Coulomb Interaction We find a set of Green's functions G(r.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. 11.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . 11. e > 0 small. Thus we consider the contour integral taken around the contour C+.6. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . displaced slightly away from the real axis as indicated in Fig./2 dk'd(—cos 6)dip iki.242 CHAPTER 11.
2Z + 2. If we proceed from the radial Schrodinger equation. (11. Before we introduce the general form of the socalled partial wave expansion. i.21 + 2.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). (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. 2ikr)] [Wi{l + 1+11. b. (11. under the stated conditions 2/07 = Z\Zie2 = —Mo. .176) " 0—A ~ (I + 1 + iy)<t>i = 0.178) (11.0 + W2(l + 1 + iy. Then yW (11.11.2l + 2. Then £^<fc + (2l + 2 £ = ~2ikr. Eq. b. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. z) + W2(a.e.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. z). z) = W^a.9 Scattering Phase and Partial Waves 243 11. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems.z2 27fc 1(1 + 1)' yi = 0. (11.g.140) (previously we solved this only for the bound state problem) § .179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij. e. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. 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. whose continuation to infinity is again given by (cf.£). 21 + 2. i. Thus we proceed as above. (11. Eq. b. where Mo is a parameter in Chapter 16.e.
244 CHAPTER 11. Then e2i5< = Reie Re~ J2i9 9 = Si. (2kryr{i" V 7 .146) and (11.«7) „ .i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 . T(2l + 2) (2fcr)*T« (Alliy ikr . The Coulomb Interaction Using Eqs. +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.183) r(Z + 1 + 17) := Reie.l e*"" + T(l + 1 + H ) T(l + 1 .180) ^ ' (11.' . .147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 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( _ .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 . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.181) so that eldlT{l + 1 .ii)' (11.ry) = e"^r(Z + 1 + ry). (11.
(11. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. g = e«* = rSi + 1+ *V*ri (11.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). where the continuum of the spectrum starts. One defines as regular Coulomb wave Fi(j. i.n ' . indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.187) contains almost the entire physical information with regard to the Coulomb potential. i.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase.1.e.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .2..188) in the form y/E. In the present quantum mechanical considerations I is. For instance the poles of S are given by Z + l + z7 = . with Eq.e. considering I —• an(E). i.. 245 (11. kr) the regular solution with the following asymptotic behaviour: r 0 (11.e._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). . (11.139) n' = 0. That the energy E appears in Eq. considering I as a function of E extended into the complex plane. as we saw. However.114a) reveals).. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element. i.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function).e. (11. a positive integer. .11.
190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. For the Coulomb potential a typical such trajectory is indicated in Fig. 11.7. we obtain trajectories which are known as Regge trajectories.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." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections.246 CHAPTER 11. In these cases. (11. 11.189) where as before the incoming wave in the direction of z is ikz ipin = e (11. and thus in general. (11. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. we obtain the partial wave expansion of the scattering amplitude as follows. Singh [252]. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. The Coulomb Interaction and plotting these for different values of n.7 A Regge trajectory of the Coulomb potential. 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.
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. The potential disturbs this phase relationship (by delay or absorption or redirection). Oberhettinger [181].192) Thus the incoming and outgoing spherical waves have a definite relative phase.11.196) "See e. 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 ' . When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg . 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. W.eikr\P^cos9).195) with the scattering matrix element S(l.22.g. (11.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. 21 . Magnus and F. pp.r ^ V ^ Wostf).^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. .i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. (11. The mathematical expression can be found in Tables of Special Functions (there cf.i f c r . 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„ ' .k)=me2iS^k\ in which r\i^\ indicates absorption. r e.~ J > Z + l)Lil*e~ikr .^2l + l^S^k) lik (11. (11.
This page is intentionally left blank .
one of the main and now wellknown methods to perform such calculations. These newer methods will be presented in detail in later chapters.e. Such a vital quantum mechanical phenomenon is socalled tunneling. therefore. For various general aspects and applications of tunneling we refer to other literature. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. however. in texts on quantum mechanics is rare.* *See e. states of a finite lifetime. which a particle in classical mechanics would never be able to intrude. to explore first much simpler models in order to acquire an impression of the type of results to be expected.g. It is not surprising. Similarly difficult is. Razavy [236]. in general. M. i. namely the instanton (or more generally pseudoparticle) method. was developed only in the last two to three decades. the computation of finite lifetimes of a quantum mechanical state. It is very instructive.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.e. can be found in any traditional text on quantum mechanics. as — for instance — for trigonometric or double well potentials. i. that the explicit derivation of such quantities in nontrivial contexts. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. that quantum mechanically a particle may have a small probability of being in a spatial domain. 249 .Chapter 12 Q u a n t u m Mechanical Tunneling 12. In fact. which in one form or another.
ihr(x. r (12.t) that the one particle concerned is somewhere in space at time t. According to Eq. so that p = T/>2 is the probability density for a spatial measurement at time t. i. t) was developed by Born in 1926. Quantum Mechanical Tunneling 12.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.t)\2 =1 (12.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x.e.250 CHAPTER 12. M^ Then dt + Vj 1=0.3) we can obtain the continuity equation which describes the conservation of probability in analogy to the case of the conservation of charge in electrodynamics.t) (12.1) as the probability described by ijj(x. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.t).4) (12. . (12..t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x.t) = HiP(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.
29 . Tarrach [119].12. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). where „ mpVo V(x) =+V0S(x). 2m E 2 V0 > 0. However.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. pp.33. the problem is analogous to that of the scattering process from a potential in a onedimensional case.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. (12. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. 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. Schiff [243]. 12. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. the equation can be integrated at every point. Nonetheless. even for discontinuous potentials. although in general small.6) . we can do this also here. The Schrodinger equation for this case is* ip" + [A. Gosdzinsky and R.2 . which is the case we consider here. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. In much the same way that we define there reflection and transmission coefficients. Thus it is possible that the probability for the particle to be in such a domain is nonzero. This effect is known as "tunnelincp. If ^>(x) and V^>(x) are known at every point x. higher dimensional delta potentials do not have properties like that in one dimension. 'Maybe the reader dislikes being confronted again with the onedimensional case. This case plays a special role. as shown by P. 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. Such a region is for instance the domain where V is larger than the total energy E. Delta potentials in more than one dimension do not permit bound states and scattering. as we mentioned above. If there is no absorption. 1 0 ~x o a= f See L.2a6(x)]il> = 0. t For the singular delta function potential a modification of these demands is necessary. From this follows that both p and j have to be finite and continuous everywhere. 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. ~hr> k = n^ (12 7)  .
[<//]0+ . 2 (12. i.e 1 + B = C.= 2a^(0).9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo. i.: (i2iQ) Ce for x > 0.13) ~A R= for x > °(12. .(ik . (12.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—. a — ik £Qr < { eikx a eikx •t" . Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see.[^']o.12a) (12. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .ik(C . i. (12. e Q (12.ikC = i.e.' : It follows that ik . (12.252 CHAPTER 12. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).ikB) = 2aC or 2a(l + B).11a) ikC .11b) 2aC.1) . 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. The continuity conditions at x = 0 are therefore for V : and according to Eq. except in domains of the potential V.2 a J 5{x)4>(x)dx = 0. a 2 — a 2 a — ik a2 + k2 ' .e.14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.8) (12. (12. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x .e.9) for ij.
a l s l. f Ce~KX c. K = a.e. ( 1 2 . = . as coefficient of the outgoing wave. when normalized to 1.g. .(K)C = . and C follows from the normalization: OO A.17) da#(s) 2 = 1.K C .e.) and V>'(0+) . We write the solution ^ so that ^(0+) = ^ ( 0 . . poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. See e.e «x = J \ CeKX for for x > 0. (12. Omnes and M.2 o ^ ( 0 ) . / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0.K 2 < 0. Thus ^ = Ce.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R. Proissart [223] or other books on scattering theory like R. i. i. 2m 0 (12.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.i 8 ) .15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. Newton [219]. at k2 = —a2.2 = . or h2 E = —a2.19) Thus the delta function potential supports exactly one bound state. z<0. 'Note that the physical pole has negative imaginary part. possesses simple poles at a ± ik = 0. R.ip'(0) = . oo (12. the quantity T.12.2 a C .
The latter contains also specific applications in nuclear physics. 55. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0. 12. We want to consider the case of stationary states for E > 0. Fermi [92]. 12. p.a].1 The square well potential. V(x) V=E a v=y Fig.254 CHAPTER 12.1. x€[a. \x\ < a. (12. (12. or in the lectures of E. Schwabl [246]. x G [a.21) The solution of the Schrodinger equation then consists of the following parts: Aeikx + Be tp{x) = { Ce~ iKX lkx iKX with with with xe[—oo. —a]. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.20) for for \x\ > a. Quantum Mechanical Tunneling 12. . p. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. oo].22) + De Feikx + Ge~ikx *We follow here largely F.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. 58.
12.ikGe~ika.ikBeika = inCeiKa + DiKe~iKa. Fikeika .27) v ' Similarly Eq. (12.26) K^gifoa—i/ca fc. (12. (12.28) 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 \ . ( V C n vD ). 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .24) The first set of equations can be rewritten as / ( p—ika / p—ika pika pika U pika A \ \ / \ / 1 pina Kpina p—ina Kp—ina \ \f ° ){D ) \ B ) ° or A N \ Bt i. . M(a) . ).25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. (12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.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. fc Af(a) = ..f )e" t\ K\0—tKa—ika .iKa ? ikAe~ika and .24) yields F ^ G.e. I (12. (12.
a]. x € [a.a .34) into Eq. xe[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.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. i.a ] .35) with with with x G [oo.4T(£)e ifci.37) .e.e sin 2na IJlika e B = r)F sm2na. = < Ce~lKX + DeiKX . (12.22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.e.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.32) [cos2«. we obtain A = F( cos 2na — . the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 . (12.31) we have A AT(a) = )MFo B ^r)sin2na . i. (12.(1 + e 2 ) sin2 2na 1 . Quantum Mechanical Tunneling Hence with the complex quantities IK (12.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. \R\2 = T(E)^sm2Ka (12. (12.256 CHAPTER 12.
2m0\E\ h? ' (12.e. p. Li ( — H— I sin 2na = 0. which is satisfied if either / = g or / = — g'1. (12.e. i. Dwight [81].12.44) .34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12.e. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. (12.4 Scattering from a Potential Well On the other hand according to Eq. 2\k K. . for VQ < E < 0. (12.42) These equations have no solution for K. 83.e.41) i. However.31): 1\T(E)\2 257 1 .39) 2 i. they have solutions.43) Thus we set k — —i ' H .38) as expected. in the domain of A imaginary and K real. nor when both parameters are purely imaginary. (12. (12.40) we obtain cot(na) — tan(/«j) — i K IK . cot(fca) = i— or tan(fta) = K K . and k real. an equation of the type f~1f = g19.35) and using Eq.12. (12. i. (12. B. formula 406. We see in particular from Eq. (12.
. ± l . At their positions cos(2/«z)£. (12. 2m0\E\ Ktan(«a) = — y——^— . 2Ka = nn. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — .47) These states in the continuum are called resonances.20). n = 0 . t2 2 (12.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . and is itself infinite there.fl = ( .g. Schwabl [246].48) We expand T{E) around these energy values. (12. t&n(2Ka)ER = 0. Then according to Eq. . 2m0\E\ and Kcot(Ka) = y—^—L. (12. (12. (12.45) These are the equations that one obtains from the connection relations for the case of Eq.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. Quantum Mechanical Tunneling . „ . k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12.' We observe that T(E) has poles at those values of E which supply binding energies. around the values given by Eq.258 and obtain the equations CHAPTER 12.l .50) See e. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave.34). i. Eq. = 2/rn (EER). T(E)e2ika = — r .65. The amplitude of the transmitted wave assumes its maximum value where (cf.46). pp. (12. . (12.l ) n . (12.e. i. from Eq.c o .49) Taylor expansion of the denominator in powers of (E — ER) yields . A = 0.e. 64 . We now return to the scattering states. F. (12. The remaining wave part becomes exponentially > decreasing. . "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1. ± 2 . Also.42) which then leads to bound states with associated even and odd wave functions.35)) for E > 0 sin2Ka = 0. i. This can be understood.e.
2a<x<2a. = 0. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig.5 Degenerate Potentials and Tunneling In the following we consider a potential which is very similar to a periodic potential and can serve to illustrate a number of aspects of the process of tunneling in quantum mechanics.2. 12. This result is physically very plausible. T(E) has poles at E= ERi: (12. We restrict ourselves to the period — la < x < 2a.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. A potential of this type is known as a PenneyKronig potential.e. The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. 12.^ l / . Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. 12.52) Here ER is the energy of the resonance and 2/F the lifetime of the state. a 3 a. These are precisely the resonance states.' ' 5a \V / Fig. s \ V0 /• ~\ 'X .12. . V(x) x=2a x= =2a .2 The PenneyKronig potential.51) i. .' ''' 5a 3a \ a ~~~. V(x) = 9(x + a)VQ6(ax).
. and we choose (12.2 = W h 2 _„2 = 2mo(E . Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.e.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.. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. 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.53) *w«w^( « j)'*^{i i ^ . Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo.2. We set . (12. with V as parity operator. In view of the periodicity we have ^(x) = ^(x + 4a). HV^(x) = H. and hence for = Eip(x). .. (12.2 < a .e. i. E^En = ^ ^ t n = 0..55) The Hamilton operator of the problem. i.a ) = 0 = ^(a). x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. the differential operator is even in x. = EVip(x).1. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). .e There is the additional boundary condition (for Vo = oo) ^ ( .Vo) n..260 CHAPTER 12.
The central region of the solution is the classically forbidden domain. 12. V'(O) = const.2a]. are ^_(0) = 0. We expect.a]. At x = ± a we now have to connect ip and ip' of neighbouring domains. xe[a. This wave function is no longer restricted to the interior of a box. (12. 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. We > obtain even and odd functions ip± by setting ^ = ±1.. (12. = Ka ±Asin(k2a). of course.12. so that (W meaning Wronskian) W[tl>+. VV(°) = °.2a]. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ .il>]?0.60) The boundary conditions that ip± have to satisfy.58) D sin k(3a — x) with xE[a. (12. V+(0) = const. which.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] .61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].63) = KBe~ F = .2a. ±Asink(3a — x) with x€[a. We therefore have different boundary conditions. (12.a].5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a.2). — a].59) Asink(x + 3a) with x & [~2a. — a].Be Ka (12. This means that i>±(x) = <  = ±1. (12.. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. that these are such that for VQ — oo they approach those of the previous case. ^kAcos{k2a). however. BeKX±Be~KX with xe[a. we shall not exploit here.
12.68) .nir + 0[{2ak . Quantum Mechanical Tunneling Fig. say. i.4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . for VQ — oo. (12.— (1 ± 2e~2A + O (—). However. i.65) and we expand tan(2afc) about nir.64) we obtain ~[(2ak k . is the one belonging to the even solution. (12.67) The value of k belonging to the upper sign. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. (12. i. say A.even.3 Level degeneracy removed by tunneling. (12.nir)2}} = 1 ± 2e~2Ka. It is plausible therefore to use appropriate expansions.e.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt.+ • • • 1! cos^(n7r) 2ak .mr)2}. we set tanhx = l . (with reinsertion of nir/2a for k) 2ak ~ nir . k = fc0dd.nir) + OU2ak .e. and is — as we see — less than that belonging to the lower sign case..e.2 e . where Akoce'2™. in the present case it is easier to continue with the above equations. In general one now rewrites the equations in matrix form.e. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.262 CHAPTER 12. For large values of Vb. + 0 ( e . we obtain k v ' eKa^eKa  tanh(«a) v ' These are transcendental equations for the determination of the eigenvalues of the even and odd eigenfunctions. = AA. F ^odd (12.2 a . i. Taking the ratio of the equations in both cases. the odd solution with.66) Inserting these expansions into Eq.
Classically the car is unable to overcome the bump. but quantum mechanically there is a finite probability for this to succeed.4 A car meeting a bump in the road.70) xo = (a + b) 2' .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. . V ^ 0 are given by the equations ^o + " J . 12.4.2 dxy/V(x) (12..69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E. and is a reformulated version of a problem set in lectures of Fermi.e. . 0.E)da amplitude divided Thus with E ~ 0 the probability P is given by the square of the transmission by the square of the incident amplitude. ^v + ^[E. y(x) = sin and y(xo) = 1. 12.** " W ~^7~ Fig. . b > a. Fermi [92]. .. 12. '. exp r rb rb / •J a 2mo (V(x) . Solution: The wave functions i/>o>Vv in regions V = 0. i. a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. w(x0) •• = = mogy(x). Example 12. 2m0E „ 2mnE . 57.^ 0 = °' Hence in dominant order •00 — exp . (12. exp . 2mn. p. Compute this probability.V{x)\i.v = 0.3. 1pV 2ro 0 4>o '.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. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation.
y) is the beta function (see Eq. p.384. one evaluates an approximate probability of exp[6. 14. **For instance E. .807 m s . Emde [143].2. Ryzhik [122].264 we have ro / dxyfV(x) Ja CHAPTER 12. (12.9191 ~ 1. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. S.l . h = 1.72) Assuming a mass of 1 ton of the car (i. We find™ (with ( . p.1. (15.1 / 2 ) ! = v^r) TT/2 /.4 x 10 3 9 ]. we obtain ( . We can now evaluate the probability. ^ ) = 2 ^ . Gradshteyn and I. which give /•TT/2 dx sin'M .e. 1000 kg) and using the following values of the natural constants: g = 9. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. p. n I .2 .1. dx sin ' x •• Looking up the value of the factorial in Tables. Jahnke and F.621.**.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.055 X 1 0 _ 3 4 J s . : 2 ^ B ( ^ .17)). 950. and assuming a length of b — a = 100 m of the base of the bump.71) The integral can be looked up in Tables of Integrals. formulas 3. M. 369 and 8.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.
Then considering the probability distribution determined by the squared modulus of the wave function. in Chapter 15. 265 .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.1 Introductory Remarks We distinguish here between three different types of linear potentials. Example 14. the last case allows only a discrete spectrum. In the present chapter we consider in some detail the first of these which is the potential of a freely falling particle in one space dimension. we consider the corresponding case of a particle above the flat surface of the Earth. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves. Thus.Chapter 13 Linear Potentials 13.3. each of which propagates with the classical momentum and energy of a Galileian particle. we consider the linear potential in three space dimensions that forbids the particle to escape. in the book of G.* In Chapter 14. 13. 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. * A highly abridged treatment of this problem is given. whereas the first case permits only a continuous spectrum. This is the quantum mechanical version of the problem of Galilei in one space dimension. for instance. p. 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. Siissmann [266]. This is the problem of Galilei in quantum mechanics. 144. Finally.
t). i. Therefore we use the Fourier transforms il){x.2. 13. g > 0.t) = . We see " that the case g < 0 can be related to the reflection x — —x.t) = m0g with potential V(x. (13.1) where g is the acceleration due to gravity and we can put Vo = 0.1 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. (13. (13.t).3) In general the momentum representation is of little importance. (13.t). In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x. c> 0. (13.e. V(x) = —mogx.2) which is the case with a discrete spectrum.= l V 27T Joo dxelkxip(x.t) = c\x\ + V0.4b) . and yields rriQX = mog. from d_ m x2 0 dx + V(x) 0. Recall the classical treatment of the freely falling particle. The present case is an exception and a good example of its applicability. The equation of motion follows for instance from the derivative of the constant energy.t) = mogx + V0.= l V 27T J_oo 1 dkeikxijj{k. We consider here the case of Eq. A different > "linear potential' is the socalled "confinement potentiaF V(x. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle.t) = . (13.266 CHAPTER 13.4a) f°° ${k.1). x > 0.
t) h2k2 d ~J(k. d \ ~f1 dkr[ k + m0gt ^r't (13.3) yields us now by inserting for ip(x. Applying the total derivative d/dt to this function. (13.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13. This can be achieved by first converting this partial differential equation with two partial derivatives into an ordinary differential equation with the one total derivative d/dt.6) the latter equation becomes d ~( m0gt ^(^M<=+=?.6) the argument k is then replaced by this.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k.< (13.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq. (13. (13.7) In the wave function on the left hand side of Eq. • fd ~(' m0gt %h .t) the expression (13. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i. (13.e. To this end we consider the following substitution in Eq. i.6).t) dk nikx k=oo '2K (im0g)il)(k.d m +imo9 .e. We rewrite this equation in the following form th (13.8) With this result and the substitution (13.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.t)im0g^k. multiplication by a test function and subsequent integration).t) m0gx /2TT m0gxip(x. (13.t).13.t).6) and wish to solve it.t) dketkxiP(k.9) . we obtain the differential operator on the left of Eq.5) m+tm9Wcr{ktt)=2^^ktt)t (13. Equation (13.7) applied to every quantity in Eq.
(13.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc.6) is then $(k.6). (13.= k rmt. the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.12) .1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. E x a m p l e 13.11) *[k?f.i) . t ) .10) k by m0gt back : k —> k h The solution of Eq. To obtain the solution of the original equation (13.6).e. we have to reverse the substitutions and replace in (13.10) Note the amplitude function ipo(k) in front. i.11). Thus we have first (the second term arising from the upper integration limit. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential.11) solves the partial differential equation (13. the following result h 2moi tp[k\ ^—.6). Thus addition of both indeed leaves the expression on the right hand side of Eq. (13.6) to the solution (13. We define the wave number kt as kt. (13. On the other hand the second operator expression implies dtj>o imog^—ip(k. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.11) in a neater form.1 we verify t h a t this solution indeed solves Eq.6).t In Example 13.268 CHAPTER 13.gh [ d t ' j f c + ^ ( t ' . (13.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. Our next step is to reexpress the solution (13. We have thus obtained the solution of the transformed equation. Solution: We apply the two operator expressions on the left of Eq.t).t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.t ) ' U ( f c .
4a). we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .2„2+2 mfaH 3^3+3 . 27T ^ .t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With .11). (13.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.t) = .19a) . Inserting ip(k.o o 2 —  exp iktx h mZgH' ..t) into this integral and recalling the property (13.11) can be written j'A ^.16) Next we evaluate the Fourier transform of ip(k.13) rn0_t+2 + mlg2t3 . m0g (13.t) defined by Eq. we obtain rp(x.15) we obtain / = .gfci + 2 g2m0 2/i 2 ' (13.} fA ' _ _tkt 2 k+ t) 2 k . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13. (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential. .— [It(k) .14) (13. we have ip(x.o ut = 2mo —kf 2mo A. _ k t  h2 3 Inserting here the explicit expression for fct.= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.18) h =k mogt p h S' h .+^f (13. .I0(k)].13.17) = J_/ C X ) h2 [It(k) .
23) . These wave vectors vary in accordance with Eq. Linear Potentials f dt'.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 . ipo(k) is the probability amplitude with wave vector k at time t = 0 and varies in the course of time with the timedependent wave vector kt = k — mogt/h.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .v = kh/mo. the relation (13. (13.The associated kinetic energy of the particle is mv+ = huJf.2.ut> = Jo t2 g2m0 i 3 k t .18) into a • different form in order to be able to perform the integration with respect to k. The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.20) (13.21) In other words.270 and CHAPTER 13. Such a decomposition is possible only for homogeneous force fields.o o Jo This result can be interpreted as follows. 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. (13.t) as a decomposition into de Broglie waves. 7 u>t'dt' (13.20) expresses the wave function tp(x. 13. fm0gt\ \ 2h ) (13. (13.19b) h3 3 we can rewrite ip(x.t)\2 for t — oo (distant future or past).2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x.t) =y= dkip0(kt)) exp i< fctx V27T J .19a) exactly as for a falling particle.t) as 1 I00 ip(x. To this end we convert Eq.
integrate and then take the limit fi —> 0).26) d£e ie £ kt^ m0' 0./oo dkip0(kt) exp l x ixI exp oo .26) with the formula 1/2 (13. /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. fi > 0.24) We now set fit / 2m 0 V" 0 V m0a. being now independent of £.QX 2 ^ 0 7 (m.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 \ ./m 0 a.2 The Freely Falling Particle: Quantization The integral (13.t) 1 exp r .29) . 2h h 24ift ' mogt^ r m. It follows that ip(x. may be put in front of the integral. We evaluate the Fresnel integral in Eq.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. (13.18) thus becomes ip(x. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. rnogt m0g2t3 tX^r~. (13.t) = 1 1 oo 271 27T /.13. Then ip(x. TOO#A ft and. m0xcl ht m0gt 2h gt O v: °° 2hti (13.e.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i. (13.
dA 4>o(h (13.1) has only a continuous spectrum. V(x)=m 0 gx V=E Fig.t) ~ W —exp i V irlt m ktxd J cot.272 CHAPTER 13.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. xc\).30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i. (13. For x exponentially increasing and decreasing solutions are possible.33) oo for E <C \mogx\.t) = eiEtlhct>{x) (13. kt and ut 13.31) and obtain from Eq. this expression can be brought into the following form: if *\ 0 ip{x. 13. Linear Potentials Replacing here x by the large time value x c i.e. . (13.1 A turning point at x = —XQ. For stationary states we set xl){x. 1/2  ±^(2mlg/h^x^ / 2i (13. These are incoming and outgoing waves.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 .3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0.
and hence is continuous and extends from —00 to +00. 13. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. 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. 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. we .34a) Then d &k 1 f°° ih—il>(x. so that ih dtp{k.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.6) that also for 4>{k. i. E)dk.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. i.4a) and (13.34b) It therefore follows from Eq. its momentum p would be imaginary there. 1 f01 dketlcx^{k. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. J \ „ (13.00]. probability. is reflected. Akx (13. The point — XQ.e.e. at which E — V = 0. t).e.1.4b). (13. i.0 0 dkelkxE^(k. The Fourier transform is given by the relations (13. but also d 00 dt Eijj(k. Fig. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00.e. cf.t). H2k3 4>{k) — cexp Ek (13.13.t) = dt (13. However quantum mechanically this is possible with a small.e. t) = Eil)[x.36) m0g \ 6m0 with c = const. i. i. at which the classical particle bounces back. t) = — = / V27T J .35) It follows that /W tm0g \ 2m0 i.e.t). which is not possible for real particles.
(1340) i.L [ dkeikxMk).43) y=2^x+2^E.a : ' ) .36) into (13. (13. In a similar way we can verify the validity of the completeness relation. h? h2 .L dxeikx(f>E(x).39) dk4>*E{k)4>E. (13. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf)... (13. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). </>E(x) = . V 2lT J Mk) = .37) (eikxdk.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .41) J—oo Next we insert (13.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.£') = <*(#£')• 1 27rm0g' = /o .36) (with c replaced by (13.274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. / dE(j)E(x)(t>E(x') = ^ s .e.38) we have for <j)E{k): OO /"OO / oo (13. i.e. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk'). 13.42) The Airy This integral can be rewritten in terms of the Airy function function Ai{s) can be defined by the following integral: 1 f°° Ai(s) = — / eft exp i ( st .M *K Joo Setting dt cos ( ts n Jo st). / <>~ikx VZ7T J (13.40)) into (13.38). apart from a constant phase which we choose to be zero.
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. and we shall see t h a t one branch has periodic behaviour.^ .^x + ^ E (13.43) / • '**«(» + * ) * • < * + *) ds dtexp i{s(y —TIT (27 + x) s3 ^Y J ds_ f dt_ f J exp — i< t(x + z) r 2^rJ 2W dxe^Qe^^eU*3*3) _^Pi(Sy~tz)ei(s3t3) S(y = j£j**w I ^LJt{yz) 2TT _ We will encounter the Airy function Ai(s) again in Chapter 14 in the matching of exponential W K B solutions to periodic W K B solutions. 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= ^ .J * . For a further solution Bi of Airy functions we refer in Chapter 14 to Tables.. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s. (13. (13.46) This follows since with Eq.45) One can now derive normalization and completeness integrals for the Airy functions. and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. the function 4>E{X) i and setting s y = — = 1 . . For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z).44) is then expressed in terms of the Airy function: ./ 2m 0 E (13. / s M .
(13.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.49) (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .f dzJ*W 2TT JC with z = reie. o (13.276 CHAPTER 13. or n <6 <TT o for r > oo. i.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . Linear Potentials 13.TT < 30 < 0. i. .51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0. . (13.r 3 sin 30 ). (13.51b) ..e. (13.. Thus we must have sin30<O.4 The Saddle Point or Stationary Phase Method The integrand of Eq.TT < 6 < 0 for r > oo.r 3 cos 30 J + i ( sr sin 9 . (13. 13. We want the integral to exist and so desire that lim e**^ = 0.e.2 The contour C for s < 0 with ends in the angular domains (a) and (b).\zz.43) contains the phase (with x — z): > *(*) := sz .
55) into (13.54) i ¥~s)3/2' (13. where the derivative of the phase function vanishes.57) $"(z s ) = 2% s.e. —i ${zs) = szs .e.z3s and V^s + . (13. at which the phase becomes stationary.48). i. so that exp(±2i9s) — 1 and — s — rl. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.13. (13. (13. (13. 0 = &(z8) = s~zt s = za „2„2i6>. provided its ends at infinity satisfy these conditions.( \ / ^ ) 3 2z.55) s < 0. Such a piece can be found in the neighbourhood of the socalled saddle point zs. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.S )V2 (13. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. oo] as indicated in Fig. i. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. and the main contribution to the integral comes from the a region around the saddle point.54) where a is the new variable with —oo < a < oo.2. Inserting (13.52) = exp(±i7r) Let s be real and s < 0.51b). Then the ends of the contour lie in the domains specified by Eqs.58) .4 The Method of Stationary Phase 277 The integral (13.53) Since we shall have the contour C in the lower half plane.51a) and (13. (13. we choose zs = iy/^s. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. 13.45) exists for the contour C.56) 2n(s)^Jc But I daefr"?*"^. Then 6S = ±7r/2. and we set with a real: z := zs + a (S)V4' (13.
M.44) we can convert Eq.42) into an equation with a more appealing form. V^s / 4 V3 (13.44) and (13. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle. This is an important aspect in the WKB method to be considered in Chapter 14. Sec. cos .64) tSee I. .63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). S.32) with the solution (13.60) The reason for describing the point zs as a saddle point is that the integral in Eq.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. V 2h2 (13. (13. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction._ 1 .278 CHAPTER 13.59) In Example 13.32) therefore becomes dz2 + zd) = 0. With the help of the substitutions (13. as one can see from the following equation^ ri' + Ffz •2^2. Gradshteyn and I. (13.491. (13. > so that Ai(s) 2 v ^ ( . Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. 2 s 3/2 . (13.s ) V 4 exp \(s) 3/2 (13. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. 8.56) decreases exponentially around a — 0 for real values of a.2/32u = 0 (13. For (—s) — oo the integral converges towards T/TC.45) 4>E(X) ~ Ai(z). (13. According to Eqs.2 we show that for s — oo: > Ai(s) 1 . Ryzhik [122].62) This is the Airy differential equation with one solution 4>{z) = Ai(z)..
Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. with *"(z) = 2z. .43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).^ e ^ / V2 4 [°° Jo dae'2^. 27T Ic Jc 3 This time s is real and positive. Thus there are two symmetrically placed real saddle points on the real axis. Choosing the contour C to lie along the real axis.z ± ) 2 * " ( z ± ) . Then. Eq.4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. Example 13. * " ( z ± ) = =F2Vi.51b) are satisfied.. cos  . we obtain f° Joo dp=—[a+ i\a\]da. (13. We have therefore r /*0 poo zs = ±y/s = z± / Jc dze^M = / Joo dze^^ + JO dze^z\ In these integrals we set respectively P with *(z) = #(z±) + i ( z .59) and (13. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .2: Periodic asymptotic behaviour of the Airy function Obtain with the saddle point method the asymptotic larges behaviour of the Airy function Solution: We have (cf._ .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. changing to the variable p.13.51a) and (13. the conditions (13.s 3 / 2  . P Introducing a new variable a by setting p = ai\a\. It follows that for s —> oo: Ai(s) ~ .
e. 13. ~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq. / " ( z o ) = —n/z2 = —1/n.3: Derivation of the Stirling approximation of a factorial Obtain with the saddle point method from the integral defining the gamma or factorial function the Stirling approximation. and hence integrating as described above through the saddle point indicated in Fig. Linear Potentials E x a m p l e 13. i. since f'(zo) = 0 and (z — ZQ)2 = —x2. (11. Jo Jo \ .3. Then. 13. Thus here / ( z ) = — z + n l n z .111).280 CHAPTER 13.*.Be 2 Fig. Solution: Briefly.3 The saddle point at z = n + ix. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. . z o = n .dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. ^/2KJW1. n!= / e~zzndz= / e~z+nlnzdz.
where *G.295. named after its first promulgators in quantum mechanics: Wentzel. Brillouin [37]. It goes without saying that such comparisons are very instructive. Froman [99]. N. O.Chapter 14 Classical Limit and W K B Method 14.g. ' T h e most prominent earlier expounder is H. § See e.e. 316 . been developed previously by others in different contexts. i. Froman and P. 281 . For the history see pp. Jeffreys [144]. the linkage of solutions above a turning point to those below. B. Dingle [70]. A critical overview of the history of the method. *R. will there be that of a coupling parameter. H. The WKB result is therefore frequently described as the semiclassical approximation.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs.317. 292 . ^ The central problem of the method is the topic of "connection formulas".§ 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. A. for misunderstandings pp. however.* The method had. Wentzel [282]. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. of misunderstandings and other aspects may be found in the monograph of Dingle. and the role played by h in our considerations here. Kramers and Brillouin. 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. Kramers [153] and L.e.
2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. In this equation we set.1) (14. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. (14. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).7.3) .g.* We illustrate the use of this condition by application to several examples. the energy has a definite value E with V(r.V(r)]^(r) = 0. the derivation in Example 18.282 CHAPTER 14.t)\2 = \iP(r)\2. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . For stationary states ^/.t) = eiEt/h7P(r) and \V(r. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. 14. A(r) and S(r) being real functions of r. V>(r) = A(r)eiSWn. Classical Limit and WKB Method E — V changes sign). the function S being called phase function. for which the observation of position and momentum is independent of the time at which the observation takes place.
(14.t)eiS{r't)/h.V(Y) + and 2VA • VS + AAS = 0.Q A 2i 8A 1 (14.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.4) (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.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. h h (14. We can identify Eq. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation. (14. (14.* VAVS+AAS + V(r)A.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14.13) .t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r. We set in this equation.12) VA A (14.1) and separating real and imaginary parts. we obtain the following two equations: E . h2 ih— = H$ A + V(r) * . (14. (14.14. V(r. These equations are equivalent to the Schrodinger equation.t)}2.10) m 0 ^ .+ VAVS + AAS = 0.
(14. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. (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 . (14. we have mo—(A2) + V(A2)VS + A2AS = 0. ^ + V • J = 0. + h2 AA V = — ^ .17) With Eqs.21) . (14.n. the implications of this equation are also those of the equation of continuity. Since J = — A2VS = PVS. TTIQ (14. (14.. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq. (14.J = 0. (14.14) A2VS. mo <™> (14.10).10) we obtain = — [V(A2) • VS + A2AS}.16) BA 1 moA—.9) as 9S 1 mo^ 2 + T.284 CHAPTER 14.15) and (14. .+ AVA • VS + A2AS = 0.20) With this we can rewrite Eq.01. (14. Next we investigate the significance of Eq.9) for the phase function S. m0 But now (14. Since we obtained this equation from Eq. ie.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. (14.
(14. In the first • place we require the ansatz (14. . This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics.e. (14. V • J = 0. the wave function \I/ describes effectively a fluid of particles of mass mo. and hence''' ^(m 0 v) + m 0 (v. 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. a singularity different from that of a pole). 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. where. d — = (14. We deduced this equation solely from the phase S in the limit h — 0. that the quantum theory implies an essential singularity at H = 0 (i. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). dt 2 Constructing the gradient of this equation. with the help of Eq. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. This result is very general.20).22) (14.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. The equation of continuity then describes the stationary flow of a fluid. r o t v = ( l / m o ) r o t g r a d S = 0. we obtain ^ V 5 +  v ( v v ) + V V = 0. However. This means. which for very small values of h yield the dominant contributions. i. The particles have no interaction with each other.24) d 1 dr d •— = d Lv _ . One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.23) (14. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents.1/h. they are asymptotic expansions. Such series are typical for expansions in the neighbourhood of an essential singularity.e.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ.14.2) for the probability amplitude. so that V ( v • v) = 2(v • V ) v + 2v x rotv.V ) v + V 7 = 0. This expression with h in the denominator shows. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2.
26) We observe already here that an expansion in rising powers of k.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.5)) and 2 ^ f dx ax + All=0. B.27) (14. We consider the onedimensional timeindependent Schrodinger equation y" + ^. similar to our earlier procedure with A and S real functions. (14.3. is an asymptotic expansion.286 CHAPTER 14. (14. (14. i.2.e. .[EV(x)}y = 0. There are also other more specific descriptions. The expansion in powers of h2.§ y = A(x)eiS^'h where W(x) = S(x) + . Messiah [195]. Dingle [70]. Sec. the method is widely familar under the abbreviations of the names of Wentzel. i = eiW(x^h.30) ax *For details see R. 6. §We follow here partially A. z (14. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).28) The calculations we performed at the beginning imply the following equations (compare with Eqs. Classical Limit and WKB Method 14.4) and (14. such as "LiouvilleGreen method' and "phase integral method'. 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.$ 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. p.3 14. Kramers and Brillouin. or h2/2mo(E — V). As remarked at the beginning. 318.In A(x). We set.
(14. (14.A = const.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .33) 3 (S")2 . + (14.32) is therefore (S'0)2 + 2H2S'QS[ .14. Eq.31) \a. (14.3 The WKB Method 287 Equation (14.4 (S>)2 1 S'" 2 S> (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.29) becomes (note that this is still exact!) (S')2 .35) . (14.2mQ(E .30) corresponds to the equation of continuity (i. (14.32) s" = sz + h2s'( + •••. 3 / 5 .\ In S'.V) and (14.2rn0(E V) = h2 lS'A 2S'0 (14.35) can be integrated. Li This expression can be substituted into Eq.28).10)).36) Equation (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 . But first we note that: A dA dx2 2 = c{S')ll\ = A 15^ 2 S' + ^ = A ax 3 (S") 4 (S')2 2 2 S' With this Eq. . Since the expression (14.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .e.
1 S e e R.19).33) to make sense as an asymptotic series. y/2mo(V(x) . and instead is asymptotic which we shall not establish in detail here. E > V(x).37) For Eq.E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. The solution is seen to be periodic provided k(x) is real. (14. p = h/X) k{x) :Then S'n +.39) and (14.' For the expansion (14. (14.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). (14.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . (1. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====. (14. we expect that in any case So » ft2Si. i. Dingle [70].± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2).27) we obtain now with Eq. Classical Limit and WKB Method has the form of a momentum. . B.and So = ±h / — — A J k(x (14.288 CHAPTER 14.39) where a and /3 are constants. Eq. (14.e.40) rx dx v "7 W)\.41) We now ask: What are the conditions of validity of Eqs.38) h y/2mo(EV)' (14. Chapter XIII. it is reasonable to set (cf.41)? We note first that the expansion in powers of h2 does not converge.
Eq.3/2 s&T + 3 {So Il« we have (cf. 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 (14.36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.1 / 2 ]" [§0SS).3 The WKB Method We have 289 S0 = ±h and Si follows from (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.45a) .42) [(^).14. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .36). Since k(xy 1/52\S'0 (14. E > V(x).V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14. (14.
*2 A(z) = V / 2m 0 (£ . Roughly speaking one therefore requires E to be large in both cases. (14.V(x) = V(a) .46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. (14.a)V'{a).45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l .2 Turning p o i n t s a n d m a t c h i n g of W K B solutions Let x = a be the value of x at which E = V(a). In a similar way we have for E < V(x): l'(x) <C 1. at which E — V{x) changes sign? We note first that in this case near x = a: E .290 CHAPTER 14. In the neighbourhood of E = V(x).48) behave in the neighbourhood of such a point called "turning point'.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. and hence (with UIQ = 1/2. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.48) The (dominant) WKB approximations are exact solutions of this equation. h = 1) . (14. V(x)) (14.e.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. around the classical turning point. Classical Limit and WKB Method we can rewrite the inequality (14. i. 14. In terms of X the original Schrodinger equation is y" + = y = 0. the WKB approximation is in general invalid (see below). How do the solutions of Eq.V{x) ~ Or .3.
1 The regions around the turning point at x = a. 14. which connect one domain with the other.3. The transition is provided by matching relations. and on the other side. 14.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. i.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. 14. (14. since this requires different approximations there. 14. 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. (14.1.e.1.3 after stating them first here.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq.14. V=E Fig. in the domain E > V the WKB solutions are oscillatory. We shall not derive these conditions here in detail but rather make them plausible in Sec. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . 14.1 there are WKB solutions on either side of the turning point and some distance away from it. As indicated in Fig. On one side. in the domain E < V the WKB solutions are exponentially increasing or decreasing.48). We assume we have a turning point as indicated in Fig. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.
52) For a potential rising from left to right. This means the asymptotic part suffices only in t h e exponentially decreasing case. 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. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. Note also the extra factor 2 in yi. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi.292 CHAPTER 14.50) x S> a. a interchanged. as in Fig.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . in order to make the formulas independent of whether the potential is rising or falling. 14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . (14. (14. one could simply replace everywhere /•a / pa • • • b y / Jx J x .1) is given by the following relation with the + sign in the argument of the exponential on the right and no factor of 2: VACOS I / —. has the same asymptotic behaviour in the domain i « a a s Ay\.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. Thus. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. 14.
(14.2 The overlap regions around the turning point at x = o. 14. 14.3 The WKB Method 293 In that case Eq. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14.46).14. i.3.54) . whose exact solutions are the dominant WKB approximations. the equation V approximately linear around turning point V=E Airy solution WKB . is singular at a turning point x = a.a). the matching relations have to be altered accordingly. the equation becomes approximately y" = (xa)xiy.e. 2 1 X (14.48). 1 ~ (x . as we saw.53) Since at x ~ a.51) — always with the decreasing exponential on the left — is valid in both directions.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.Airy overlap* Fig. If higher order contributions are to be taken into account. (14. (14. Therefore we consider now this original equation in the domain around x = a. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. y" + Toy = o.
(13. In the following we require the asymptotic behaviour of the solutions Ai(—z). Dingle [70].Bi(—z) for \z\ — oo. 448. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. i I 772 \3 + > for ia>0. This is given by the following expressions. or M.57b) UV2 3 11 by J dx .56) We have discussed some aspects of solutions of this equation in Chapter 13. R.55) the equation becomes the Airy differential equation d2y dz2 zy. formulas 10. p.^ 4 I 7T x — a <C 0. J (X (i4 58)  Replacing in Eqs.57b) To the same degree of approximation as the equation y" = (x — a)xiy.60). > for a.a)3/2 ^ /2 =^ 3/2 o Q. (14. as > may be seen by referring back to Eqs. Stegun [1]. Cf. (14.59). Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. A. p. we have r x * f x ' /2(x a)i/2d*=^(x .4.a)X\/3.5960. (13.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. (14. Abramowitz and I.1 / 4 by \^. (14.294 CHAPTER 14. (14. In particular we encountered a solution written Ai. B. A second solution is written Bi. . With the substitution z = (x .57a) and (14.291. — a ^> 0.
l.2. and hence must be proportional (in view of the uniqueness of the solution there). As an illustration of the use of the WKB solutions we apply these in Example 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 / — = . The latter domain is a small region where the solutions from either direction overlap.l v/2m0(EV(x))dx 7r = (2n + l).14.51) and (14. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. Example 14. In this way different asymptotic branches of one and the same solution are matched across a turning point.2. by Ai(z).(zf2 by J we obtain the matching relationsX (14. (14. 14. e. 14. These solutions.3 The anharmonic potential well. n = 0.59a) . We summarize what we have achieved. 14. As we approach a limit of this interval in going away from the turning point. however. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions.g.3 The WKB Method and 295 •. are only valid in a small interval around the turning point at x = a as indicated in Fig. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation).1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution. In the direction to the left as indicated in Fig.52). In this sense we have now verified these relations.1 to the quartic oscillator.. we enter the domain of validity of a WKB solution.
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 . /n = s u ^ TJcos^ya TJ+cos^a T J . 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 ( .296 CHAPTER 14.59b) Solution: We consider an anharmonic potential well as depicted in Fig. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a. The condition for this is ya{x)=yb{x) for xe(a. (14. .61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique).. . Here we are interested in the discrete states (the only ones here). i. j/i. 14. ^ A = (2n + l ) . regions 1 and 3 in Fig. The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained.44) this implies large values of n.b). (14. Expressed as an integral over a complete cycle from one turning point back to it the relation is: <b dx^2m0{EV(x)) = (n+jh. Let E = V(x) at x = a and x = b. . Classical Limit and WKB Method where x = a. We argued earlier that the WKB method is suitable in the case of large values of E. 1 . for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. n = 0 . 14.y o dx TT T provided 6 _ 4 /. extends only over a length of very few wavelengths and is therefore valid for n small. . Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. as in the case of the harmonic oscillator itself.2. Hence we search for solutions which are exponentially decreasing in the far regions.51). yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b.l. 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. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. (14. 2 . According to Eq. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin.60) Using the matching relation (14. (14.(x) = c ' v Acos I / — I for a < x < b.3. Then around these points the potential would be well approximated by a linear potential (as we saw above).^ . (14. Sometimes this distinction becomes imprecise.e.3.. . b are the two turning points. the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. 2 n = 0.
14.64) (14. The corrected form* is. Bohr. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0. equivalently. (14. B.e. Sommerfeld and W. .63b) *The author learned this in lectures (1956) of R. n = 0. . This relation is sometimes wrong. with n — 0.x number of turning points h. Wilson by supplementing classical mechanics by Planck's discretization.4 Three cases with different pairs of zeros.63a) or. 2. (c) V Fig. 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. pdq = n + 1 — . Dingle. .x number of turning points TTH. and knows of no published form or script. ri2 I Jqi :pdq n + 1 — .4 BohrSommerfeldWilson Condition 297 14.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N.1. (14. A. 14. 2 . In this socalled "old quantum theortf (i. .1. .
14.298 CHAPTER 14. this is an exponential zero.4(a). i.e.65) We use the knowledge of these solutions to find the eigenvalues of the equation. 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.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. Case (a): The case of two trigonometrical zeros. Since sine and cosine have zeros spaced at intervals of TT.66) J<1\ Thus in this case the socalled old quantum theory is correct.e. as illustrated in Fig.e. 0(g) ex . In this > case the wave function for g > go is . this means dqyf—f = (n + l)7r. where /(g) is negative. i. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. i. i. / • n an integer > 0. (14. (14. 14. where n —. Jqi ^ fpdq J =2 pdq = (n + l)h. the case of tp(q) — 0 at q = gi. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. There are three cases. In quantum mechanics we have dq2 + Ki/j(q) = 0. together with q — — oo. this is a trigonometrical zero. if /(g) is negative. We have the case of trigonometric solutions. where /(g) is positive. Before we continue we recall from Eq.67) Case (b): Next we consider. (14.4(b).
Hence we have pdq= i n +  .77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). ri\ / J do y/f(q)dq (n + 1) 1" (14.63a) or (14. i.63b) to obtain the eigenenergies of the quantized harmonic oscillator. Thus in this case the socalled old quantum theory is wrong. Therefore their arguments differ only by (n + l)ir. For the wave function to be exponentially decreasing to the left of qo and using Eq.2: The harmonic oscillator Use the relation (14.4 BohrSommerfeldWilson This will vanish at q = q\ if rii / V^f{q)dq+j^ Jan It follows t h a t i 4 Condition 299 = {n+l)TT. this is the case of two turning points.e. (14.70) We can therefore summarize the results in the form of Eq. E x a m p l e 14.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. (14. or Pdq (n + l ) " 2 h. i.3.2 and 14. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . As illustrations we consider Examples 14.63b). and • • for the wave function to be exponentially decreasing to the right of q'0. 14.4(c).69) pdq = (n + l ) " 4 h.14.68) (14. Since both cases refer to the same region qo < q < q'0. Solution: As in the case of the quartic potential. (14. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. the sines must be proportional. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r.e.
in principle. where g is the acceleration due to gravity. 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. and h = 6. 14.71) For m 0 = l x 1 0 _ 3 k g .2 . Thus there is one turning point at E = mogq. one obtains E ~ 2. Calculate its quantized energy. g = 9. as illustrated in Fig.2 3 J x (n + 3 / 4 ) 2 / 3 . permit this although with rapidly diminishing probability. whereas a turning point does. . 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.5. 4 / x CHAPTER 14. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B .28 x 1 0 .807 m s .63 x 10~ 3 4 J s .5 A particle above a flat Earth.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.V(q))dq The result is E/u.300 With potential V(q) = 2ix2mov2q2. 14. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. :: V=m0gq Fig. Example 14.
75) which verifies the formula immediately.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 £ = . h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .5 Further Examples E x a m p l e 14. In the case of screened Coulomb potentials (cf.74) p hjtl *2 —t\ _ mo f 2rr 2K J" X2 dx p~ ~ d dE For a period T from x\ to X2 and back to x\ this implies T 1 mo / d a .5 Further Examples 301 14. d dE V)={n+)n.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.= — <*— = h(n+\. 9 . E x a m p l e 14. (14. with p = modx/dt. 1 . Solution: The proof is contained in Example 18.5: 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 ( — = .( 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 (14.72) \\2dxp=Qdx^{Eh Hence.14.f V ft h JXl h 2TT (14. and u> is the oscillator frequency in either well.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB. Chapter 16) one obtains the quantization relation (for Mo = const.7. ra 2/ = 0. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V).l.73) Jx! rp2a dx I^<E 2 v) = ^ n — = . (14.
6.78) (a) We determine first the three turning points at qi = <?i.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q).<?2i<?3 for E > 0 given by E — V = 0.133)).~ \ (14.6 The cubic potential.a .76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. 14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig. 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 . The q term in this equation can be removed by transforming the equation to a cubic in y.77) Determine (a) the turning points. Example 14. which determine the imaginary part of the energy. where . q = y+ a . where 1 2 • 1 2 y = Q. Eq. 14. (11.e. (14. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14. (14.y2)(y . Fig.79) . i.302 and T CHAPTER 14.y3) = 0. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.
120°)] = a 20? 2a2 3 2a 2 • cos 0  91 . p . 0<6O°.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. 6 = 0 i m p l i e s E m a x of E q . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 .80) I n s e r t i n g t h e r o o t s yi.' : ± 6 0 ° w h e n E = 0. 92. V cose' + sin 8 (14. (14.e. H e r e cos(<9 ± 120°) = . § T o h e l p we cite H. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .c o s O + cos(6» . a n d t h e a n g l e 0 = 0° t o E = £ m a x .\tan0^ v/3.9 2 [ .78)) c o s 30 •• 4P3 ( ' \ aeb2} . 93.c o s ( 0 + 12O°) + c o s ( 0 .2 c o s ( 0 + 1 2 0 ° ) ] .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 . f o r m u l a s 236.q)(q .k 2 2tan# \/3 + tan! .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. 80 a n d 361.81) m a 91 .6.04. _ \ / 3 — tan(9 2 V92 .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 . 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.84) 92 . B y r d a n d M .= s i n t 3 v 3 (14. F . e > 0. i.gi = 1. 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. p . tan60°=%/3.9 3 — = = 2 " 3 " [ . 92 = a2[l . g 2 ~y{3}>0.14. D w i g h t [81]. M .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . V3 + t a n 0 2 . (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0.9 l ) f f / du s n 2 u en2 d n 2 u .08.93 2a2 [ .g i < 1.< 1.0^6O°. D . M i l n e . B . ui=sn 1 \l— — = l.1 / 2 . (14. T h u s we c a n n o w w r i t e t h e W K B i n t e g r a l from > o n e b a r r i e r t u r n i n g p o i n t t o t h e o t h e r as (recalling t h e factor y/2/b in front of E in E q . (l_fc' 2 a + fe'4)l/4.cos 0 + cos(6> + 120°)] = a2 92 . 91 < 93 < 9 < 92(14.T h o m s o n [196]. 212.g i ) ( g 2 . 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 .79) a n d t h e n s e t t i n g m o = 1.5 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. q3 ~  . F r i e d m a n [40]. (14.83) 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 . .ui=K(k). 78: sin 120° = v ^ A c o s 120° = .3/2. q3 = a2[\ . p . ' P .2 c o s 0 ] . gi ~ y {V3e} < 0.{ V 3 e } > 0. we set 0 = 60° — e. p .93).2/3 i n t o q = y + a 2 / 3 . (14.a 2 [ l . we h a v e qx = . 37.120°)] = . 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. 14.
87) Inserting G(fc) together with the expressions for the prefactors into (14.84) we obtain Carrier = 2a 2 sin6> 2 b( .fc'2 + fe'4)"1/4.00. V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . 2 2 . V 93 . the integrand of the integral across the well is effectively the inverse of this momentum.86) 15 tan. with K(k = 1) = oo.91 F(TT/2.90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .~ y ^ ) a< [ COS 5 + . Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. g= 2 .?)(?3 . G(fc) =86aS (ITS£ li[fe. Friedman [40].89) This expression will again be obtained later with the help of configurations called "bounces" (cf.91) a2[cos0^] y/3tanfl ? L . p. (24. Byrd and M. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 .k). 91 < q < q3 < 92 bJc. Next we evaluate the required integral across the well from 91 to 53 at energy E. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! . (14.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — .2(fc2 . D. E(u = K(k)) = E(k).^ + ^ ) (l + fc'2)2 . E(u = 0) = 0. (14. Eq. 1+72 = and J o s 0 + ^ = (1 .32)).qz) (14.S .^ V3 / V V3 y a / c o s e J. Whereas the integrand of the above barrier integral is effectively a momentum. Thus Iwen=lgF(ip. .= —^ = k a2[cos0+^] v ^ + tanfl V=". . The WKB Method elliptic integral of the first kind. (14. V?2 . sinV = q3 qi ~ = 1. F. 72.92) ''P.sm t %/3 £\. A:') = # ( * ' ) • (14. as in (14. We can evaluate this integral again with the use of Tables of Integrals.304 CHAPTER 14.) .2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. 0 (14. E(u) the incomplete elliptic integral of the second kind.91 . formula 233.85).
a°b n = 0. 906. so that even in the case of the ground state the zero point energy will contribute to the prefactor. (14. . 32 (14. M. we obtain 15 .2 IE 4 fc' ln 8 (" «V iU^fc' 4 . To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).80). (14.14.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.3. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. S.** However. . (14..96) Expanding the coefficients of the elliptic integrals in Eq. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. Gradshteyn and I.4 2n + 1 fc' = 4 — ^ — .85) (14. 2a 5 b K(k) + 2 E(k). formulas 8. 905. pp. to represent a physical decay rate.88) in rising powers of fc' .5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .3 and 8. Thus one obtains from (14.113.2. (14.63b). With q = 1/z the integral (14. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. and hence at E = 0.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) .93) where w is the harmonic oscillator frequency in the well.114. (14. we obtain > .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. we obtain E a6b2k'4(l + k'2).94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. Ryzhik [122].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.l. 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.
/25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l . 8 Bfc . also Eq.ba / 25a5b\n+2 ±l_l eT5 a f >. i.73).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 [ . n ~ V27T .306 CHAPTER 14..Wigner formula (10. (0) i hw En = E\> . _ exp[2/barrier] >.7 ..101) (14. this has not yet been done. the result agrees with the ground state path integral result (24.e.fc'2 + fc'4)1/4 * ± t ^ . It follows that (for one complete orbit back to the original turning point) .100) 'barrier = ~0°b In [ ^ — j .„.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq./^~^l^ab ^aub . (20. _ T T • (14.2 7 b a r r i e r ] Z Z7T where (cf. 2n + 1 — / 26a5b \ . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .e.178)) call the Furry factor set equal to one. so that we obtain 4 . 15 a .> tf = = 2/wfiii = ±tv8a°b——P. (18.j . .( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) . We observe that with \j2je as 1.. . oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.96) for the imaginary part of the energy as in the Breit. i.— (14. 7 ~ —.102) .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. V (14. The WKB Method For k' —• 0 the first and the third terms dominate..J + 0(VoJb). (14.
now known as quantum chromodynamics. Gottfried.1 Introductory Remarks The particular linear potential we considered in Chapter 13. See particularly E. C. in fact. [232]. describing free fall under gravity. D.* 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. after the realization that quarks as their constituents might not exist as free particles and.e. at least indirect. Rosner [269]. Yan [82]. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. indications from the nonabelian generalization of quantized Maxwell theory. This potential became widely popular in the spectroscopy of elementary particles.M. led to a spectrum which contains only scattering states. [231]. C. the classification of the various states of nucleons and mesons and other particles. Quigg and J. Eichten. Lane and T. Rosner [230]. which permits only bound states. Thacker. T. Quigg and J.Chapter 15 Power Potentials 15. that this is indeed the case. L. The study of this potential. K.g. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. There are numerous. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. B. [233] and H. along with inclusion of angular momentum and spin effects and their interactions. 307 . Kinoshita. the Coulomb potential. K. 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. i. L.
and the normalization [ dr\y(r)\2 = 1. R(r) = ^ . For a central potential V(r) we write (15. (15.308 CHAPTER 15.r acting between two particles is constant and directed towards the origin (of the relative coordinate).V 7 = const.1. ~ (I + l)rl (15. mi + m2 and r is the relative coordinate.u ~ r i + 1 (/ + l)R(r).4) Instead of considering the linear potential.2 The force The Power Potential F = . (15.£ ] * ( r ) = 0.4>)ru{r) = Ylm(9. We leave consideration of the logarithmic potential to some remarks and Example 15.e.m2. 15.3) with the boundary conditions''' u(0) = 0. we return to the Schrodinger equation in three dimensions. (15. and so u'(r) .1) i. where \x is the reduced mass of the two particles of masses mi. These particles therefore cannot be separated by any finite amount of energy. v > 1. </>)R(r). This force maintains this value irrespective of how far apart the particles are separated. f dr{u(r)}'2 1.5) u'(0) u[r) r=0 = R(0). h2 2n V 2 * ( r ) + [V(r) .2) tt(r) = Ylm{9. m\m<i M= : . i. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). Power Potentials general power potentials. In order to have a clear starting point.e. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. and u'(0) > [u(r)/r]0.6) = (I + l)u(r)/r = R ~ rl. A > 0.
7) k(x) = h/y/2m0(EV(x)).2. 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.9) Threedimensional case: We transcribe the above considerations of the onedimensional case into that of the threedimensional case by restricting ourselves in the latter case to S waves (no centrifugal potential!).15. d5.9) into the threedimensional case in the following form: / Jo C dry/2n(E . (14. (15. Then k(x) = k(x).4) in the threedimensional case. x > 0. . We therefore proceed as follows. We now want to find a quantization condition for the threedimensional problem. (15. (15. . Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality.V(r)) = (2nl) +1 N. N = 0. so that with Eq. (15. .63a).2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. which has been reduced to an effective onedimensional problem in the polar coordinate r.63b) for the onedimensional Schrodinger equation. or f l H " ^ ) ! . This means that we transcribe Eq. is to be interpreted as the principal quantum number in three dimensions. (15.10) Here rc is the turning point and n — 1..8a) we can write the quantization condition (14. (15.63b): 2 and V(0) = 0 (15. 3 .2. odd integer ^=(nll)vh.8a) ^ JO = (2Ar + l ) £ . and the turning points are at a — — b := — xc.1.8b) This condition implies that we have only one pair of turning points.. We also assume that V'(x) > 0. In order to to obtain the ."MA..
y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. (15.17) The integral on the right is the integral representation of the beta function B(x. Now.13) dt . s.14) Prom Eqs.y) defined by (xl)\(yl)\ y ^ (x + y . .\rvc = 0 . E A fv.'(1*)1/2= J1F\lt)y1dt. c (nj\irh.15) I ' = ^ 1 (( 1) B{x.12) and (15. dr = . _ —vr" *dr.13) we obtain _ X_ „ _ so that A£_ (15.11) rc = (15. (15.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15. (1516) (15. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2. (15. Power Potentials eigenvalues E —• En for the radial potential (15.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .310 CHAPTER 15.
i r ( i ) = r ( l + i ) = ( i ) ! . $T(z) = {z. The case v = 4 is that of the pure anharmonic oscillator also discussed by M. for instance simply graphically as indicated in Fig.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. One can convince oneself now.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. (15. and n — — ]7r/i.1 Approach to square well with v —> oo. ^ '"Compare with Eq.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. Weinstein [281]. . t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. (12.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.1.15.1)!. 15. T(z + 1) = z\.55).
In the case v = 1.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15.21) (4nl). x > 0.2) implies A s fj. discussion at the beginning of this section). since these are the ones we are interested in (cf. we obtain E„.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. > and with Eq. eigenvalues (JV + h)tkj — (2n .oj = ^/2X/JX.25) "Comparison with Eq.2.g — —2. Power Potentials In the case of v = 2.41) for the .61) h2 = I. (13. 15. 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). The potential V(x) = X\x\ (15.312 CHAPTER 15. the case of the linear potential. 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. (6.oj2/2. (6. we obtain En = (n .24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 . For the following we set in Eq. Physically it does not make sense for the probability amplitude to have a discontinuity there.> r ( 2 ) A V 2 in \)*\W \it^T& (15. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. (15. the harmonic oscillator.mo = \.1 + \)hx> = {An • l)£fiw.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.
this wave function has the required exponentially decreasing behaviour at infinity.29) . and for s —> oo the trigonometric behaviour of Eq.e. E > V or s > 0. In the domain —a < x < + a . (13.56) and (14.60) applies. (13. (15. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. i. i.26) For z = s := E — x —> —oo. (14.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig.15. dz2 According to Eqs.27) i.x).e.2 Behaviour of an odd wave function at the origin. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E .x)\x=0 = 0.28) In particular at x — 0. Ai(s) "v^ G ' cos s 3/2 _ t (15.59) applies. s = E. Ai(s) 1 2v^(s)V4 exp x —> +oo. i. Eq.e.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .e. i.e. (15. \(s?» (15. 15.
2. L. Power Potentials Table 15.82878 1 2 3 4 5 6 7 8 9 10 2. It is possible.08795 5.51716 6. = im — (7r/4).02137 10. i.^ = !(2nl).29) implies that cos (  ^ . The expression (15.00767 11.93528 12.78445 7.e.03914 11. so that n — 1/4 ~ n. Quantum Mechanics with Applications to Quarkonium.e. Quigg and J. It is interesting to note that in the dominant approximation both methods agree.314 CHAPTER 15. .1 demonstrates the quality of the WKB approximation by comparison with the exact values. p.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. (15. Rosner..30) ! ^ / 2 . 201. (15. n = l.32025 4.e.52056 6.82814 i..3. (15. (2/3)E3/2 or 3 / 1 2/3 E — En — . Co.En from Ai(£„) = 0 2. to obtain the zeros of the Airy function numerically.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .= ) = «. but for large > values of E Eq.08181 5. Table 15.94249 9..33811 4. the eigenvalues En are determined by the zeros of the Airy function. Table 2..94413 9.93602 12. with permission from Elsevier.00852 11. copyright of NorthHolland Publ.28) is really only valid for s large and s — oo.78671 7. *This Table is reprinted from C. [231].04017 11. i.02265 10.31) (an odd function has an odd number of zeros!). of course.
3 The ThreeDimensional Wave Function 315 15.** r ( # ^ e e ) o c *(0) 2 . r <53) 1 3  dr cos 2 () (15.e. i. e.g. of course.34a) **The most frequently quoted reference for this result is R.e. We saw previously (see for instance Eq. ee.3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. D. .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.e. the charmanticharm meson ^ . See also J. (14. In order to be comparable with experimental data these investigations. have only a finite lifetime and decay into other particles such as electronpositron pairs. P. i. In general these quarkantiquark pairs. which are inversely proportional to the lifetimes. We shall encounter WKB normalization constants at numerous points in later chapters.g. can be shown to be proportional to the modulussquared of the particle wave function at the origin. in Chapters 24 and 26. cos2()^— / 1 fr° 2 > w . Jackson [140]. The resulting masses agree in most cases very well with those extracted from experimental observations. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. F. Weisskopf [239]. van Royen and V.15. e. We indicate briefly here — without entering into further details — how these quantities can be calculated.• •) by a mean value. *Note that this makes N a W K B normalization constant. The decay widths T.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions.
so t h a t approximately l N2 / Jo k(r)dr . (15.l N n ..32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 . & (15.rcu w X{r)dr = Kh .38) N : sin V ^ Jr dr' 7T (15. (15.. (15.2) and (15. (15. (15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15. dE a—— / dn J0 so t h a t with Eq.35) N2 .1 )n . From Eq. i.316 In accordance with t h e relation CHAPTER 15.40) VWY . Power Potentials this averaging implies a numerical factor like 1/2.2..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.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .e. A = (15.39) in / .34b) VMEvy Variation or differentiation of Eq.\ \ * n = l.37) (15.36) \i dEn irh2 dn According to Eqs.
38) implies 317 (15.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. (15. Hite and H.15. 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). (15.44) (15.* Using the result (15. § S. W. Itt^pocn2*"1^"^. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. The case of arbitrary positive power potentials as above has been considered by R.l ) n .23) for the energy levels of the linear potential.41) I*(0)2=_L^)^. . MiillerKirsten [147]. . MiillerKirsten [28]. the expression (15. J. S. MiillerKirsten [29]. Bose and H. K. 3 . Kaushal and H. J. K. . we have ^'(0) = ( . An enormous literature exists on the subject in view of its relevance to the spectroscopy of mesons and baryons made up of quark constituents and the necessity of checks with results derived from experiments. W. G. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. E. W. S. for various potentials: (a) For a confinement potential of power v. J. Bose.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. .g.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. for instance. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e.37) we obtain: 1 (2M^)1/2 47r ft. Hence Eq. and hence with Eq.1 ^ ( 2 ^ „ ) 1 / 4 .2.
L2 h4 = 4/3exp[(2a — /3)/f3]. 7 With the substitutions r = exp(z — c). E' = E + gln(r0).^ ( 3 ^ + l)55p^«(3^l) + . rewrite the equation as ^ z + [ .j . J.^ +•••.. K.. in agreement with Eq.. Bose and H. *=^f.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro). We observed there in the case of the Coulomb potential that the period is proportional to n 3 .— oo < z < oo.133).318 (c) and for the Coulomb potential^ CHAPTER 15. Lifshitz [157]. MiillerKirsten [29]. and there Eq. (15. q = 2n + l. S. = 1(1 + 1). U(z) =  dz zf3e2a/?e2*.e. n = 0. M. (15.46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14." 1 11 L . W. ip = exp[(z — c)/2]. .g ^£+(<*Pl»riy = 0.46) and the classical Kepler period of Eq. . i. Power Potentials *(0) 2 oc . Solution: Details can be found in the literature.2. Landau and E. Example 15.I l n f ^ p . . (11. {S = ^ > 0. 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 . (14. Eq = i q . D.5..75). = (I + 1/2) 2 .1.L 2 + U{z)]<f>.c = —a//3.
Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . so that like charges are repelled and unlike charges are attracted by each other.1 Introductory Remarks We observed previously that the infinite range of the Coulomb potential — the concept of range being defined more precisely below — leads to a scattering phase with a logarithmr contribution. it will always notice the latter's presence. 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. 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. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. and in fact. This effect sounds unphysical. This implies. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. A very similar potential is the Yukawa potential V{r) = g2 — . it is. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state.Chapter 16 Screened Coulomb Potentials 16. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. It is clear that for ro — oo the > potential becomes of Coulomb type. In reality. r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units.
ii k . Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. and that. from the relation Pi + P 2 + M2c2 = 0. in fact the parameter [i represents the mass of such a spinless meson. 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. Since. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. the mathematical expression with an exponential is not so easy to handle analytically.e. (16. i. all coefficients Mj are real and independent of the energy E = k2.3) where for the real potentials we consider here.1) We encountered Green's functions earlier. What is meant is that the fourdimensional Minkowskian Fourier transform of the propagator is given by the relation dk (2TT) J 4 e x ^r—2 = 7Z^u o)^n k + fi2 4TT v y x 2 ikx ^ e^W ( 16 .* The relativistic equation of motion of such a meson when free is the KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x.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. . Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). one frequently resorts in calculations to the expansion of the potential in rising powers of r. (16.320 CHAPTER 16. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory.2 ) or (p = hk) 1 /" . A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. Heisenberg [133]. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. however.
Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. . Kinoshita [21]. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. Frautschi [97] and E. (16.{l.16.r)=0. (16.1.e.5) where K — ik and A n is an expansion in descending powers of K. . They are usually written I = an(E). S. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential.k. i. Potentials expandable as in Eq. . 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. Omnes and M.Q. T. Masson [180].mo being the reduced mass of the system.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. 2 . "These were first considered by C. f H. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R.3) have been considered by various authors. bound state eigenenergies and the 5matrix. i. Squires [258]. A. l+n+l = ^P. Lovelace and D. . 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. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. to be of the order of 1/k. Wu [47].1 in the above. J. only for the cases of n = 0.r).3) in the radial Schrodinger equation for the partial wave ip(l.e. Cheng and T.§ 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. S S. (16.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.e. C. if>oo. Bethe and T.e. however. Mandelstam [185]. 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. (16. physical) values of angular momentum as functions of energy E. See also the other references below.4) where E = k2 and h = c = 1 = 2m. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.k. Froissart [223]. i. £+**PVV t/.
starting with the power r _ 1 of the Coulomb potential.t Considering such superpositions of Yukawa potentials with /•oo / a(/j.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.19).4) to z = —IKv and set TP(l. Screened Coulomb Potentials as a function of the energy.k.z) Then x is a solution of the equation VaX= (16. in the present case at fc2 = 0 or k2 — Mi = 0. 'In the case of the Coulomb potential (cf. as may be seen from Eqs.z) = ez2/2zl+1X(l. J.9) the cut starts at E = k2 = 0. 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. Muller [201] and [202]. tf H . Squires [258]. < oo for all 0 < vr/2. A. < const.k. In the following we follow mainly the last two of these references.)/j. Regge [31] and E. .6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. one can assume an expansion of the potential V(r) in ascending powers of r. Miiller and K.ndp. o rV(r) I dpp\V{peie)\ regular at r = 0. J. Section 11. M. Longoni and T.322 CHAPTER 16. for all n. Muller [204]. J. W. (16. H. Bottino. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. *See in particular A. Under these conditions the S'matrix is meromorphic (i. 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. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r.e. W. 16. (16. J.15) and (16. W.** The energy is later obtained by reversion of the resulting series. Schilcher [203].
J.j)$(a + j). so that to leading order we have VaX{0) = 0.a+ l)$(a + 1) + (a.11b) ^ The associated boundary conditions are: So(a.1).i 0) = 0.a + j)Smi(a.awhere (a.0) = 1. which means that the hypergeometric series breaks off after a finite number of terms. W. (16.j) may be computed from a recurrence relation which follows from the coefficients (16.* Recurrence relations for coefficients of perturbation expansions for Yukawa.9) (16.5) is equivalent to a = —n.j) (16.a — l) — a — l. Sharma and H.2 Regge Trajectories where d2 Va = z^ and (cf. a + 1) = a — b+ 1. MiillerKirsten [249]. {0) X = Ha. anharmonic and cosine potentials have been derived in L. as in the case of the Coulomb problem.a)${a) + {a.b. K. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).j +(a + j + l. We also observe that the ansatz (16. Eq.j) for \j\ > m are zero.7) is seen to be of order 1/K.j) = (a + j l. (16. and all Sm(a. a) —b — 2a.10) zm$(a)= ^ j=m Sm(a.j l) + (a + j.a + + l). (16.5)) a = + 323 . (a. all other So(a. The function <&(a.z) = Ha).z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. By a repeated application of the recurrence relation (16. (16.b.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.9) we obtain m l)$(a .11a) The coefficients Sm(a. (a. where $(o.16.a + j)Smi(a.10): Sm(a. . j)Smi(a.
oo _ . (16. We first cite the final result .aj 2 \ [a.l.a + l ] 2 [ . #(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.13) one obtains the quantity An(K) and hence with Eq. when n = 0. of course. (16.' — . [a.a\4. and hence Va$(a + n) = ra$(a + n).a]l^a) + ^2(2K)i+1 Y.7).[ a + l. . (16.[a>a + J]i+i*(a + 3). 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). so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . (16.An{K).13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion.aj 2 (2K)A _ [a.5) the Regge trajectories I = ln(K).a + j]i+l = MiSi(a. This follows from the fact that D a $(a) = 0.1] 2 . a . Va+n = Va . 0 < \j\ < i. + (16.j).324 CHAPTER 16. 1 . i R »] = 2K[a. Evaluating the first few terms of the expansion (16. (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.e.<&(a + n) on the right hand side of Eq.o]i = M 0 .12a) where [a. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. [a. Any term /j. the latter can be written .n. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. i. for the coefficients M^> of the expansion Details can be found in the references cited above. [ a .
. P.1) + 3M3M02 + M22n(n + 1) (16.^ 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. However.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . Thus the expected appearance is that shown in Fig.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .1 by evaluating the first two terms.5).1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] .l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n .2. .e. 16. Inserting this expansion into Eq. i. including an exploration of the domain of small energies E.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . Ahmadzadeh.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n . we obtain the expansion for the Regge poles. Obviously the expansion — which. Tate [6]. (16.1.^ p [ 3 M 4 ( n . (16.( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n .§ This is an important observation which indicates the equivalence of the methods. With numerical methods one can achieve more. G. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. Burke and C. 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. of course. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.16.
Hence Si (o. ai is small and A6 is large. C. [a. . where the imaginary part of I is very small.13). through which the particle orbits during the course of the resonance. For a resonance with a long lifetime. 114. and the angle A#. and the lifetime At of the resonance satisfies the relation TAt ~ h. p. The conjugate variable to energy is time. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. Example 16. In fact.1). just as a decay width T represents the width of the resonance in energy. 0).—" + In = — .O) = Other terms vanish since So (a. (a.a) = b . this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. so the imaginary part ai of I — an represents its width in angular momentum.326 CHAPTER 16. (16. satisfies the relation ajAO ~ h.a)So(a. 16. 16. a] 3 = M 2 S 2 (a.2 a = In See S.^ .0) = (a. However." Re I 1 0 1 2 Fig. 0) and Prom Eq.11b) we obtain Si(a. For a bound state a j = 0 and the orbit becomes permanent. ±1) = 0. From Eq. Frautschi [97].0).12b) we obtain [a.1 Typical Regge trajectory for a strongly attractive Yukawa potential. one can expect a resonance with the lifetime determined by this imaginary part. Similarly the conjugate variable to angular momentum is angle. Solution: We evaluate the first three terms of expansion (16. a] 2 = M i Si (a.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. (16. .
al)S0(a. a)Si (a.a)5o(a. . One sets z = 1/r so that. . Sec. (16.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . We obtain the quantities S i ( a .1) Si(a.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . i ) again from Eq. (n+l)U+^pJ .2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6. Solution: Ignoring higher order terms in r. 0) + (a + 1. Thus (with terms which are 0): Si(a. using the Cauchy formula (6.65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 . 1) Hence 52(a.1.2 Regge Trajectories 327 Analogously we evaluate S2(a.16. 14.^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .6 + l)a + A. 1) (a . 1).n ( n + l Inserting these expressions into expansion (16. 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 . we have to evaluate (cf. (a.13) we obtain 0 (Mo .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.a + l)S0(a.0)+ 0 + 0 = (a1).2o)Si(a.11b). 0) + a S i ( a .2 a ) .b)(a .65) 2m dz V'A + 2Bz + Cz2 • Z— 0. a)Si (a. A" An An A" = = = (a.1) + (a . .1 ) + (6 . a)Si (a.6)Si(a.0) + 0 + 0 = ( a .0) Si(a.1 ) + (a.An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .6 + 1 ) .0) = (a .0) (a . (a.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity".0) + 0 + 0 = ( 6 .
5) together with the expansion (16. I. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. Solution: For details of the solution we refer to papers of Boukema.e.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. Singh [252]. The turning points in this case are at r = 0 and r = oo. Boukema [34] and [35]. Example 16. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential..1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ . We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.328 Thus here dr K' CHAPTER 16. t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. we can use Eq.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6). n = 0. i.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16.3: Calculation of Regge trajectories by the WKB method Use the WKB method to verify the first two terms in the expansion of A „ for the Yukawa Regge trajectories. ±2HK in agreement with our expressions above. *J. Screened Coulomb Potentials Mo . This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .15). however. Here we do not perform this procedure. by expanding the right hand side in rising powers of v and evaluating the individual integrals.t First. (16. tSee Chapter 11 and V.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z .* 16. .l. (16.2.3 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. 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 .
§ See H. n = 0.. This has been done§ and one obtains . H.19) +2M 3 M 0 (3n 2 + 3 n .. M 0 2y/K2 + Mi n(n + l)M 2 4 ( # 2 + Mi) 3 /2 (2ra + 1)M 0 M 2 8(K 2 + Mi) 2 + 16(if2 + Mi) 5 / 2 3M 4 (n . J.16.19) we obtain the energy. Miiller [211]. We observe also that the S'matrix is unitary as a consequence of the result (16. for instance.18) of the Smatrix now to explore further aspects. W.e. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16.4 The Energy Expansion It may have been noticed that in the above considerations we could have combined the constant Mi with the energy into a combination K2 + M\.l. Miiller [204]. .* 16. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16..l (two additional terms are given in the literature). the behaviour of the scattering phase in the domain of high energies.l)n(n + l)(n + 2) + 2M2M02 .17) and is therefore real for real values of the potential coefficients Mj.g. such as. Paralleling the case of the Coulomb potential in Chapter 11. i. One can use the explicit expression (16. J. 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. . which are precisely the expressions yielding the Regge trajectories of above.14). the following *See e.2. Now reversing the expansion (16. W.17).0[(KZ + Mx)'6} (16.
Zauderer [286]. C. 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) ^ — . W.11) + 2M22M03(9n2 + 9n .1) + 2M2M. Watson [280] in 1918 and later resurrected by A.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. 107.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n .< 3 M 4 M 0 ( n . Frautschi [97].10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . Sommerfeld [256].2)(n . the transformation was introduced in the form given below by G. Large coupling expansions derived in H. p. Iafrate and L.1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .l)n(n + l)(n + 2) . VahediFaridi [212] contain as special case the expansions of the first pair of authors. J.* We recall from Chapter 11 the definition of the scattering amplitude F(9. B. .l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). Screened Coulomb Potentials expansion. E. A.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. in the following expression with an "See in particular G. 'According to S. i.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 . p. J.330 CHAPTER 16. Mendelsohn [135] and E. Warburton [279]. N.^ 16.5 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. 282. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A.e. MiiilerKirsten and N.
2 The integration contour C.k) = 2zfc L " v "'' v .1.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.k) = ^T(2l + !)/(*.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. Actually 5j lrr/2 as in Eqs.''' The idea is now to consider each term of the expansion (16. \x\ <C l .21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.24) Si(k) being the phase shift.187).184) to (11.22) = \F(9.25) Setting Note the minus sign in the argument of the Legendre function.k) — e 2i5[(k) ~" p—iirl (16. For this purpose we consider the following integral taken along the contour C shown in Fig. 16.fe)P . I = n + x. Eq. .k) 0ikr (16.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. c = 1.23) f(l.26.. (11. n = 0.iJ 11 —[S(l. 16. z=o in which (cf.kmcose)..k)\2.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. (11. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n)..2. ~:' sin nl (16. The scattering amplitude possesses the partial wave expansion F(0.k)~l] = ~ k" PiSiW sin 5i(k) and S(l..187)) (16. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig. (cos 9).
n) oo {l)1 Pt{cosO) v ' (16. n=0 which is the usual partial wave expansion.3. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.28) . Consider the reaction of a particle a{p) with fourmomentum p^ colliding with a particle b(q) having fourmomentum q^ and producing a particle c(p') and a particle d(q'). For the momenta indicated we set (with metric +.3 Reaction channels and their respective Mandelstam variables. In order to see the relevance of Regge trajectories in a reaction of particles. t = (qq')2. 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. we obtain m® = ± I dl^±^f(l.27) a(p) © b(q) Fig.26) P((+COS0) = 2 ( 2 n + l)/(n. Thus the direct reaction may. a + c —> b + d. a + d — • c + b.332 CHAPTER 16. u={pq')\ (16. in which case c = a and d = b.k) Zi Jc TT{L . 16.—.—) s = (p + q)2. d(q') (16. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem.—. 16.3. for instance. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. 16. be elastic scattering of a off b. as indicated in Fig.fc)P„(cos0).
2 ^ ( 1 + cos6 s ).t) = y^Fs(s. F.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s.32) t = 4(q? + ml). S.(<fe)^(cos 6t).t). .g. the only kinematics we require in the following is given by the relations (e. the variables t and u would describe momentum transfers in the crossed channels. Khuri and S.The variables s. one then has A(s.t. the dynamics is described by the exchange of quantum numbers. except for poles and cuts which characterize the reactions in the three channels. the momentum k there would correspond to qs) s = ~4(ql + ml). A(s. there will be several such amplitudes which together describe all three processes).16. one can write the amplitude as depending only on two. 16.u are not independent. A(s. t = 2q£(i . In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). One now makes two hypotheses.u) (if any of the external particles has nonzero spin.^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories.u are known as Mandelstam variables.t.g.9s)\2.c o s 0 s ) . Ft = £ ( 2 J + l)F.31) i l Taking all particles to have the same mass mo. B. etc.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s.t. which is an analytic function of the variables s. if s describes the square of the total energy in the s channel. Blankenbecler.t) = VtFt(t. In a Regge theory on the other hand.§ Since the three variables s. e. if we had been considering the s reaction originally. N. Chew [48].et) (16.. u.0s). Thus.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo. u = .g. § See G. T See e. Mandelstam [184] and R.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). Goldberger. s = 2qf(lcos0t)(16. t. (16. the quantities carrying these quantum numbers are the Regge trajectories. L. For a Lorentzinvariant normalization of the initial and final states. (16. Symbolically we then have the situation shown in Fig. M.4. Treiman [25]. N. (a) Mandelstam's hypothesis: All three processes (described by the s.t.
However. 16.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. Screened Coulomb Potentials Regge oc(t) Fig.)e~ •fj.4 Yukawa versus Regge theory.5 The Lehmann ellipse. 16. the Legendre polynomial Pj(cos 9) does not possess a cut.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. . It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^.e. But for integral values of I. So what can one do? 1 H.334 Yukawa CHAPTER 16. For s — oo : cos 9t = 1 + s/2q^ — oo. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. 16. the partial wave expansion is not valid in the physical region of the (i. in the domain of schannel physical values of the kinematical of the ichannel). But > > wave expansion X)(2Z + l)F.5 for Yukawa potentials V{r) u: lm cose dr a(/j. Lehmann [162].
^ B 1 A Fig.6 The integration contour C".e. Moreover.6.33) .t).26) but for a different choice of the contour. 16. contour C / / D ^ ^ .e.5 The SommerfeldWatson Transform 335 T h e answer is. 16. k) have to the right potentials one can show t h a t the integral along the curved porA' of the circle at infinity tends to zero. i. B Iml 1 \ i All \ r\ 0 / closed . and there is only a finite number N + 1 in the domain of 9W = . all poles of Ql > 0.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. to use the SommerfeldWatson transform. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. the contour representation (16. N F(6. Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.16. „ i.1 / 2 . Thus consider now the same integral but taken along the closed contour C shown in Fig. sin nl [16.
32)) and finite. we have (16. W. Apart from more refined details. Regge [229]. In particular. Regge trajectories have. as we discussed briefly in the above. 42. Bose and H.e. 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.e. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo.336 CHAPTER 16. the amplitude can be represented as a sum over ^channel Regge poles. . of course. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. K. i. Omnes and M. p. It follows that under these conditions. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. Limic [177] and E. for high energies in • > the schannel). (16.t)&B(t)sa°V. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. Foissart [223].35) This result is valid for s — oo and t negative (cf. Predazzi and T. J. if ao(t) is the Regge pole with largest real part. this behaviour has been confirmed in high energy hadronic reactions. i.6 Concluding Remarks The potentials we considered above have wide application under the name of Yukawa potentials in nuclear physics and as screened Coulomb potentials in atomic physics. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. for plots and discussion see R. 16.e. at high schannel energies. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. Eq. MiillerKirsten [29]. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics.
MiillerKirsten [209]. The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. Singularities appear also in the form of cuts in the plane of complex angular momentum. and also as socalled "fixed poles" . The way to do this is similar to calculations in Chapters 18 and 20. and is composed of quarks.g. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum.16. For an overview see e. . The above treatment of screened Coulomb potentials is incomplete.6 Concluding Remarks 337 important parameter in string theory. W. 20.** Whereas the former are related to absorptive properties of a reaction. a composite particle being one with structure. called "Regge cuts". like a meson or a baryon. the latter are related to the distinction between "' elementary and "composite particles".1. H. Related aspects are discussed in Sec. J.
This page is intentionally left blank .
Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. and are themselves periodic with period e. Arscott [11]. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. example 14. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type. See F. motivated mainly by the regularities of the crystal structure of matter. 339 . 19 and p.0 < k2 < 1. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. The reason for these difficulties is. In essence these periodic Jacobian elliptic functions. however. depending on the parameter k.* These elliptic equations involve Jacobian elliptic functions like sn(a. M.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. called elliptic modulus.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. The Schrodinger equation with this potential.g.Chapter 17 Periodic Potentials 17. are not much harder to handle than trigonometric functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. 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.) which are functions that interpolate between the n. separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. the most immediate candidate is a trigonometric form like that of the the cosine function. p. K being the complete elliptic integral of the first kind. 25. 2K or AK. since many analogous relations hold.
the three Jacobian elliptic functions s n z . and to relate these to the weak coupling bands or regions of stability. Dingle and H. i. 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. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. Like the trigonometric functions sine. In fact. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. those of integral order. Muller [73]. c n z and d n z are handled with analogous formulas. W. Periodic Potentials vide periodic potentials like trigonometric functions. . Miiller. W. [205]. or Sw&ve Schrodinger equation with elliptic potential. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc.7.1. We might mention already here the much less familiar but more general Lame equation. cosine and tangent. ''The reader who encounters these Jacobian elliptic functions here for the first time. need not be afraid of them.340 CHAPTER 17.2h2 cos 2z]y = 0. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A . J. The most important formulas for our purposes here are collected in Appendix A. it is this difference which implies in the case of the Mathieu equation the existence of the parameter function v called Floquet exponent^ The Mathieu equation can be obtained from the Lame equation in the limit of n — oo. J..4. Lame equation: H. each with its own characteristic eigenvalue.1) where h2 is a parameter and — IT < z < IT. This is a significant difference compared with the Mathieu equation. The first of these conditions is already tMathieu equation: R. The Mathieu equation. The cosine potential of the Mathieu equation can therefore more precisely be described as Mathieu or trigonometric potential and the potential sn 2 (x) of the Lame equation as Lame or elliptic potential. which is conveniently written ^ + [\2K2sn2z]y = 0. the Lame equation possesses 2 n + l polynomial solutions. (17.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. 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. which does not possess any polynomial solutions. B.e. which one looks up in books like that of MilneThomson [196] or Tables when required.1. (17. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. For large coupling. 8.2 . 17.. Both of these domains are important for a host of other considerations.
1. 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. Adding the negative term — 2h2 cos 2z to A. In classical mechanics stability is treated for instance in connection with planetary motion. i. 17. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.e. since the solution there is of the form cos vAz. we compare the For interesting related discussions see M. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. as in Fig. we see that in a plot of h2 versus A.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials.17. a bounded function. Salem and T. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms.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. . Looking at the Mathieu equation for h? —> 0. Vachaspati [242]. 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).e. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). i. these concepts are rarely explained as such there.e.2. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability.^ 17. h2 = 0 belongs to the domain of stability. the case of h2 small.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. the semiaxis A > 0. Although the mathematics literature on the subject uses the physical concepts of stability and instability. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. y" + \y = 0(h2). 17. i.e. i.
2/i2 cos 2(2 + vr)]y(z + vr) = 0. (17.e. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. we obtain y(vr) y(2?r) a = — — a = —^^r. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. TT. y~(z) = 6sin vXz. sin V Az or linear combinations.2h2 cos 2z]y(z) = 0. ° ^ [1 + 0(h2)}.342 CHAPTER 17. 2/(0)' yfr) . — • . a = const.e. (17. 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). (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). Periodic Potentials equations ^ p + [A . Setting z = 0. y+{z) .B and a.3) The parameter a is therefore the eigenvalue of the operator TV Our first objective is its determination which amounts to the determination of the parameter called Floquet exponent below. y(2vr) 2 a = —7—^ = 1.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). d(z + TT) cos 2z We see that there are solutions which are proportional. i. i. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. d2y(z + 7r + [A .
a2b2\ a± ~ 2a6\/A . [y+(v) .2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. y'_(Q) = bV\ with Wronskian W[y+. y{z) = aty+(z) + 0y(z).y+(7r)j/_(7r)} = 0.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. y'(z + n) = a[ay'+(z) + py'_(z)}.4). the determinant of the matrix must vanish. (17. \y+>y\ = y+{ir)y'{ir) . The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. 7 y\z) = ay'+(z) + (3y'_{z).y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. we choose these with the following set of boundary conditions: y+(0) = a. This expression is equal to the coefficient of a2 in the quadratic equation for a. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. yV(0) = 0.y+(7r)y_(7r) = 0 or abVXa2 . 2/'(7r) = a ^ W + ^ . whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. 2/_(0) = 0.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.{ay'_{ir) + bV\y+(ir)}a Now from Eq. (17. We can write therefore. and by'+(?r) = —ay/Xy{n). Using Eq. This means. i. with constants a and /3.aa][y'_{n) .e. Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions.17. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .3) we obtain y(z + ir) = a[ay+(z) + @y(z)].oWX] .
<7__cr_ = 1. 8.e. 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. cr+ = 1/<T_.2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .e. Periodic Potentials g= 2aby/\. i.1. (5i/ 2 + 7)/i 8 vz + — .344 Setting / = 2by/\y+(Tr). i..= — . Prom this condition we determine later (cf.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). <T__ + <T_ = 2y + (vr) . this implies y+(Tr)/a = ± 1 .0 This series may be reversed t o yield the Floquet exponent.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. This calculation is demonstrated in Example 17. (17. (17. (17. Thus this Floquet exponent is determined by the value at z = K of t h e solution which is even around z = 0 and is independent of the normalization constant a of t h e even solution.5) a The parameter i/ is known as Floquet exponent.8) "Observe that for integral values of v (in lowest order of h2).22)) the boundary conditions of periodic solutions.7. (17. T h e result for A is the expansion A = h4 2 . Eq. Setting cr+ = e . we have <7± — CHAPTER 17. _ / ± v 7 ^ V .6) We can derive various terms of this expansion perturbatively with the method of Sec. ** An important consequence of Eq. 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 . **This point is not immediately clear from mathematics literature. .455A2 + 1291A . where the constants are usually taken as unity from the beginning. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 .+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. .
(17. = J—^ + o(hr). With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. W.11) or smvz or e±ivz.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 .3). These cases therefore have to be dealt with separately. Example 17. It follows that in these cases i? = n2[l + 0(h2)]. .7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. Meixner and F.2 how the expansions for small values of h2 may be obtained perturbatively.„„ x n.17. (17.35A + 15A2)/l8 /. (17.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. (17.35A + 8 .9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A . (17. 124.jl2. as will be done later. we must have a2 — 1 and hence e 2iv/A7r = 1} i.1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec.e. cos TT\/A 32A(A1) 2 . (17. Solution: We write the eigenvalue equation A = v2.10) A = n2.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. Schiifke [193]. n an integer. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. For these integral cases of v we demonstrate in Example 17.5) and (17. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.1) 15A2 . p.12) S e e J. 8. VA = ±n. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.
Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z. W. for instance.14) therefore leads to the following nextto leading order contribution (v. (17.16) *These smallh expansions are convergent.(2) + . of course.11). h2(u. and (v. as shown.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. .!/) = 2A We now observe that DvVv = 0. .h2). v .e.2)j/„_ 2 + {v. where except.3. see Example 17. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.2) which applies when v is nonintegral.e.346 CHAPTER 17. ("• ^ + 2) (o) 2(2^ . 17. Schafke [193].v±2) = l.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. (17. i.16) directly into the original equation in the form of Eq. (17.3 and 19. se and me as* oo i y = „«>) + yW + i. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 .11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. v)yu + {v.v + 2)yu+2]. (17. up to 0(h2) we have A = 0. The right hand side of Eq.„ + £ h2i £ P2i(2j)yv+j (17.14) = = = cos(y+ 2)z + cos{y— 2)z.e.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. by J. Proceeding in this manner we obtain the solutions ce. = . i. since j / 0 ' = yv leaves unaccounted R\. Meixner and F.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v. siniy + 2)z + sin(i/ — 2)z. v . h2). We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration. when a = 0 or 2v + a = 0.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. e ± i ^ + 2 ' 2 + e±i("2)z. '. With our perturbation formalism — as demonstrated in cases considered in Sees.. (17. Next we treat terms yv+a in j / 1 ' in a similar way. v) = 0.15) Hence a term fMy„+a on the right hand side of Eq. i. sev(z.e. or A = 0(/i 4 ). We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. (17.2) _ __ _ (y. (17.h2). (i/. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .2. The recurrence relation can also be obtained by substituting the right hand side of Eq. me±v(z. .
14).2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. * Note. Example 17. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. 2 ( 2 i / .v)\ . . fo^ + 2) .u) + h4 (". We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• .R{ . The higher order terms naturally require a little more algebra. v) ' ( 2 v + 2) (17..e. the solutions are not yet normalized..1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3].."2).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)+. In the case of integral values of i>.17) Inserting 1 for the stepcoefficients of Eq. (17. (17.g. . 121.2: 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. Meixner and F. „ .1 with A = v2 — 2h2A but with v = 1.7). J. 0 = h2(u. .17.2 ) v(y2. one has to deal with each integral case separately as in the following example.18) Again applying Eq.. W. Schafke [193].2—r——— (i/ + 2. i. For v = 1 we then have yi = cosz = y~\. + y~i) (17. thus determining the quantity A. Solution: We start as in Example 17. See e. however. . p. (17. Following the first few arguments of Example 17.. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero.
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]. p. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. We introduce a normalization constant c and rename the normalized solution y then yc. 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 (17.348 From this we obtain 2h2A = h2 CHAPTER 17. h2). the case of A an integer. (17.24c) below) . The reverse situation is later of importance.* The associated solution is ym+ym. Schafke [193]. Periodic Potentials h4 r ••• . see below!) one obtains the additional terms given by Meixner and Schafke [193].19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). .n integers. W.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p.e. i. 120. We consider this case again in an example. 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. Meixner and F. With normalization (not to 1. 123) (there the solution is called cei(z. 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 . h4 (17. Eq. and v is to be found in ascending powers of h2. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . m. J. In the above we considered the case of v an integer and calculated the eigenvalue A. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z.
(17. S.1) .1) (A . we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i .9b) about A = 4.17.• 1 + 5n h 2 s 2932 Setting in cosm.9b) v — 2 + <5. Liang and Yunbo Zhang [189]. ^S. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above.9b) and considering the approach A — 4 gives > (A4)2 + .3 1 A + 8) ( A . 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 . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq. o 1• • • . Manvelyan. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + . Hashimoto [124]. fe47T2(A4) 4%/A(A .. MiillerKirsten. (11A2 . Gubser and A. . . H. on the left hand side of Eq.2 Cosine Potential: Weak Coupling Solutions Example 17. J.l ) 2 (A ..= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • .('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 . W. Substituting these expressions into Eq.4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . (17. J .1)2/A TT2 +h8 (15A2 . (17.4 ) + .3 5 A + 8)TT2(A4) _ 3 / / 64(A .Q .3 1 x 4 + 8) 274233 (.4)Av A2v A ~ 32A(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.
2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. M.2.w^. (17. for instance. "These conditions. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. 17. These are defined by specific boundary conditions.21b). in F.6.W4W=UtWSta±I(. i. may also be found. Arscott [11].350 CHAPTER 17. = and s+w 2tW±I. solutions for integral values of v.. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. 17. We now derive these boundary conditions. Taking in particular Eqs. 28 .2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir.4) into account. as also conditions (17. . Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. bV\ y+(0) 2/40) 07.10) applies. Fig.1 Boundaries of domains of stability.e. (17. pp. We verify these for the large/i 2 solutions in Example 17.29.21a) and (17.
f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . . .4(f) (17. 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.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. /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 . /7r\j/!_(7r) y L l 7TJ . Then y+ K 1 _ 2/+00 7T y V + (f)[i + 2/±Ml ^ ] y'il) 2y_(f )i/+(f.21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .17. ^\y±M /vr\yL(7r) . o y+(fy(f)ob\/A _ = 1 + 2I&A/X . 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+. which after some rearrangement can be written y+(vr) _ .
(17. the case of ai with solution cei (of period 27r).2 we require this relation for nonintegral values of v. 17. in this order.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . A > b2n+i(h2).2. (17. (17. (17. A 2 y • ce2n=q0i. There the second term on the right is nonzero.2): hA 7h8 29h12 + Y l28^04.21b) must be ± 1 . Mr7nA (17. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0.' fh2 V2ceo = l ..• a2n(h2).T + . . y > se 2n +2s 9o +i The functions so defined have respectively period 7r... 19..352 CHAPTER 17.24d) v 12 ' **To avoid confusion we emphasize: In Sec. 17. and whether the function is even or odd. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • . sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .23) We can now see how the domains of stability and instability arise... was treated in detail in Example 17. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.21a) and (17. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. (17.24c) 8 h2 sin 4z \ . (17. .1). y > se2n+i=qo.1. These are defined by the following boundary conditions. (17. IT. 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.— — — • • • .24b) 8 h2 z —— cos 3z + • • • . This is shown schematically in Fig. These may be subdivided into classes depending on whether the integral value n — 0.3. of v is even or odd. We see that there are four possibilities and hence four different types of functions.22) * a2n+i(h ). 27r.+ . A *• b2n+2(h2). ce± = cos 8 64 h4 5h8 b2 = 4 1 . 2K. y > ce2n+i=qo.5) that the left hand sides of Eqs. 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 .
17„. W.17. the walls are not infinitely high and hence tunneling occurs from one well into another.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R. 353 . .z3 ce2 = cos 2z . of course. The cosine potential cos 2z depicted in Fig. 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. Thus in Fig. In the case of the cosine potential.3. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator.3 17. 17.> (17. 17.24e) 17.2 The cosine potential.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . 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. We can see from the inverse of Fig.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. 13824 .2cos4. 17.1 that the boundaries of regions of stability merge to lines.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials.hz — 12 . . 17.3 we show schematically how the low energy solutions for integral values of the Floquet exponent are related to the asymptotically degenerate harmonic oscillator eigenvalues. Miiller [73]. B. 17. Dingle and H. .
simply the onedimensional Schrodinger equation for the harmonic oscillator.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .2/i cos z V = 0.2.3. Periodic Potentials minima at z — ±ir/2.2 and 8.26) 17. Thus in this case the quantity on the left is not an exact odd integer.. 4 (17. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. We therefore wish to expand it about these points. Ion (17. 8. where A / 8 is a remainder.3) and is. Sees. but only approximately so.27) . there is tunneling from one barrier to the next. of course. as in the case of the cosine potential.1. we therefore set 9 A A = 2h2 +2hq + —. n 0. Taking the remaining terms of the cosine expansion into account. (17.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. In the case of finite heights of the potential barriers.354 CHAPTER 17.. and we set this equal to q. (17.i r ) ~ 1 — 21 z ± 7r Changing the independent variable now to the equation is approximated by A + 2/t2 dx2 + 4/i 2T y = 0.2 The solutions We insert Eq...
27) we )A = 0.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. 16/i/ (17.17. 17.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . another solution. 2 (17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig. can be .29) is known. and substituting this into Eq.3 Schematic picture of Mathieu equation eigenvalues. the linearly independent one. z). (17. (17.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z). Choosing y(z) = A(z)e2hsinz.31) We make the important observation here t h a t if one solution of the type (17.30) D A): COS « = 4z + 1 iq sin 2. (17.
e. .356 CHAPTER 17. (17.9) (9.g + 4) . Aq+4 2Jh .4 ] . (17. ? .30) amounting to Ri ] q = .^ 6T [(«>« + 4 ) A ? + 4 + (Q.9) = 0. The solution Aq(z) of the first order equation Dq Aq(z) — 0 is c o s ^ 1 ) ( \K+\Z Aq{z) = sml(9+i) ( ivr + \ z ) COS 2 9 ( D I\n+\Z (17.32) cos^'+i) [\K .94) = (9 + l)(9 + 3).35) not yet taken care of is set equal to zero._ 4 . (9. 1 + 4 ) ^ + 4 + (^. 2 A A = («. (17.4 ) ^ . 2[(92 + l) + A]. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.9 + 4) = (9.4)A. when i = 0 — the terms not involving Aq in (17.\z To lowest order the solution A is A^> = Aq. With some algebra one finds that 16^ + where (9. (17. ^ ) A + ( ? . D{q%Aq+4i = 0.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.36) Then ^4. (9l)(93).Q)Ag + ( ? .35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. 2 /i D^Ag = 0.35) But since and so a term iiAq+n on the right hand side of Eq.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except.33) (17. i. (9.94) \ — Aq4 (17. of course. ? .
4. h) = A(z. h). A(z.39) which is the equation from which the quantity A and hence the eigenvalue A is determined. Q .g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. +4.4) (q .g4) 4. (17.. . h). Thus in particular A(z. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q._ 8 (17. q . In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. (q. q . q.)+ & ! = « ( . h) = A(z. the sum of the terms in Aq in Eqs. q.4) (q . We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. except those involving Aq..q 8)Aq+8 (o)' ^±V(q 1 4. A.38) Aq8 + 1 2 Clearly we can continue in this way.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation. q. Then for A = A(0)+A(D+A(2) + .g + 4)(g + 4. Thus 0 = (g.q + 4)(q + 4. .4 .30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ . q.37) and so on — left uncompensated — must vanish. (q. to satisfy Eq.40) . an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.g + . q (17.37) Here all the terms.q 4)Aq+4 q + 4 1 q + + + + + <^(. . (17.4.17. (17..4) 1 1 (q.<J+4 Aq+& + 1 2 1 . (17.q + 4) (g.35) and (17. q .8)A. ) ^ 4.
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 ." 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. (17. (6. i. (17.43) 9 2eh . this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. 2 w2 + H* (17.) i ( Q . 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.d2B w dw2 + w(l w dB_ dw d dw 1.46) 2s(«. Note different definitions in use. The differential equation there is Eq.67).44) where D(B) 9 = dw2 $_ !)• (17. Before we study the eigenvalue equation in more detail.42) (17. and discussed there. . We return to Eqs.41) so that the early successive contributions decrease in magnitude.e. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). We observe that for the large values of h we are considering here.45) {B) We recognize Dq as the differential operator of the Hermite equation. it is found that D(B) z+ — ]B = 0. 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.i ) We encountered Hermite polynomials earlier in Chapter 6.358 CHAPTER 17. where Hen{w) is a Hermite polynomial when n is an integer.
i. Finally we obtain a third pair of solutions. Comparison with the case of solution A now shows that the form of that solution can be taken over here except that everywhere Aq{z) has to be replaced by Bq(w).e.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. (17. q)Bq + (q.34).47) where the coefficients are the same as before. with the equation C" + 4h cos z C + 2h ( q . forms a rapidly decreasing expansion provided that \w (z)\ « Vh.51) . A{wz (17.7 T H 1 2 Z 4 <1. . (17.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. i..17. The solution with contributions B^°\B^>.e. 16hJ (17.48) This shows that this solution is valid in particular around z = ir/2. also the same coefficients in the case of functions C below.e. 1 COS I . the expansion for the quantity A being identical with that of the previous case. y — Cexp(2hsinz) and Cexp(—2/isinz).50) C = 0. i. as in Eq..3 ) \He* ! (9+1) H. (17. i.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17. which is an amazing and unique property of the Mathieu solutions — cf.d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. Again a change in the sign of z throughout yields the associated Mathieu function y = i?exp(—2hsinz) with B[w(z)] = B[w(—z)] and domain of validity in particular around z = —ir/2..q + 4 ) £ g + 4 + (q.qi)Bq_4.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .e.
A closer look at that expansion is our next objective.(w2 + A)Hen. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged. (17. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. q)Cq + (q.53) This solution is valid in particular around z = —7r/2. (17. 2' the remainder 72. ie 1 COS I . These domains of validity are indicated in Fig.h).52) and the coefficients are again the same as before.4: Hermite function linearization of perturbation Using the recurrence relations (6.q. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). 17.360 CHAPTER 17. (17.4. (9+3) (2n 2 + 2n + 1) + . He'n(w) = wHen(w) Hen+i(w). show that the perturbation remainder of Eq.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .q. We note that C(z. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq.44) can be linearized with the same coefficients as in the case of the trigonometric solutions. (17.n2(n l)Hen2 . The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i.59d) of Hermite functions Hen(w). can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. 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.h) ocB{z.7 T 1 2 Z 4 <1.44). (17. 2 J where n=(ql).A Hen . q + 4)C 9 + 4 + (q.4)C g _ 4 .54) where the proportionality factor is complex. We have thus determined three pairs of solutions of the Mathieu equation.w2)He'n . each of them associated with one and the same expansion for the eigenvalue. . Example 17.q.
3 T h e eigenvalues We saw above that along with each solution the following expansion results: 2A = M 1 + ^ M where Mx = 2(q2 + l). (q.3. the solution y = exp(2/i sin 2)^(2).57) . (q. (q.q + A){q + A.g) 1 1 and so on. 2 + ^ M 3 + (17.' 361 in 2*f*1>[i(gl)]! = (q. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.q4) (17.0 and —4.17. q .55) M2 = M. ) . 2)Aq_8} + • (17.i ) ( w ) 2l(.4) = (q . q + 4) = (g + 1)(? + 3). We can write the solutions obtained previously.)Bqi(w). .1i .3).g4)(q4.l)[l(q _!)].q 4. (b) The allowed moves are +4.q + 4)Bq+4(w) + (q.. ^ 1 Hv" '^ ^' ^ ' . 17. l)Ag4 + P2(q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. e. In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. q) = 2[(q2 + 1) + A]..3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . r j=r. with (. in the following compact form with coefficients Pr(q.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .q + 4) (q + 4.l)(q .g.56) (94. q)Bq(w) + (q. (?) 1 1 + (9. . no intermediate move to or from q being allowed. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.
j) = (g + 4j . and the contribution from a sequence of r moves from q + 4j back to g. Miiller [73].g + 4) 2 1 (g.j)Pr+1(g. .g + 4)(g + 4. .g + 4) + 1 1 1 and so on.4.g + 8)(g + 8. 2) 1 2 ' ' ^ ' ^ 1 (g. g + 4 j ) P r .59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. for instance.j ^ 0) = 0.j) = (l)rPr(q.34) we deduce that Pr(q. j + 1) (17.j).e.362 Clearly CHAPTER 17.58) with P 0 (g.1) + (g + 4j.P 2 ( g . (17. g + 4j)P r _i (g.j)] = 2j]jPr2(g.3) ^3(9.g + 8)(g + 8. 0) = 1 and all other P0{q. P2(q 2) Pl(g. W. x)_ (g.j). r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. J. l ) .g + 4)(g + 4.60) M2r+l = = *R. the factor j removing a duplication of factors in the denominator at the junction. p = (<?><? + 4 ) ( g + 4 >g + 8) . j) +(g + 4j + 4. i. M2 = P 2 ( g . (17.l ) . j . and in general* M2r = J]j[P2(g.j).j). Thus we have.g + 4)(g + 4. g + 4j)P r _!(g.j)^2(9.g + 4) ^3(9. Dingle and H.g + 12) 1 2 3 (g._1) = Mzil.g + 4 ) ( g + 4. Periodic Potentials PlM)=(M+i).Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j.j).g + 4)(g + 4. 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. as jP r 2 (g. From the coefficients (17.i (g. B.
. It will be seen in Chapter 18 that in the case of anharmonic oscillators. Dingle and H. q is only known to be approximately an odd integer qo = 2n + 1.17. B. as pointed out earlier. remains to be determined next.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above.* We observe that the expansion remains unchanged under the replacements q — —q.^ . 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. Its precise deviation from qo. J. 17. W.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) .284g + 57) + (17. In fact we demonstrate in Example 17. L. *E. Up to > > this point.3. Goldstein [115]. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness.92q3 + 70g2 .2 * +3) + ^ 2 (9g4 .4. where solutions overlap. as a result of tunneling between wells. of course. h — —h.( 3 ^ . 17. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig.5 the following proportionalities there: B[w(z)] = aA(z).^q(33q4 1 220 hA 1 (63g6 + 1260g4 + 2943g2 + 486) 225/i5 rg(527g6 + 15617g4 + 69001g2 + 41607) —3T76 (9387g8 + 388780g6 + 2845898g4 + 4021884g2 + 506979) —WTfs q ( 175045qS + 9702612 ^ 6 + 107798166g4 (17. Ince [136]. [137]. S. a= (8/i)3(9x) [1(91)]! 1 . Miiller [73]. [116].62a) f R . We observe that there are regions.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . t Terms up to and including those of order l//i 5 had been obtained by others and by different methods.
364 and C[w(z)] =a~A(z). a .32). and determine the constants a and a.1 ) ( 7T + . Example 17.1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . The first term of the expansion of A(z) is the function Aq(z) given by Eq.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .4 Domains of solutions and their domains of overlap.7T H Aq(z) = = C O S ^ .1 4 .62b) *z domains of overlap Fig. 17.(17.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA. Solution: We begin with the solution containing the function A{z)._ ^ l CHAPTER 17. We write and expand this function as follows: .
53) for the expansion. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. with e.284g + 57) + • The other relation is obtained similarly. (8.61) for the connection and Eq. with (8h)li''1) (3g2 .g.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec.46).2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed. A*1) given by Eq. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. we obtain Bq[w(z)} H.3) 1!(8™2) )[i(gl)]!L Inserting this into the expansion of B[w(z)] = B^ + BW f • • • where w{z) — 4 / i 1 / 2 cos I IT + z J. we have to impose the appropriate boundary conditions.36). 17. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.17. Since the solution with A is obtained from that with A by a change of sign of z.i ) 1 (g. the coefficient of the dominant factor cos' 9 _ 1 ^ / ' 2 (7r/4 + z/2) in the result is seen to be J (8/i)3 ( g .3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •.63) In Example 17. i. These are the conditions of Eqs.g + 4) (q + 4. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2. these are conditions imposed on even and odd solutions at the point z = it/2. s . As we saw there. \ql)/2(w) 2i(9D[i(. (17.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.e.92q 3 + 70q 2 .2. (17. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 . (17.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 ._i)]! w*te» 2l(91 (. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. the coefficient of the dominant factor c o s ( .22).!)(. (6.
W. Oberhettinger [181]. i.65) For the evaluation of these conditions we require the following expressions involving Hermite functions which we obtain from Tables of Special Functions:^ ^ ( =0) = 2TT2I//2 Fi(^ijii' He2n(0) = ( Ur w=0 •W+m' (17.» a \dw _ C[w = 0}c_2h a w=0 = Q (17.22) and to correlate the solutions to those for small values of h2. He'u{w) = wHev{w)He„+1{w). He2n+1 (0) = 0. 80.66). The functions defined as in Eq. i.59d).63) to the domain around z = ir/2 by using the proportionalities (17.366 CHAPTER 17.62a) and (17. a ± C[w{z)}c_2hsinz^ a (17. He'v(0) = Hev+1{0). Then y± oc ^ M e ^ i n .66) ^E. give the following expressions for polynomials with argument zero: ~19)"(^)!.e. 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.g. p. and then replacing 2n by i>. . Magnus and F. (17.66) then follows from the recurrence relation (6. Periodic Potentials We can extend the solutions (17. The second of relations (17. With the help of the duplication and inversion formulas of factorials.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. one obtains the first of relations (17.62b).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\ ! ' . Thus y+ = ce and y_ = se.
69) If the right hand side of this equation were exactly equal to zero. (17. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.136g + 9) + • • • . and using the reflection formula (8.82g2 .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17. Hence expanding the cotangent about these points.( g .7.Qo) + 0[(q .3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.67b) <f ~ 82g^ .11.! ) ] ! [ . .87 4 2 ¥w (17.68) Inserting here expansions (17.67d) Considering now the second of conditions (17.39 2~^h? (17. the solutions would be given by q = go = 3.14) 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 .67c) 4 2 \dw)w=Q C[w = 0] [\(q .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .106g . we have cot j ^ ( g ~ 1)} = ~\<(l .» 2 6 /i the condition reduces to the equation cot{ .1) 7T . f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .40g3 + 18g2 .39 (17.g0)3].67c).67a) and (17. B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .17.67a) g 106g 87 2 14 /i 2 (17. and similarly for C and its derivative.
528gg + 3307Q .. (17.40g3 + 18q2 .73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.(17. .120g£ + 467g .9. .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .70) + ^ 2 (9?o .e. i.70) except that now qo = 1.65) leads to the same result except for a change in sign.71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .5. = _a a 6 _4h (17.136g0 + 9) + • Clearly the last of conditions (17.40gg + 18gg ..65) yields the same result except for a change in sign.136q + 9) + . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .. Considering now the first of conditions (17.40gg + 18gg .408g0 + 1089) 2 /i +• (17.72) This leads again to the result (17. The remaining condition (17.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg . Thus we have 2 (16/i) qo/2 T2 4. we use \(q)~\(q0) + (qq0)(^J ..
n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.. (17.e. The tunneling effect. See also F.74b) The above results for the cosine potential are in many respects important.17. minus sign). quite apart from applications.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . The complete. (17. though approximate. . .2. In the literature that we follow here* in all cases several more higher order terms have been given.t From Eq. 120. n = 0. W.A_(g0) ^ .1. Dingle and H. i.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq.e. Arscott [11].3 we call this difference 26E„n.. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential.61). 20.. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. We see here that *R. + ^ 2 (9<?o + 89o " 78g02 . just as any solvable problem is important for the insights it offers in a concrete and transparent form. M. and not the separation of these as a result of tunneling. 9o = 2n + 1. example 3. results from the boundary conditions. The dominant contributions were also found by S. p. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. B.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.3. In the literature one finds frequently the statement that the splitting is a nonperturbative effect.3. Miiller [73]. Goldstein [116]. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.87) = Azpteo). (17. J. 17. and that for the odd Mathieu functions se qo+ i and se9o (lower.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w .88q0 . (17. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. i. *In Sec. this becomes 369 = (16/i)^+1e4h l .
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. the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft).370 CHAPTER 17. evident through the exponential factor exp(—4/i) in Eq. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . M±TT) . ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). (17. In fact the perturbation theory developed in this context and tested by application to the cosine potential is of such generality that the recurrence relation of the coefficients of the eigenvalue expansion can be looked at as a difference equation. _3/+(z) + ____2/_(z). results from the imposition of boundary conditions. As mentioned at the beginning. and one can compare the methods. The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential. We consider for simplicity only the dominant contributions and replace throughout —1 to some power by exp(i7r) to that power and combine such terms into cosines and sines. From Eq. The exponential factor represents. Then we obtain first y±(z) = yA(z)±y^(z). (17. the Schrodinger equation with the cosine potential is a special case of several other more general cases.74b). We begin with the solution containing the function A(z).2. .74b) is also important for various other reasons. Naturally one expects the solutions there to be related to those of the periodic case considered here.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). Periodic Potentials this nonperturbative effect. Thus we do not reproduce every step. The explicit derivation of the level splitting (17. The splitting can also be obtained with some methods of large order of perturbation theory. in effect. T y'_(0) y'A±*) y{z). . The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion.6: Translation of Solutions For the largeh 2 even and odd solutions (17. yA{z) = Aq(z)e2hsin '. Example 17. 17. 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.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues. y^{z) = Aq(z)e~2hsin *.
1 )  .7 r ( q = F l ) } j / .• Then y'+ (0) = 0. From the equations for the sine and cosine we can deduce the value of the Wronskian at z = n in the leading approximation for large ft: W[y+.c o s  ~(<1 T 1) f y .31).17.Ah) ~ 32ft.4ft) ~ 8(q .[A'qM + 2hAq(n)] = iV2(q . y'_ (0) = \/2(4h .g A ? ( z = ±TT).4 17. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .) cos < 7r(q 1) [ = .4ft) sin j ^n(q .3 The Ellipsoidal Potential Replacing VA. Aq(z) The derivatives can be handled with the help of Eq. = ±ir). The Lame equation has not been a widely known equation of mathematical physics so . we obtain the conditions stated at the beginning. which is easiest by comparison with the foregoing treatment of the Mathieu equation.7 r ( g ^ l ) U + ( z ) q : i s i n { . Similarly we find y(z±ir) = = and by putting z = 0: V.%/2(g .T ( ? =F 1) p + (0). 17.y]z=„ = [y+yL .( z ) From this we deduce for z = 0.( 2 ) . and y+(z = ±TT) = V2(q . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials. (l^ifr^i^O + Cl^ifo^j/aCO cos  . etc.4.( ± 7 r ) = c ° s  .q). ( . Replacing sines and cosines by the functional expressions. A ?q( Z = ±TT) = .VA DV 371 1 V±. Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . but also to enable a brief familiarization with the Lame equation.yV+\z^ = [j/+(0)] 2 (<? . y+(0)~2V2. (17.. the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r).z ) . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .4ft.4ft) cos i TT(<J + 1 ) 1 . where Aq(0) = Aq(0) = V2. A' (z = ±ir) = (.2hAq(n)] .
. H. This line of investigation. Arscott [11] (p. J. 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 . for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.* As Arscott [11] (p. Periodic Potentials far. 194) remarks.372 CHAPTER 17.fi2A.1. inverted double well and cosine potentials (J. W.K2sn2u . the nonoccurrence of the Floquet exponent. H.5 < z < 5..5 The potential sn2(z..75. 194) remarks to parallels with the Mathieu equation:". Tchrakian [165]).0.* and the work of both prepared the ground for presentday investigations. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials.0 and ." Fig.4sn4u]y = 0. Liang. (17. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). *A. p.Q. The most conspicuous difference compared with the Mathieu equation is. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . 17.. as alluded to at the beginning.98. however. L. k) for k = 0. Arscott [11]. Erdelyi [85]. but also cubic potentials. so that Floquet's theorem cannot be immediately applied . M... f E .75) ' T h i s means double well. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940. Ince [138]. 19. MullerKirsten and D.. But more recently it has been observed to arise in various contexts.5. § F . .0.0. and was continued by Erdelyi.
The function sn2u is plotted in Fig. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). i.2k. W. in front of the second derivative has to be kept in mind (mo being the mass).76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K.\k\ < 1.4. In comparisons with the Schrodinger equation. (17. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. MiillerKirsten. 17.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. (17. J. .1.77) into Eq.K) = qK+^±. (17. in the case n2 — oo.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). the usual factor —h2/2mo. K being the complete elliptic integral of the first kind. 17. cmx and dnu. 2 . is the elliptic modulus of the Jacobian elliptic functions snu.78) We follow in this brief recapitulation the description in H.2 = n(n + l)k2. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.e.5 for several values of fc.e. If we put fl = 0. The first step is to write the eigenvalue A as A(q. N = 0.77) where q —> go = 2iV + 1. The second step is to insert (17. i. . Eq. Although the results have been calculated for the ellipsoidal wave equation.76) which can be looked at as a Schrodinger equation with periodic potential K2sn2u where snu is one of the Jacobian elliptic functions of period 2K.17. The range of the independent variable u is 0 < u < 2K. In order to distinguish the above equation from that with the Lame potential. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. Jianzu Zhang and Yunbo Zhang [206]. the equation ^  + [A .K2sx?u}y = 0. . we consider mainly the Lame equation. (17. .75) reduces to Lame's equation and one writes K. (17. and of the derivation of the level splitting'.
K). i. / z J . C which are valid for dnu±cnu <1.374 where CHAPTER 17. by transforming the equations for A. one obtains again the same expansion for A.e. (17. Periodic Potentials . the other in terms of those of an imaginary variable. k' = Vl~ k\ \dmi±cnu/ 17. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k. dnu± cnu K Thus one can construct solutions Ec(u). more precisly for dnu = cnu F 1 » . but solutions B. ^ (17. A are derived.q. \/8K (dmz = cn« \ ' F . Thus in the third step two more pairs of solutions B. C replacing yl. A into equations in terms of the variables z(u) = ^—{k' . /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.q. i.K — > A(u)=A(u + 2K).Es(ii). . _ Solving the resulting equations iteratively as before. (17.79) The very useful property of these solutions is that for the same value of A (which remains unchanged under the combined replacements q —>• —q. B and C.82 . dnu =F cnu .77).81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed).K)=A(u. A.e. one has to derive new sets of solutions there and match these to the former (i. one pair in terms of Hermite functions of a real variable. A second solution is written y = A(u) exp < / nsnudu >.80) The domain of validity of these solutions is that away from an extremum of the potential.e. which are respectively even in u (or snu) or odd. determine their proportionality factors) in domains of overlap (their extreme regions of validity). A(u. . (17..
" A = qK±(l 1 + r {(l k2)(q2 + l)^{(l + ifc2)3(5g4 + 3 V + 9) + k2)2(q2 + 3)4k2(q2 + 5)} 2WK2 Ak2{l + k2)(5qA + Uq2 + 9) .86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H.63)} + ••• . Es(2f0 = Es(0) = 0.e. W. *A. C = a~A.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. a of _ _ B = aA. (17. J. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. p. as well as \duJ2K 11 (17.6(5g4 . Oberhettinger and F. i. This expansion is found to be in the more general case of Eq.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. 64. Miiller [205]. Tricomi [87]. G.384n2k4(q2 + 1)} q r {(l + A. W. 17.^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) 8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) 64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£.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. Erdelyi. (17.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . £2 = 0).75).4. . who obtained this expansion for the eigenvalues of Lame's equation (i.17. the ellipsoidal equation.85) The first three terms of this expansion were first given by Ince [138]. Magnus.e.38g2 . F. (17.
88g0 . from which the difference q .l n 2^+ i J + 0(«i) (17.g 0 ) ( ^ . (17. Periodic Potentials These conditions define respectively functions EcX°.89) one obtains the eigenvalues from which the level splitting can be deduced.2 /c +256fc2g0(g2 + 5)} (17.j 26K2 (1 85) = A(g0) + (q . W.136g0 + 9) n 3.91) tt H.40g03 + 18g02 . + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ .87) (17.i Here the upper sign refers to Ec*> l or Ec£°.Ec£ 0_1 and Es2° of periods 4K.2 n K 2 {3(1 + 1 +128fc2(2g^ + 9ql + 10g0 + 15)} For the two lowest levels go = 1 and one obtains for their separation AA(1) 2(4^) 3 / 2 (lfc)*.Es^ 0+1 . Evaluating these one obtains (from factors of factorials in q and —q) expressions cot{7r(g — l)/4} = • • • and tan{7r(g — l)/4} = • • • (in much the same way as in the case of Mathieu functions). J. 2K. go being an odd integer.g0 is obtained by expansion around zeros.:J {3(1 + k2)2 (9q* . 2K and 4K respectively. Finally expanding A(g) ~ A(g0) + ( g .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .78g^ . 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 .4k2(q2 + 2g0 + 5)} + • • • ]. .376 CHAPTER 17. and the lower to Es*>+1 or Es*5.90) + 3. Muller [205]. 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 25.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .
One can see.f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. . .* Apart from the choice of notation. Thus in the case of the dominant contribution of Eq. W. 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.4 h ( i e M 9 0 7 2 \. K = 2/i.* M G .87) agree with those of Ince.93) (17.Ah2 sin2 u}y = 0. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . V. but also more easily generalizable. Dunne and K. Ince [138]. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite.2h2 cos 2x}y = 0. + E. in which they dubbed the classical configurations Lame instantons. Miiller [206]. 17. (17. (17.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. that calculations with the Schrodinger equation are not only easier. J. i.2h2 = A. (17.74a). h .92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.86).90) one has in this limit: /l±±\ ~K/U = e . K2 = h2 (say). A .2h2 .J1 O.17. the conditions (17. Rao [79].3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. Replacing u by x ± IT/2. however./2i and hence A^ A/ A AU I2 . Hence the conditions fi = 0. this equation becomes y" + {A . (17.e.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. *H.4. it = x ± ^ . A = 0. L.
i. Gu and S. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. . W. Periodic Potentials 17.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). MiillerKirsten and D.^ The equations considered above also appear in diverse new problems of physics. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine.Q . 1 1 See Z. H.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. all basic potentials. in the problem of two parallel solenoids the lines of constant electromagnetic vector potential  A are elliptic with the Hamiltonian separating into a Mathieu equation and an associated Mathieu equation. § . See Chapter 25. Quian [123]. This is an interesting problem. Rana [287]. A generalized associated Lame potential has been considered by A. de Veigy and S. D. Ganguly [103]. double well. **Y.Y. S.e. See e. A.5 Concluding Remarks The potentials we considered above will reappear in later chapters. H. H. N. A. for instance.378 CHAPTER 17. Cho. 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. MiillerKirsten and J. Liang. Thus. W. Khare and U. Kan and K. inverted double well and cubic potentials. J . A host of related elliptic equations has recently been discovered and studied. Shiraishi [49].W. J.g. Ouvry [276] and Jianzu Zhang. M. J. Tchrakian [165]. Sukhatme [148].
1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. In particular the investigations of Bender and Wu. 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. 379 .Chapter 18 Anharmonic Oscillator Potentials 18. This seemingly simple problem revealed extremely rich internal structure . Wiedemann [4] and a revised version of parts of this reference by J. ' T h u s A.. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. T. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. Liang and H. M. T. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. both of which make the calculation more difficult. J. Wu [18]. f C . Achuthan. MiillerKirsten [163]. Bender and T. MiillerKirsten and A.Q. The recent work of R. Weinstein [281]. [19]. D. Bender and T. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C.". Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. W. J. H. W. Priedberg and T. Wu..t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest.* 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.
380 CHAPTER 18. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . described as the case of the double well potential. V(z) = \\h*\z2 I 2i 4 \cr\z . (3) Complex eigenvalues with tunneling: In this case. These contributions may be given different signs. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation.1. and thus lead to very different physical situations. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. 18. The three different cases are: (1) Discrete eigenvalues with no tunneling: In this case V{z) = \\h'\z* + \\<?\z\ (2) Discrete eigenvalues with tunneling: In this case. *~z (1) (2) (3) Fig. 18.1 The three different types of anharmonic potentials. with the potential described as an inverted double well potential. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [EV(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. V(z) = \\h*\z* + \\<?\z*.
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. (18.1 Introductory Remarks 381 normalisable wave functions. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. however with decay as a consequence of tunneling. except for a change of sign of c 2 . . Thus we are mainly concerned with the double well potential and its inverted form. The level splitting will be rederived from path integrals in Chapter 26. since the potential decreases without limit on either side of the centre. We begin with the latter. 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. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). We do not dwell on Case (1) since this is effectively included in the first part of Case (3).175) below. The result is an expansion in descending powers of h?. 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. as is sometimes hoped. 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). We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. it will not be possible to obtain the exponentially small contributions with convergent expansions.86) below. i. The shift of the eigenvalues is best calculated with straightforward perturbation theory. The eigenvalues of this case are given by Eq. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. 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. 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. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined.18. that the general applicability of the method becomes evident.e. (18. Case (3) is seen to be very different from the first two cases. The eigenvalues of this case are given by Eq. The imaginary part will be rederived from path integrals in Chapter 26. In this case our aim is to obtain the aforementioned complex eigenvalue. 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.
we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei. 18. We take here h = 1 and the mass rriQ of the particle = 1/2. 18.2. Anharmonic Oscillator Potentials 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.3) "1/2 harmonic oscillator h8/25c2 .1) for h4 and c 2 real and positive.2 The inverted double well potential with (hatched) oscillator potential.1 and more specifically in Fig. This implies t h a t results for THQ = 1 (a frequent convention in field theory considerations) differ from those obtained here by factors of 2 1 / / 2 .382 CHAPTER 18. v(z) = ^h4z2 + ^c2z\ (18.z Fig. a point which has to be kept in mind in comparisons. . h 4 2 2h1> ~1/2 2c?. If suffixes 1/2. (18.2. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. The potential in this case is given by V{z) = v(z).2 18.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. (18.1 refer to the two cases. 18.
then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. _i\(w) and q = qo = 2n + l.1.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.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. i. and a variable w defined by setting E= b* 1 + A A and w = hz.18. /i). to match these in domains of overlap. \h?\ — oo and c finite.2.1/7  Vg(w)y(w) = o(J^j. n = 0. the harmonic part of the > potential dominates over the quartic contribution and Eq.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. (18.. y{w) oc Di..2. (18. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig. (18..7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w).5) becomes . The problem here is to obtain the solutions in various domains of the variable. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. 4 .4) we can rewrite Eq. The positions z± of the maxima of V(z) on either side of z = 0 in the case c2 > 0 are obtained from h2 v'(z±) = 0 as z± = ± — with (18. 18. (18.e. (18.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small.d2 . The result will be that derived originally by Bender and Wu..
Before we return to solutions we derive a new pair which is valid in the adjoining domains. •A~2 hrz" 2^\ c^z V2 —— + (18.10) into (18. as we shall see. In order to arrive at solutions we set in Eq.14b) .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 = 0.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. as encountered and explained earlier.2.9) we obtain 1 h4z2 c2z4 A L2 (18.10) E= qh2 ^ j. Inserting (18.384 CHAPTER 18.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.14a) 1/2 VA( ) = A(z)exp (18.11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.4(2. (18. h) is obtained from the perturbation expansion of the eigenvalue.+ ^ . (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 ) .2 A (18.2 We are concerned with the equation d2y(z dz2 where ^ 1 . This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. be and A = A(q.9) z =0 these Thus these c2z4 1/2 + • 2 (18. This observation allows us to define the pair of solutions yA(z) = . these solutions are not valid around z = 0. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions. We observe — before touching the square roots in Eq.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18. 1 d r A!{z)±iA{z)±{ 0. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18.
13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0. We take the square root by setting z2h^1/2 +.20b) . Since these higher order contributions are of little interest for our present considerations. (18. (18. 385 (18.15) where A{z) is the solution of the upper of Eqs. 1 (18. (18. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq..e.18.13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.13) and A(z) that of the lower of these equations.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18. 1 2 4 2 Aq(z) + 0 (18. (18. i. / c z Aq{z)+0 1/2 . (18. Clearly Aq(z). Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. /H 4 Z h +~C Z 21.zh2 =()*2"(18.19) We see that one solution follows from the other by replacing z by — z.4 .1 and as a verification again in connection with the solution y# — as the other solutions. we do not pursue their calculation.16) For large h2 we can write the Eqs.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h.34) below) — to be derived in detail in Example 18.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).Aq(z) approximate the solutions of Eqs.
Vq+ii = Vq + 2i. z)). (18. 03 = .20b). a2 = 2 1 .. A = . i.! ) ^ . 2 . (18.. W i t h proper care in selecting signs of square roots we can use the solutions (18.17) and (18. 1 ..^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl.20a).e. A A _ 3 . z) ± yA(q.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q.20a) and (18. c*i = 1 1 .e.. Anharmonic Oscillator Potentials i. h2.• Using Eqs. one observes that with £>.l)A(z) 1 = tfA"(z) A . 128 8 16 A = i. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. away from the central minimum.21) Example 18. ft = .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). 04 = 5 .18) we obtain A » = H^rL^z) 2 z . h2 —> —h2).2 ( * ) 2 and from this or separately A"(Z) : 1 (q .[yA(q. . .1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. h2. (18. (18.2 .e. to obtain in leading order the eigenvalues E. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).v ' = o (« . (18. zA'(z) where the expansion coefficients are given by ao = l. X » / A q + 4 i = Mg+4i . i.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q . procedure.f .:=*£ +i(5l). „ .386 CHAPTER 18. we write > V±(z) = . * = — .
—77 [i(9l)]. w = hz.2 The Inverted Double Well Potential Thus a term /j. give an equation from which A is determined. 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). We return to Eq. _1J±w) and D_\.. r ( CJw) = ^ . (18. +1J±iw) (observe that Eq. (18. when i = 0.e.'2H9i) ^faDfr™) and .Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. = •>— r.(q + 3)(q + l)Aqvri(q '^ ' . 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. i. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= .. [1(9 + 1)]! . The solutions yq(z) of the equation Vq(w)yq(w) = 0.22) are parabolic cylinder functions D i . Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .. of course.5). c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B.18.22) is invariant under the combined substitutions q — —q. Rq '.w — > > ±iw) or functions R B w q( ) . (18. (18. ^f( g + i)(±H2^ + 1 > \ — 1.. those in Aq(z). and the coefficient of terms with i = 0. In this way we obtain the next order contribution A^1' to A^°>.23) .
Magnus. Actually these factors can also be extracted from W K B solutions for large values of q. (18.28) (A + c w )yq(w) E 2 4 1  2 ^ [q.±4) S4(9.±8) 5 4 (g.q + 4j] = c2SA{q. we also have Vq+^yq+tj = 0.3)y g _ 4 . .4j)yq+Aj.0). (18.^ For higher even powers of w we write i w2l V<i = Y. pp. See Example 18. Erdelyi.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. Comparison with our notation is easier if this reference is used. (18. (18. F.l)2/?4.( < ? .123. l(q2 + l).23) one can choose the solutions as Bq(w) with B QM = ~ 77[(93)]!24(91) These satisfy the recurrence relation w2yq{w) = (q+l)yq+4 + qyq + . j=i (18. (q±2)(q±S). and for j ± 0 : [q. (18.29) ' A . As an alternative to Bq(w) in Eq. G. Vq+4j = Vq + 4j.q}=A + c2S4(q.388 CHAPTER 18.3. Tricomi [86].26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.0) = = = ±(q±3)(q±7). 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. Oberhettinger and F. and so Vqyq+Aj(w) = Ajyq+Aj(w).25) where in the case i = 2: SA{q.24) The extra factors in Eq.Aj)(18. 115 . since Vqyq = 0. W. S2i(q.q + 4j}yq+4j.27) Now.
(18. . We can now write the solution y(w) in the form y{w)=yq{w) + Y. Thus the next order contribution to yq is y H = ie 2^ —z. and c 2  replaced by — c 2 . i.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.[rp ) Y. A = ^(q2 + l)c 2 + o(J^. and even powers of q in combination with even powers of 1/h2.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. i.1.—A(q + 1) .e. (18. .—v*+*j ( 18 .e.31) Proceeding in this way we obtain the solution y = yW („. 2 . Equation (18.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.3 °) For the sum y{w) = y(°)(«. (18.(!) («.q]=0. This type of invariance is a property of a very large class of eigenvalue problems.32) Evaluating this expansion and inserting the result for A into Eq. (18.• q. ( 18 .10) we obtain E(q. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. h2 —> h2.2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq.) + j. so that the entire expansion is invariant under the interchanges q .) ty^^w) to be a solution to that order we must also have to that order [q. n — 0.34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. p i(9»9 + 4 J'WK7».18. . h ) = qh . (18.35 ) .
? T 4 ] [ g T 4 . q + At] (18. W. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t.37) For further details concerning these coefficients. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz.)]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. > "p. g ± 4 ] T4 ±4 and so on. their recurrence relations and the solutions of the latter we refer to the literature.argz=0 (18. w — ±iw. i. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.11) is invariant under a change of sign of z. H.—.g±4] ±4 ±4 [g. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. q + 4j) = 0. Wiedemann [4] . o. Achuthan. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q. and for j ^ 0 all other P0(q.36) with the boundary conditions p o(q.q±A) = MM!i±MM+[M±8][g±8. ^VO(<L<?) ^i(?. We thus have the following pair of solutions VB(Z) VB(Z) = B w . (18. MiillerKirsten and A. We observed earlier that these are obtained by making the replacements q —> —q.390 where for instance v J CHAPTER 18. 2 4tPi{q.38) [yB(z)]argz=7T = [?/s(2. q) = = 1. J. we may infer that given one solution y(z).i 1 + 4j) in complete analogy to other applications of the method. there is another solution y(—z). Again we can write down a recurrence relation for the coefficients Pi{Q." Since our starting equation (18.e.
4 . (18. see e. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.z2 . the solutions of type A being valid away from the minimum.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. with normalization since the normalization constants are also asymptotic expansions.2. . Thus in the transition region some become proportional. Such contributions arise. 18.g. like w of Eqs. Integrating and expanding as follows since h? is assumed to be large.18. Mauss [192]. We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e. in fact.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. J.39) are with the same coefficients Pi(q.q + 4j) as in ys. one has to stretch each by appropriate expansion to the limit of its domain of validity.40) "Variables like those we use here for expansion about the minimum of a potential (e.h^ih. h2) in which odd powers of q are associated with odd powers of h?.3 M a t c h i n g of solutions We saw that the solutions of types B and C are valid around the central minimum at \z\ = 0.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. Vc(Z) = [V!B(Z)}q^~q. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18. (18.4)).g.h^ih 391 (18.Vc suitable asymptotically decreasing expansions in one of the domains \z\ < O 7T argz We emphasize again that all three pairs of solutions are associated with the same expansion of the eigenvalue E(q.g. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0.The solutions yc. Aq(z)) does not appear in any of the higher order terms. In order to be able to extract the proportionality factor between two solutions.
42) we obtain for the solution ys(z).l ) ] ! ( .z).7 r . not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. 2 ) i ' ir! i iHim 1 3 j argw < 7T. (18. The function Di. y^(z) we see that in the direction of z = 0 (of course. 2 ^i![i(g4il)]!(2«.(«.42) H W2 (ql)e$W I „ .41) with the solution ys(z) of Eq. > n. (18. eh* /12c* e\z*h? (18. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=. 4/T 4 Prom (18. we would have to substitute correspondingly the expression (18.w yB(z) ~ B. 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.44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.392 CHAPTER 18. Comparing the solution VA{Z) of Eq. (18.43) W + 1)]! w^+V A . We do not require this at present. .2 « .2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18..44) In the solution y~g(z). with z in ys(z) replaced by — z. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > .43) (since z — — 2 implies > argz = ±7r).) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.45) .4 i .0x ! [ £ ( g . Anharmonic Oscillator Potentials Considering the pair of solutions ju(.
49) d yc{z) Again there is no such simple relation between yA(z) 18.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.. the ratio of yA(z)iVB(z) i s n ° t a constant. The first of the conditions (18. Looking at the potential we are considering here — as depicted in Fig.e. i. (18.41).yc{z)pansion (18.1 ) e 12c 393 Kgl)]^"1) l + O h? (18.50) and y+(0) ^ 0. (18.2 — we see that near the origin the potential behaves like that of the harmonic oscillator in fact — our largeh 2 solutions require this for large h2. (18. — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0.. we see that in their common domain of validity yA(z) where a = =Vc(z).y'_(0) ^ 0. We proceed similarly with the solutions yc(z).42) into y~c(z) w e obtain Vc(z) = (h*. 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.21) as even and odd about z = 0.50) will be seen to imply qo = 2n + 1 = 1. hence we have to .5. 18.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.46) However.7.18. Recalling the solutions y±{z) which we defined with Eq.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. large probability for the particle to be found thereabouts. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.11.. At z = 0 the solutions of type A are invalid.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 .9.2. and the second qo = 3. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.
(0) (18. we obtain i V7T24( g .SxW) = \ i»fl(0) .s i n { .57) [i( 9 _l)] ! 2 i(9i) B.l)(°) Thus with the help of the reflection formula cited above: D B. e. with the reflection formula (—z)\{z — 1)! = IT/ simrz./o^Tpif (q+i) ..56) D 4(. A.2.(0) = and htoi)(°) ^—'. (18. + 1)]!' D ' j .u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin..' '.5 ( 9 + !!)(*"')• i ( 1) .i ( g + 3)]l (18. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i. Z (18. Stegun [1].2: Evaluation of yB(0).i ) . y c ( 0 ) .51) and 0 = V(0) = Jim i»xM .i)H = e.(iu)]. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1. one finds that [ 1 IT (q + 1)1! = .fc(0) Thus we obtain the equations (18.54) dio q(w) .oi(. ( 9 + 3)]! s i n { ..( g + 3)}.(0) + ^ ( 0 ) l z—>0 2. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin. [ . Expressions for C 9 (0). [C. C.! ( . We leave the detailed calculations to Example 18.(0) = .^{(9+1)}. IT 4 dio In fact. M.(«."*(«!) ^ ' ~ [ .52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.( g + 3)}.) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18. y'B(0).l ) ] ! [ .. Example 18.(0) Cq(w) y' c (0) r^ Show that — with w = hz — the leading terms of the quantities listed are given by B n\\iam' 4(«Z + l)]l[i(8l)]!' i\ sin{(q3)}.g.55) Solution: Prom the literature.1)('») + ^ r + l)]! ^. Abramowitz and I.394 CHAPTER 18.58) diu [  ( g .F ? . — C. . 1 [7(93)]! n «(°) = .
using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z .2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.sin{^( 9 .49).(0) = *>i(„+i)(0> 2 . + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.6 ) } ^ — i t a n < — (q + 3) >. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs. (18.+i)(°) [ _ I ( .1/2)!. isin{f(g + 3 . 4 r T (18. Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .62) . we obtain 2 i(»+i)[_i(.53) in dominant order and insert the appropriate expressions for a and a from Eqs. (18.+1iH]«=o = . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(. (18._l) From this we derive C.18. _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.46). ~ ^ 1 5(9+1) 2i(''.sin{^(g+ 1)} = J .i .)[i(g3)]! and ^i(.i ( « + i ) [ _ i ( .3)}.56).3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^. W . 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) _^_ .
L ^ i .9. (18. With a Taylor expansion about go the left hand side of (18.. ^ .65) but now for the odd function with these values of goWe have thus obtained the conditions resulting from the boundary conditions at z = 0. + 3)} = .e..64) for q = g0 = 3.1). (18.. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. 1 1 . the vanishing of the wave functions at infinity).^ . (18.2) with potential (18.i . this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. we again obtain (18. equivalently. . Thus we have to determine these conditions first. Anharmonic Oscillator Potentials Then the derivative relation of Eqs.63) becomes (? .63) and (18. (18. and the left hand side of (18..2.64) the right hand side is an exponentially small quantity. . (18. 7 .9o) ^ cos  . the energy E is real.63) Proceeding similarly with the second of relations (18.1.64) In each of Eqs. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo. i. (gg 0 )^±^ y e5?..53). (18. Our next task is to extend the solution all the way to the region beyond the shoulders of the inverted double well potential and to impose the necessary boundary conditions there. we obtain cos((g + 3)) = .. z2 > z2. . . 18..( g 0 + 3)  + • • • ~ (g .63) vanishes for q = go = 1.^ >^_>m e J*.go) ^ (1) It follows that we obtain for the even function with q — go = 1.11..9.53) becomes 8 m(I(. In fact the left hand side of (18. .65) Expanding similarly the left hand side of Eq.._ e =*.5.5. This is Case (1) of Fig. 18. by the rotations E > einE = E..64) about go — 3.. 7.5 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.396 CHAPTER 18. z^ ein/2z. Recall > the original Schrodinger equation (18.
cos 36.18.66) z3 = \z\3em. we set z w r = \z\eie. i. 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. i. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. 6 6 In the case of the inverted double well potential under consideration here (i. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied.e. the resulting wave functions vanish at infinity and thus are square integrable.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0.e. 1/2. r / c 2x 1/2^3 = exp{ ± i — —\.68) .2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.e.e. (18. i. the case of complex E). Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. in the domain — TT/2 < 36 < > 7r/2.e. (18. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. i. if TT/3 < 6 < 0. replacing sin 30 by " 0 + ^) 1 IT = . —}. Now. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. 2 c2>0. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — . or — < 6 < . y{z)~exv\i[) 7T dV'V. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36.67) Rotating z by vr/2.
and there impose the boundary condition (18. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. Looking at Fig. Our procedure now is to continue the even and odd solutions (18.21) to + infinity and to demand that they satisfy the condition (18. We therefore explain our procedure first.e. (18. We have to remember that we have various branches of the solutions y(z) in different domains of z.21)) were defined in terms of solutions of type A which have a wide domain of validity.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero). This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator.68) for c 2 > 0. for z > Too. (18.Z\.Z\.3 The inverted double well potential with turning points ZQ. 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. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. Eq. 18. we match the WKB solutions .398 CHAPTER 18. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.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. We do this extension with the help of WKB solutions. Anharmonic Oscillator Potentials V(z) Fig. This is the boundary condition also used by Bender and Wu [18]. 18.68) will lead to our second condition which together with the first obtained from boundary conditions at the origin determines the imaginary part of the eigenvalue E. Our even and odd solutions y±(z) (cf. i.
and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\. ^2? 1 2qc2 /2? .71) We now come to the algebra of evaluating the integrals in the above solutions.^ T rz\ 1 + yh* . i. p. t To the right of the turning point at z\ these solutions match on to (r. 18.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ .*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . Dingle [70].2.70).. Vol.18. z2h4/A > c2z4/2.^ 4 1/2 + icV 7T Jz + (18. p. The distant turning point at z\ as indicated in Fig. We begin with the exponential factors occurring in Eqs. 6. ^ 2 4 •K + —(r. 4 2 T ' i.e.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i.3 is given by (using Eq.4) for E and ignoring nondominant terms) t o . h2 (18. (18.69) The W K B solutions have been discussed in Chapter 14. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. *R. f R . Sec. 291.70) where z < z\. equations (21).4. Messiah. A J24 1 ? 2 • "^ •z h \—c z ~ — q h . 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.=c z 4 dz 1 L2 9 + 2 4 1/2 x exp .e.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A. (22) or A.zi)/ \ 1/4 ywKB\ ) z — g/i* . 291. . I [195].4 h + c z 2 1/4 < . In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. Dingle [70]. z\ ~ . (18.B. B.
2.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.\qh2 + \z2h± .. Anharmonic Oscillator Potentials E± = exp = exp . CHAPTER 18. . In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c. (18. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\).2 \ 1/2 .40)).73) .^ I 2 2 1/2 zV = F 8c2 Jz [i . Eq.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. Eq.e. (18.400 i.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.40)) E± = exp _/^_2 f ± In A 2c2 .^ i ^ ^ c 2 .211/2 1 \ /j.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . (18. Thus the above factor yields '2c2\l/2 h4 so that (cf.2. zi 2q 1 c 2 [1 .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.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.
2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (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.*l) (3y^(z) (18.18.\z2h4 + J Z\ nl/2 \M fZ . > » Returning to the even and odd solutions defined by Eqs. h2 — — h2 is maintained.*l) (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q. (18. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .76b) apart from factors [ l + 0 ( l / / i 2 ) ] .77) Now in the domain 2 — oo we have > \qh2 .20b).1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.21 1/2 .78) Inserting this into the solutions (18.79) +S_(±)exp .74) Comparing these solutions now with the solutions (18.77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) .20a) and (18.71) and these into (18.76a) P P r 2/1 2 " 9 ( .(J*i) VA 0 ) = PVwKB (Z C.
6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.. (18. i.65). Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq. E(q.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( . (18.81) Inserting expressions (18.82b) This is our second condition along with Eq. .^ j .82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.^(q2 + 1) . (18. i.5.. Eq. (18.h2) + (qqQ)h + (18. 18.76a) for (3 and /3.h2) + {qqo)(^pj = E(q0./ ? = 0. (18. (18. i / 3 ± .e./i2) = E{qQ.34).. (18.. we see that we have to demand that 5 + ( ± ) = 0.^ § ( 4 g 2 + 29) + O ( ^ ) .85) + . (18.3.84) Expanding about q = qo we obtain £(g.2.68).402 CHAPTER 18.e.. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J . h2) = \qh2 .
The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.36) of their Ref.86) will be dealt with in Chapter 20.ZQ) .3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q. ZQ (2g)V 2 / h \ 2qc2 he (2q)1/2 In the domain 9ZE > V to the left of ZQ the dominant terms of the WKB solutions are (cf. h2).h6/6c2 E = E(q0.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 . Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0. !*WKB(Z) ~ 2 .2 . i. Example 18. Since this is not without interest (e. S W K B ( 2 ) = PVA(z).~p{q. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. where the superscript {r.e. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .3.2 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 . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. ft6 ^ = e.l ) ] ! > + = ^ 2 . above) (1. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. h2) of the relations KWKB (Z) =PVA(Z). zo} means "to the right of the turning point ZQ" . o 2 + l)^(4.g.yA{z).18.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 ( .
the (approximate) solutions* 1/2 exp • ± i ( ./ ^ 2 X 1 / 2 exp(ifi. Anharmonic Oscillator Potentials where zo > z.l c y dz 4 1/2 x exp •! / —(r>zo)/ \ 2 1/4 1 2y x exp 2 1/2 L dz qh2 2H + ~z2h4 4  2 c2z4 we nave where 2 > 20.zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . cf. 2 z 2 ) / 2 e \ . and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 .1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation ." / 4 hj (ft2 2 2)( 9 +l)/4 J'W'KB (Z) 2\1/2(fe222)^1)/4 h) exp(±h2z2) ^2ey/4 * Solutions of this type (which are asymptotic forms for q — 00) have been investigated in t h e > literature. In the domain V > SHE to the right of 20 the dominant terms of the WKB solutions which match on to 2 / W K B ( Z ) ' ^ W K B ( Z ) a r e res Pectively 1/4 (r. f x Zn 4) h2 2.404 CHAPTER 18. .ic224 2 4 qh2 + i^fe 4 .g In other words.zo)/ \ qh2 2 ZO + z2hA 4 2 ..] { hz combined with the particular constants contained in y ^ ' ^ g . On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20. 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.O. J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. Schwind [247]. N.
5 "l _ o f . i. „ / M 1 / 2 i . in fact.3 18.(r.zo). We can reexpress these relations with the help of Stirling's formula.3. Then .e.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. (18.13) of Bender and Wu [19].18. 4" ) (h?z2)^q+1) for q j£ odd integers. (j . h4 > 0.88) + [EV(z))y{z)=0 (18.DfJ»v we see Comparing now the expressions (18.. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. in Eq._ .^ 1 . 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.yA{z) with expressions 27WKB i J'WKB ' that the factors p." / 4 ( ? « / 4 . 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. But there are significant differences. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.kyil(.z ± ) = .1)! = ( 2 7 r ) 1 / 2 e .g. (18. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . (18.J ' i J .23)) appear quite naturally. we shall employ basically the same procedure as above or. as in the case of the Mathieu equation.89) .3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.41) for yA(Z).
(18.94) (18.406 CHAPTER 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 .91) becomes T>q±(w±)y(w±) = o(^)y.95a) .V(z±) l{z z±fhA + 0[(z . 18.z±) y = 0.90) V(i)(^)=0. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^.91) (18. (18.93) w± = h±(z . .z±). the previous equation (18. (18.4 The double well potential.i>5.z Fig. (18. vW(z±) = ±Qch2.
97a) U(z) = (zz±)2 + 0[(zz±)3}. 18..98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.96).97b) Our basic equation. we obtain d*y (18. Eq. qo = 2n + l.95b) with the equation of parabolic cylinder functions u(z) = D„(z). we conclude that in the dominant approximation q± is an odd integer.95b) By comparison of Eqs. (18.87).i 4 f W » = o. is again seen to be invariant under a change of sign of z..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. i. (18.95a) and (18. Inserting the expression (18.2. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].1. (18.e. n = 0.96) dz2 + 5 « 4 +  . we obtain another by replacing z by —z.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8. given one solution.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. Thus again..100) .. (18.93) for E into Eq.3. (18. d2u{z dz2 + v+ 2~r u(z) = 0. (18.
yA(q.c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t .105) "8 + 4 ^ . h2.^q(l7q2 6 4/i + 19) + (18. h2.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c). The result is given by A and E(q. (18. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —. Aq(z) = Aq(—z) = Aq(z).104) Both solutions yA{z).101b).1 6 ) . i.^ A"{z) + 2c V2 *A(z) (18.103) Looking at Eqs. z) = yA{q. n± = V2h2. h2 h4 2c' with g+ = q. (18.4. we observe that the solution y^iq.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18.c W + 1) . z) may be obtained from the solution yA{q. Vq = {z2+z2)— + (qz+z).e.e. z). z) = yA{~q.yA(z) are associated with one and the same expansion for A and hence E. h2.101a).102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. these equations can be rewritten as + (qz+z)A{z) = . (18.408 Since CHAPTER 18. We leave the calculation to Example 18. but the solution is a different one. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. Integration of Eq.h2.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18.h2) . (18. 8hw (18. ~h2. i.101b) To a first approximation for large h2 we can neglect the right hand sides.106) . (18.
101a) with A replaced by A . 92 ZZJf (18. ± 1 .106).qz+)2 . we reexpress A'q in terms of functions of z multiplied by Aq. z • .105) and (18. such as periodic potentials.e. (18. i.18.4zz+q+ z\{q2 + 1)]. ± 2 . Thus it is natural to explore analogous steps.3 The Double Well Potential 409 Example 18.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18.(l + + 2  z y* . . .z2 ) N t.+ 2 ^ ± l + 2^±i + .114a) .102). . (18.113) 2 . <4q+2 — z — z+ Aq+2iAq+2j (18. (18. J = 0.z\) + 2z(z .2 z\) • Aq$4 £Aq + Aq—4. as a linear combination of terms Aq+2i. (18. Solution: The structure of the solution (18.110) Aq+2+Aq2 9+2 A.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 . The first such step would be to reexpress the right hand side of Eq.4: Calculation of eigenvalues along with solutions of type A Show in conjunction with the derivation of solution y^ that the leading terms of A and hence E are given by Eqs.{z2 .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 .103): . (18.108) We wish to rewrite this expression as a sum y coefficientiv4.112) From these we obtain. We know the first derivative of Aq from Eq. . for instance by componendo z z et dividendo. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18. + Aq+2 + Aq Aq+2 . (18.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .Aq 1 + 2^± 2 Aq+2 (^M^f (18. = Aq+2i+2jAq. First. however. .109) We observe some properties of the function Aq{z) given by Eq. . A Differentiation yields » = ^^w (18.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials..qz+) [2z2 .
101a) the contribution R .3 ) A .114b) . (q. _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4. cf.112).l ) ( g . In particular the dominant approximation of A is obtained by setting (q. _ 4 . Anharmonic Oscillator Potentials A 1 .119) It is now clear how the calculation of higher order contributions proceeds in our standard way.34). (18. (18.<? +{q. (17. (18. (18.q + 2) = 4{q + l ) 2 .2A g + A g _ 4 ] . ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.4A ? _2 + 6Aq . q) = 0.2Aq + Aq4] 1.115) and (18.4]j.2\2 (z* .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 . <?) = 6(<j2 + 1) + z: 2 2A +72"' (18. Since VqAq = 0.+6 + • • • ] . .108).118) 2s/2c ~^j[(?.q + 2)Aq+2 (18. (18. . i.117) 4(q + l)2Aq+2 Here the first approximation of A.q)Aq + (q.115) (z*z%) 2 i2 (z 2 .116) into Eq.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.e. leaves uncompensated on the right hand side of Eq. QA?+4 _ 10 A + 6 . A(°~) = Aq. q + 4)Aq+4 + (q.33) and (17. •••].112) CHAPTER 18.116) [Ag4 . where the lowest coefficients have been determined above as (9.(0) V2 ' 2c K + h4+Ao . (18. Eqs. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.4Aq+2 + Aq+A Inserting (18.l ) 2 A .4 ( g .4)A„_ 4 + (q.4 4+* + . (q. 1_4^±^+8^±±12^±^ +.410 and from this = Hence with Eq.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.aAg+S 12+ 16 (18.9=F4) = (g=Fl)(?=F3).q + 6)A<.q2)Aq2 + (q. we obtain 5 K = ( ^ f ) [ ( 9 .
h2.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.h2.e. h2.0)=0. of course.2 ) (0) (18.e. (18. i. g .z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A.g2)(g2. (i) _ / V2c\ [ ( g . y'_(q.g) 4z+ {q q)+ 2r (9.h2. 9 + 2) Aq+2Az+ (9.105).123) ^[yA(q.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.124) . (18. i.4 ) ( 0 ) ( g .g) Az.h2. (g.h2. g) 8z 4 .101a) the contribution Rq . The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.0)=0. thus yielding the next approximation of A as given in Eq. g + 4)(g + 4. z)\ = 2^A^ h2 '>z)± VA(Q. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A. (18. for i = 0. where Rq is the sum of terms left uncompensated by AW. h2 2y^ Age? h4 (18.{^ h2.122) The solutions yA(z).0) = ^ .Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9. h2. g .94)(g4. 9 + 4) ^+48z+ (18. y'+(q. Terms jj. A' (0) = q/z\) y+(q.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.94) 8z+ (9.g + 2)(g + 2.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9. *)] (18. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .28.g) 8z+ (g.yA(z) derived above are valid around z — 0.z)±yA(q.g ( 1 7 g 2 + 19). in the domains away from the minima.(0).
yc(z) with complex variables and Cq(w) given by Eq.126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.91). We see already from Eqs.95b) that the solution there is of parabolic cylinder type.1 ) D (18. Moreover. (18. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). This means.95a) and (18. (18. Bq[w±{ ^ . (18.93) and setting w± = h±(z — z±). Since w±(—z) = —h±(z + z±). 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. yB(z).125b) It is clear that correspondingly we have solutions yc(z). and another VB(Z) ± 25^ch3wl + c2wi .3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).Evaluating the exponential factor contained in VA{Z) of Eq. Inserting (18. 2(9+) . we > . in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. Anharmonic Oscillator Potentials Our second pair of solutions.2A y(w±).We draw attention to two additional points.95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2).125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18. but w±(z±) = 0. (18.412 CHAPTER 18. we have w±(—z±) = —2h±z±.99) for z — z±.^ % [i(?l)]!24(«.yB(z) is obtained around a minimum of the potential. (18. 18. + 0(hl2). in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq.3.23) with appropriate change of parameters to those of the present case.
.127) Recalling that around arg w± ~ 0. Thus for h very large these turning points are very close to the minimum at z+. . (18.i*l(«^ 4'"+^+ 4"'+v ( « . 24(9D[i(gi)]! hj (18.• • " . Allowing z to approach z+ in the solution IJA(Z).• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I .129) . Later we calculate the coordinates of turning points ZQ.172) obtained later. . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. 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. 2 An+z+e 4w+ (18.frW.« + ) 2 (9+1) e (18. we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i .97b) ~ exp .172)) we add here comments on this approximation. (18. (18.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. ein+z+e in+(z 22+(<?+i) . .^/w + ± 1 . and comparing with Eq.z±)dz . .18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.146). .164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. (18.3 The Double Well Potential (cf.ZI and find that these are given by z+ + 0(l/h) for finite q (cf.147)). (18. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. a= ^ . Eqs. Here z± = ±h?/2c are the positions of the minima of the potential.125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z). (18.
Bhattacharya and A. and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima. The involvement of these WKB solutions leads to problems. Various investigations^ therefore struggle to overcome this to a good approximation. H. g . they assume large quantum numbers. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. The next aspect to be considered is that of boundary conditions.414 and CHAPTER 18. We achieve the same goal here by demanding our basic perturbation solutions to be interconvertible on the basis of the parameter symmetries of the original equation. even above the turning points. D. 18.7r and the solutions of type C around argz = ±7r/2. de Deus [66] and W.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. we naturally expect the wave function there to be similar to that of the harmonic oscillator. K. The pair of solutions of type B is defined around argz = 0.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. S. R. Rau [22]. The solutions in terms of parabolic cylinder functions have a wide range of validity. K.\h\)^+l\\{q l + l)]\ 1+ 0 J_ .(18. We have to impose boundary conditions at the minima and at the origin. Thus 1 E . and this means at both minima. Furry [102] . but it is clear that none of the above solutions can be used at the top of the central barrier.z+)]fr+V ' (18. S. J. Since it is more probable to find a particle in the region of a minimum than elsewhere. P.3.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . since. Bhattacharya [23]. basically.
134) y'+{z±) ± 0.Zii yB{z±) a ± a yciz±) (18.132) Fig.135) Hence the conditions (18. (18.132). We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0. 18.5. (18. The odd wave function then exhibits a correspondingly opposite behaviour.18.5. y{z±) £ 0. 18. as indicated in Fig.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. ^(*±)#0 (18. We have y±(z±) = ^VA{z)±yA{z)]z^. as indicated in Fig. and y+(z±) = 0. However. as indicated there.136) . At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. an even wave function can also pass through zero at these points.133) y+(z±)^0. y(z±)=0. 18. y'_(z±) = 0.5 Behaviour of fundamental wave functions. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.
.134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 . . .. (1.142) (<Z9o)J(l)^°+1)(l)1/2 . Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs.„*{!<. (18. Thus sin {^ +1) } = ^^>^30fe^i.«*{=(.2 we can rewrite the second of (18.54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula. _ 8)} s .136) the dominant approximations we obtain (cf. (18. +1)} = 4 m. Eqs. .m Using Eq. 1 1 .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. 7 . (18.9. also (18.138) Using formulae derived in Example 18. (18.136) as  M ..416 CHAPTER 18.5..
5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin. 18.18.» ^ B ^ ' . We do this with the help of WKB solutions. Since the type A solutions are not valid at the minima.144) Thus we require the extension of our solutions to the region around the local maximum at z = 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. y'+(0) = 0. Hence we have to impose at z — 0 the conditions j/_(0) = 0.3 The Double Well Potential Thus altogether we obtain 417 ^ • . V(z) — • z E=V Fig.3. (18.6 Turning points z$. We deduce .1 " <813 1 4)  In Example 18.z\. we must also demand this behaviour here along with a nonvanishing Wronskian. y'_(0)^0. 18. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. y+(0)^0.
In the domain 0 < z < ZQ. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18.\qh\ + \h%U{z) (18. We also note that the height of the potential at the turning points is h8 qh V(z) 20.e.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. + £\h%U(z)=Q. As a consequence.e. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0.^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.144).146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer. i.147) Since the minima are at z± — h2 /2c with /i 2 very large. Anharmonic Oscillator Potentials from Eq.418 CHAPTER 18.145) \qh\ + \zW Using z+ = h2 /2c.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.148) ("f * "The superscript (I. h8 25c2 (18. (18. ZQ) means "to the left of 20" . the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) . to the left of ZQ where V > E. one finds that .
(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ .18.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.^ e therefore have to consider the exponential factors occurring in (18.i (2q ln (18.* 4+ .151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing. .S2 ) dz rzo 1/2 'Ihl + ^Uiz) I 1/2 / Jz dz (z~z+)  w + 1/2 1 = ±lh+ 1/2 ZQZ+ Z — Z+ 2 i^$) 1/2 ~4 >'^. 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) . Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1. (18.148) and (18. 20 dz \qh\ + \h%U{z) . )+^*+)2lr} rnD h\z'2)}z Q )\ = . (18.149) In order to be able to extend the even and odd solutions to z = 0.^ j O e .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+. we have to match yA(z).^ + ) 2 + igln2(zz+).y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .
\ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+.420 CHAPTER 18. t h a t the Stirling approximation is amazingly good even for small values of the argument). of course.125a). We see therefore. using Stirling's formula (and not the inversion relation) — we have —(1.e. l/2 ~h\U{z \(zz+?h\ (18.125a) and (18. This is the aspect investigated by Furry [102].152b) [\(q 1)!] V2 K7: hi Here q/A was assumed to be large but we write [\{q — 1)]! since this is the factor appearing in the solution (18. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.126a). the results necessarily require adjustment or normalization there in the q—dependence. However. The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q. In a corresponding manner — i. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. Stirling's formula is the dominant term of the asymptotic expansion of a factorial or gamma function and thus assumes the argument (oc q ~ 2n + 1) to be large (it is known.ZQ) . 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.154) .
3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.103) and (18. The relation (18.157) is used in Example 18.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.99). \hl.155) (2z. Comparing for z —• z+ the W K B solutions (18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.158) Applying the boundary conditions (18.159) 7 i(«i) r(9^3) ^2~ j(9l) .e.e.e.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. (18.154) with the typeA solutions (18.n('>2o) 3/WKB(Z) =1VA{Z)I (18. 27T1/2 9/4 1VA{Z). (18.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .18. (18. linearly matched) W K B solutions. this result is somewhat imprecise. and is shown to reproduce correctly the result (18. (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large.5 for the calculation of the tunneling deviation q — go by using the usual (i. . However.104). as the results also seem to support. Returning to the even and odd solutions (18.157) Since the factorials [^(q — 1)]!.153).K?+1). . 7 of t h e matching relations y^NKB\z) i.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. we obtain in leading order (i.143) T^^*)) 27T (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.
G2 K [i 4. n + uY 2c „ ^6 8V2c2 (18. p. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\.422 CHAPTER 18. .t The elliptic modulus k (with k' = 1 — k2) and an expression u appearing in the integral are defined by b2 1 U kz = ^ = l + u so t h a t = u=S^q hi = GV2q. F..163b) The integral hi?) .149) near z = 0. 60 and 361. (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 (18.161) we can rewrite the integral as r I2{z) = = dz^{a2 . formulae 220.19. . D.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. Byrd and M.z2)(b2  z2). Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables.05.163a) (18. p.148).160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.iV2 ! i . (18.164) / 2 (0) z ca [(1 + k2)E(k) .. Friedman [40].k'/2. 213.. 3G2 12V2c ^See P.
.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. ± . .( ft 7. n = 0. (G\ . which implies small G2.1 " e " O T (18 169)  Y^V / 4 tfo.1.) = 2 \. [4] misses an n—dependent power of 2 in the result. .165) in agreement with Ref. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 .± ' 3G 2 ' T ( ^ (18. /2n + l 2 ± — .1)]' Gfo.r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 .166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o .. ^ ^ .3 )l ! .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n . n.18. [167].2. 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 . .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref. 1. ... where the result is shown to be (with q ~ q0 = 2 n + l . The nontrivial expansions are derived in Example 5.
Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. and obtain ( * .143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.3 concerning the exponentials.n = 0. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.173). (18.3. with the help of Eq..6 Eigenvalues and level splitting We now insert the replacement (18.157).i ) ] ! e ~ s A ' .h2) and hence the splitting of asymptotically degenerate energy levels.175) ~E0(q0. Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec. (18. we obtain for E(q0.^ W ^ / 4 t o . (18. (18.. h2) the split expressions 29o+1/i2(—W2 h6 (18. we can therefore impose the boundary conditions at z = 0 by making in Eq. 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).172) into Eq. 18. . ..»W.143) with qo — 2n + l.1.h2) + (qq0)^=. 2 . (18.. (i*™) We obtain the energy E(q. . (18.424 CHAPTER 18.159) — on using once again the duplication formula in the same form as above — 2"/ 2 (2h% 1 V/2 7 Comparing this result with that of Eq.174) .106). (18. 18.3.
e.3 The Double Well Potential where £0(<7o.174) and (18.106).176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB. (18.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.n = 0. 2h4' i. i.178) "In S.e. The result (18.2. i. which evidently supplies some correction terms. Combining Eqs. K.^ 2^ h* .18.e.h2) = E_(q0.2 = 2 . 1 7 425 2 ..e. the following expression for large q ~ 2n + 1.7 =( e N n+1 .157). T \ *n ! 1 ' (18.e..V2^\ \n+ i .q2 7..143) with (18. (18. the level splitting..^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 . 1 u2 c 2 (3 9 2 + l) ^ « b . the difference between the eigenenergies of even and odd states with (here) qo = 2n+l. for finite h2. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .1.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^. the pure WKB result of Banerjee and Bhatnagar [14] (i.h2)E+(qQ. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k.fr2) is given by Eq. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i. . i n U n / ! Thus AE(q0.
and A there with c 2 / 2 here. (18. the Furry factor corrected WKB result follows automatically. however. This. as is also explained by Bhattacharya [23]. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. as we do here. h2)\mo=i = .. (18. (66). that if this symmetry is taken into account from the very beginning. Then the splitting A ^ W B of Eq. (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. We see therefore.2 / 7. and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18.6) to yield an improvement of WKB results for small values of n. h = 1.178).180) h8 E0(qo. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. we identity the parameter k there with ft. ** Comparing the present work with that of S. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. H.h2 <> — h2." Of course.176) together with Eq. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn.6 \9o/2 e~h&l^\ with (m 0 = l) (18.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. Furry [102]. .^ 2 + ^h2 If in addition the potential is written in the form A / . we would have obtained the result with E. Bhattacharya [23].** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). K. Eq. h4 and c2 replaced by 2E. (18.4/4 here. A>0.3)).h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o».2X2 V(z) = \z2tL) .426 CHAPTER 18.181) "W. 2/z4 and 2c2 respectively (see Eq.
ZinnJustin and U. . Patrasciociu [110]. Bhatnagar [14]. (2. J. these are presumably not of much interest in physics (for reasons explained in Chapter 20). M. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation.3. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. M. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. however.18. m 2 <• p 2 / 2 . Lee [98].^ *See E.15). K. P. Mansour and H.34) and (E. 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. D. Liang and H.181) therefore implies the correspondence rj1 <> A/2. H. D. Gildener and A. ^To help the comparison note that in J. J. first paper. Equation (18.—Q. the level splitting is (with h = 1) This result agrees with the ground state (go = 1) result of Gildener and Patrascioiu [110] using the path integral method for the evaluation of pseudoparticle (instanton) contributions. formula (4. so that by comparison with Eq.^ Thus for the case of the ground state Eq. (18. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. (18.182) implies AlE(l. and g2 is given as g2 = r]2/m3. The comparison with Eq. Friedberg and T.W. Banerjee and S.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26. D.c2 = A/2. See R. 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.* A similar correspondence is also found in the case of the cosine potential. J. ZinnJustin and U. MiillerKirsten [186].§ 18.88) h4 = 2/x2. (18.h>) * 2V2^^^y/2e^3/3A.11). Eqs. §J. We considered above only the symmetric twominima potential. Jentschura [293]. W.3 The Double Well Potential 427 a form frequently used in field theoretic applications.
has also been the subject of numerous numerical studies. B.428 CHAPTER 18. Ashbaugh and Harrell [12] and Harrell [130]. Pradhan [182]. as he discusses. A reasonable. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. P. Simon and Wightman [178]. Perturbation theoretical aspects are "See e. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. can be found in an article by Coleman [54]. Santi and N. 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. as some relevant references with their view we cite papers of Harrell and Simon [129]. N. As references in this direction. 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. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. The double well potential. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. [19] — that the expansions considered above are asymptotic. for instance. mostly in connection with instantons. In fact. like those for anharmonic oscillators.g. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. [132] and Loeffel. There is no reason to view the anharmonic oscillator differently. Mahapatra. in both symmetric and asymmetric form. we cite papers of the Uppsala group of Froman and Froman [100]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. . Martin. though not exclusively numerical. 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]. 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]. B. 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.[131]. Very illuminating discussions of double wells and periodic potentials.
[122] as 13 1. Datta. Example 18. fc ~ 1 . Srivastava and Varma [24] and Bhatacharya and Rau [23]. Biswas. Gutschick.u)1'2 = V2u{ 1 .187) We now reexpress various quantities in terms of u. q ~ 2n + 1.\G\ + • and 1k2 =2Gv/2q4. (18.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14].f + • • •) 2u\ . k! (18. Thus k' and hence k' = v ^ ( l . Bender.192) 'F 12 16 . + (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. which is assumed to be small.183) S o l u t i o n : We have k2=1^. 1+ u : Gy/2q./4 E{k) = 1 + • l n In (18.^ + • (18.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .18. (18.189) =2u 2u2 (18. 2 a..190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. We obtain the expansions from Ref.186) (18. T h e r e f o r e it is convenient t o deal w i t h this here.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 .2G^2q + 4G2q .191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . We have 4 ~k' 2u(l . Cooper and Strottman [221]. Saxena.G2q + . 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. X +2 + ' ^ (18.
195) With (18.4.3 « .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .3u(6u .3u(6u .1)(1 .u){24 .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .3) = = ~ .2{«(1 . ) J 4 (18.1)(1 . (18.« 2 + 3u] In [ .13)}.{ M 2 ( 1 .13)} + ~u3(u .3).u) + 2u} + i u 2 [ 2 + (1 .3u(6« .u)} + — u ( u .u){24 .194) and (18.18u 2 + 39M} + 12u 3 (u .13)} + ±u3(u .3 u .u 13 \ 1 ln(± k> fc'2 +  (18.u ) (^){« + 5^ 1 )} + 5. (18.— ) . 1 3 « 2 = . 8 + 9u 2 .194) ln '^J + 2 In i/2u )+H l + 2u(l . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + . 8 .^ L ) H[(i _ U ){4 + 3u(l .u)(u .1)(1 . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3) .197) . .e.196) Now consider the last line here without the factor 2.u)} u{u(l .u){4 + 3u(l .3 y .3) .u){24 .u ( l .3) 1 + In ^ . (18.u) 2 {8 .«) + 2} + V [ .u) + 2} + i u ( u .{l u ( l u) / W 2 + 2} .« ) ^ { 2 4 .1)(1 .u ( l .3 u ( l . 8 + 6 u 2 .u 3 ( u . up to and including u2): u(u .3 .u)} ..6 in the denominator and in the last step pick out the lowest order terms in u (i.6w2 + 13u} + 12u 3 (« .3u(6u .193) 2u(l .u) 2 {24 .430 and CHAPTER 18.195) we obtain E(k) 1 + ln uK(k) .a = (l«)(jl 2u/ 2 ) .u) + 2}] 1 H 1 o u(u .13)} + 12u 3 (u .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .
4u .173) t o be evaluated. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / . .] in (18.3) .199) We now return t o the expression on the left of Eq.197) we now obtain 4 1 • E(fc) .148) (for particle mass 1/2 and in dominant order) is given by H.2 /o 2 431 (18.^HfHGH Example 18.zo) HWKB normalized _ ( h* y^eM. + iu^Zu2)^~u+ — +0(u3) l . (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . 2 . we have.. Eqs.128) and (18.128) and (18.u H I .«){4 + 3u(l . (18.6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. (18. (18. ' 8G 2 q = 4 2 q ln f 26/2 \ o . VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .2 l_ln(1 JL\ V {2.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs..4u ..196) and (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.3 The Double Well Potential Now consider the bracket [.uK(fe) ~ 1 + In I — = Mu2(3u.e.198) 9 «. From (18. V^LJ 4 2 16 ^ .3).155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z). we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . where w±(z) — h±(z — z±). l2 = ^ ( l + uf/2[E(k)uK(k)}. [(1 .e. / to + 3GH~ \ 8 . (18.18.it)} .196).2 (18. i.200) Remembering that u = Gy/2q.195) with (18. i. Inserting here from the above expansions the contributions up to and including those of order u2.2{w(l u) + 2}] = u(3u2 .
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.143). 82.204) .** oo / dwD\. (18. normalized = We see that the normalization constant is independent of a quantum number. (26.146) and (18. (18. W.zo)./ . The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. i. Then ((r. Oberhettinger [181]. are given by Eqs. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. (18.e.432 CHAPTER 18. Magnus and F.Aw) = s/2i 2("!) V (41) and [(9 . (18.hA/2 = h^/4 this implies . Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. Example 18. we obtain (for q reasonably large) 1 (91) and hence Joo Hence with h+/V4n = (h4/Sn2)1^ W2TT SI {[UqIWH"1)}2 2TT h2+\1/2 ~VWKB \ ~^T f 4^ \V4ir ^g^ J VWKB .203) Different from above we now continue the solutions (in the sense of linearly matched W K B solutions) across the turning point at zo in the direction of the minimum of the potential at z+.f hi\ VWKB.g.l)]![i(0 . ° \Edzy/\EV(z) V(z)V4 VWKB.*o)/ ^WWKBOO (18. Solution: The turning points at zo and z\ on either side of the minimum at 24. Eq. i.35)). 93. 1 7(l.e. and obtained as Eq. by using the periodic WKB solutions below the turning points. 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 [ . We start from Eq.3)]!23<«D [5(91)]' >A.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.158). pp. oo .147).
g.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds.205) We now apply the boundary conditions (18..e.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.g. i.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1. We could.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) . Sec. from ZQ. approach the minimum also from the right.. 7T\ + ^ 4 7r\ dz . . i.210) Choosing the point z to be z+.r + ^=co S [r.18..3 The Double Well Potential We also note that d dz 433 y±{z) >2+ 1 F = 7 sln h%U(z) 1/4 f X —2 cos 7 2 2^ + 1 \ 1 / 2 71 .212) tt See e. 6. of course. i..iK h\U{z) + / 4 (18.e.207) (18. from z\.e. 1/2 . A.J^Wj +(18.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.2.211) 7 z l .3. Messiah [195]. 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. \h\U{z) 1/2 .^r + + .132) and (18. (18.. Then at any point z € (zo.( . this implies sin cos sin cos Thus e. z{) we expect tt (18.2.6..
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. K..434 CHAPTER 18.215). Banerjee and S. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . 5. (18. 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. Bhattacharya [23]. (18.5.217) 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. .Zi / r.205). (18.215) which results from the factor of 2 in front of the sine in the WKB formula (18.146)).216) Inserting this into Eq.. ' /^i . .214) {«+"i} 0 for q = go = 3. S.e. (18. The prefactor 2\/2h2/n is. and cot< (q + 1) '!}» for q = 50 = 1. In Eq. 1 1 .149). the level splitting is given by AE(q0. and L. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). 9. P.3. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. D. W "Cf. K. .143).210) assume a form as in our perturbation theory.2 .215) we can expand the left hand sides about these points and thus obtain . i. We now see that Eqs.27 9 . . 7 .215) we can insert for 7 / 7 the expression given by Eqs. Landau and E._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . (18. + l ) j } ^ 1_ 27' (18.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. M. The formula thus agrees with that in the literature.90 — =F— for q0 = 1. 7r"y in agreement with Eq. Bhatnagar [14]. (18.209) and (18. (18. Lifshitz [156].213) The integral on the left can be approximated by 1. 2IO/K.148) and (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. in fact (with our use of qo). where u> is the oscillator frequency of the wells. they become c o t { ( .
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.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. 'For a review from this perspective with numerous references see H. W. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. H. However. M. M. Morse and H. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. 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. J. MiillerKirsten [9]. Aly. Case [45]. 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. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful.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. n > 2. f K . also in P. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction.Chapter 19 Singular Potentials 19. W.* However. are not as troublesome as one might expect. Feshbach [198]. in an early investigation Case^ showed that potentials of the form r~~n. Giittinger and H. 435 .
Manvelyan.Q. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. 19. H. MiillerKirsten. D. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. In fact. J. G. which are always worth studying in view of the insight they provide into a typical case and the didactic value they possess for this reason. K. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. Park. R. § 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). S.436 CHAPTER 19. J.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 = 0.2 19. some also permitting further research. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. W. N. (19. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. . that with the hyperbolic cosine replacing the trigonometric cosine). H. Tamaryan. Recently. Liang and Yunbo Zhang [189]. In the concluding section we refer to additional related literature and applications. J. It is therefore natural that we study this case here in detail.2. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. D referring to Dirichlet boundary conditions) in 10dimensional string theory. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions.e. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. Hashimoto [124]. S. for instance. W. Maldacena [41]. Callan and J.2) § C . S. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. Singular Potentials theories in mind. Singular potentials arise in various other contexts. Gubser and A. MiillerKirsten and Jianzu Zhang [226].
and so =4 5 and n ez = r = ~ = e^4fir. 19.3a) In the case of the attractive potential (as in the string theory context referred to above. (19. H. .^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation.g. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. r = 7 e z . using the method of Dingle and Miiller of Sec. 8. MiillerKirsten.4) first for small values of h2 and then for large values of h2.( I + 0. ig h \j g (19. Eq.mo = 1/2 and h = c = 1.28) below).3b) 7 = % j. h2 = ikg. This has advantages compared with higher dimensional cases. R. Liang and Yunbo Zhang [189]. In the following we study the solutions of Eq. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. We observe that with the replacements p2 z^iz. Manvelyan.19. \[\ in the correspondence is nonintegral. h = ei^y/k^.3 below) g2 is negative and hence h? is real for E > 0. 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. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential. The following substitutions are advantageous: y = r1/2cf>. It may be noticed that since I is an integer.Q. and as we have in mind in Sec. It is clear that we draw analogies to our earlier considerations of the periodic potential. J. (19. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . (19.7 that we employed also in previous chapters. (19. This is rarely possible and therefore this case deserves particular attention. W. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. J.
6) terms amounting to 1 „ 2 d2Zu A V (19.2 CHAPTER 19.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. 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 = 0.5) so that Eq. W.e.I . .g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions.7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. —Zu — (Z„_i + Zu+i). Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.Hv (w). J. First we make the additional substitution w = 2/icosh z. MullerKirsten and N. VahediFaridi [10].2.1.11) We follow here largely H. (19.10) i # > = 2h2 2]^v ~r 2 7 2 ~r 2<\^v o o w w dw w Using the recurrence relations of cylindrical functions. (19. 8. w 2 . and H„ (w). — Zv+i). where £>„ : = —_ + i _ + J l . (19. (19. H. dwA w dw { wA ) (lg.i l . i.Nv(w).^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. (19. H.2 A / i 2 . Proceeding along the lines of the perturbation method of Sec. Aly..9) leaves uncompensated on the right hand side of Eq. (19. Eq.7. The zerothorder approximation = Zv[w) (19. The solutions Zv(w) are written Jv(w). where these are the Bessel function of the first kind.438 19.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .
i/) = 2A._2 + {v. v .14) (19.2)*Zi. . it is more convenient to use these relations in order to rewrite Eq.6).10) as a linear combination of various Zv. (19. where the starred coefficients are defined by (19. of course. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.2 ) * ( z / .21) +(i/. I / ± 2 ) = 1.!/+ 2)*4°J 2 ]. .16) ( I / .10) is now particularly simple: i?i°) = h2[(v. v2)*{y2. where We observe that DUZV = Q. v + 2)Gv+2]. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately).2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.u + 2)*Z„ +2 ]. z .2)G„_ 2 + (y. (i/.2. (19.17) Thus a term fiGu+a on the right hand side of Eq.12) (19.u (19. wz The expression (19.v + 2)*{u + 2..15) (19.20) 2)*ZV_2 + {y. (19.i/4)*Z 1 / _ 4 +(u. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . v . However. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z. and this means in (19.19.13).13) therefore lead to the first order contribution jfU = h2[{v. v + 2)*(v + 2. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.18) Now </>(°) = Zv left uncompensated Rv . = §. v + 2)*Z„ +2 (19.13) (19.10) in terms of functions G„ defined by Gv+a = —nZ v+a . v)Gv + (y. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.2 ) * i ^ 2 + (!/. when a or 2v + a = 0.u + 4fZu+A}. The terms (19.
v ( 4 5 a . we obtain 9 h4 = a 2(al) 3 2 (13a .i/±2)*. (19. v + 2)*{v + 2. Evaluating t h e first few terms..v)] (19. i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. . p4(±2) = (i/. i/ ± 2)*.g.24) P2i2(2j + 2){v + 2j + 2.17)).4)(»/2 . v) +(i/.2. . Ri. subject to the boundary conditions P2i(2j) = 0 for  j  > i. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e.26) iy — 2. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9.2 + 2 9 ) ^ 64(^ 2 . P2*o(0) = 0.v) + {v.22) where p2(±2) p4(±4) = = (i/. . etc. (19.l)(z/ 2 . .23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . i/ ± 2)* (i/ ± 2. p 0 (0) = 1.440 CHAPTER 19. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . we obtain t h e expansion .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.I/±2)*(I/±2. v + 2j) (19. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. . j=i.455a + 1291a .25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . i/ + 2j).v + 2y{v + 2. in Eq. i/)] + • • • . Adding these terms and setting the coefficient of Gv equal t o zero. . . we obtain 0 = h2{v.j^0 (19. (I/. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .v + 2)*(i/ + 2.v) + hA[{v. and p 0 ( 2 j ^ 0) = 0. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .v2)*{v2.25)/i 8 32(al)3(a4) . v — 2)*(v — 2. (19.1169)/i 12 .2)(v + 2j .4 + 58.I/±4)*.
7) into Eq.34) . Thus to O(0) in h we have (jr ' = 4>v — cosh vz.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.31) It follows that Dv+2n4>v+2n = 0. Substituting (19.v + 2)<t>v+2]. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z.2.32) (19. and is therefore real for both cases g2 > 0 and g2 < 0.4).h2[(j)u+2 + 4>v_2] h2 [{u. £>„ = — . we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A . 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 .^ + e±(v~2)z.cosh 2z)4>v = = 2h2A(/>„ .30) £>„&.3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions. (19.19. — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. 19. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .33) = = cosh(^ + 2)z + cosh(^ — 2)z. (19.2)<t>„2 + (v. (19. = 0. v . 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. 2 . Thus v is real for small values of  h21. so that J2 (19. (19.cosh 2z)4>. ±2. we can rewrite the latter as j ~ . (19.i . v)(f>v + {v.An(u + n). .v2(j> = 2/i 2 (A .29) sinh vz or e±uz.
> (f> with </v = sinh vz — Seu(z. The use of the symbols (v. F. etc.I/) = 2 A . Meixner and F.h2i *=1 J2 j=i.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions.7 — we refer to Meixner and Schafke [193]. 19. The solutions in terms of cylindrical functions are written *J. 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. 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.h) = <(>„ +Y.38) (f> with (\>v = exp(vz) — Meu(z. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. Arscott [11]. (19. we prefer to adhere to one equation with different solutions. 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. Schafke [193].J/±2) = .I/±2)*. (19.2.1 . > These solutions correspond in the periodic case respectively to the solutions and me^. 8. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z. • (19. v ± 2) etc.h).* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.h).37) where p M (±2) = (I/. W.4 Notation and properties of solutions We now introduce standard notation as in established literature.J7t0 p2i(2j)<f>u+2j.h). CHAPTER 19. Singular Potentials (I/.442 where (I/. . M.
we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. Mi. Me„(z. 17. 'See e. h) = exp{vniri)Me„(z. For the relation of Hankel functions used below. As emphasized in Chapter 8.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(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). 4>v = Hv (2h cosh 2) * M^\z. as we remarked earlier.40) does not.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z.h)±Sey{z. 16. h) are therefore proportional to each other as a comparison of Eqs. h).41) i=l We see immediately that Meu(z + niri. Meixner and F. W. (19. Schafke [193]. i. h) = exp(inwn)MP(z. h) = av{h)M^ (z. which.e. (19. h). as Mev{z.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. whereas the — as yet — unnormalized solution (19.h2) oo = Y. <f)v = N„(2hcosh z) MJ?\z.e.h). W. see this reference p.MP(z.40) Cev{z.h).42). . {2h cosh z) • MJ?\z. h). Oberhettinger [181].h). ' (z. i. (19. (19. 4>v = Hi.43) and hence similarly M^\z + inir. — oo cv2r{h2)e^+2r>. Magnus and F. p.42) The solutions Meu(z. (19. (19. 443 (19.(2hcosh z).h) and oo i M^\z.45) T Below we frequently write the normalized solution as in J.h) = exp(vz) + Y.44) (19.g.19. this normalized solution possesses contributions exp(i/z) in higher order terms.h). Also from the expansion of Jv{z) in rising powers of z.44) implies.
Meixner and Schafke [193].exp(TM"r)MW(. h) . Singular Potentials where clearly MeJQ.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. 130) to converge uniformly for all finite complex values of z. h). the first relation is obtained as follows: 2M<£1 = M^l+M^l. The functions Me±v{z. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19.h) Using further properties like (19.49) Inserting this expansion into the modified Mathieu equation (19. ^ . oiv{h) = (19.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. p.48) The series expansions of the associated Mathieu functions M„ (z.e.g.h) .50) E. i.h) can be shown (cf.h2) = ] T cv2r(h2)e{y+2r)z. As mentioned earlier. H™(z) = e^HfXz). 2M™ = eiuvMi3) + e~ivlrM^\ . Meixner and Schafke [193]. For later essential requirements (i.46) H^z) = e^H^(z). These points are important in our derivation of the S'matrix below. 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.8. s (19.444 CHAPTER 19.47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. . r=—oo (19. i. h) = M™(z. we have the following for a change from v to — v. Mi 3 .h2): oo Mev(z.e. (19.h) on the other hand can be shown (cf. p. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).43).e. MJ>>(0.
Thus now c 2r .T22r[\ A . p.[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. W. (19. 122.* For 'Actually.52) in agreement with Meixner and Schafke [193]. Eq. 117. see J. .53) /i2[A(^ + 2 r ) ] .(/^.^ + 2r + 2 ) 2 ] . + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. + 2 r ) 2 ] . . Meixner and F. 2r (19>51) 2r For ease of reading we give the steps in rewriting this. (39). \) /.2 [ A . p.^ p 2 2 ^ or c 2r+2 2r2 1 (19. Schafke [193]. Alternatively taking the inverse of Eq.19.^ C 2r+2 (19.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . eg = c~v = 1.54) 2 C2r 4 ^ h./ i 2 C ^ . \ .2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] . h~2[\ .51) we have 2r (19.(v + 2r)2} / z .
. M.8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. and also R.2.27) and truncate the continued fraction after the second step. (19. Solution: We put r = 1 in Eq. Here we insert for A the expansion (19. 128(i/+ l) 2 (i/ + 2 ) ( ! / .[v^. but follow our earlier reference H.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.+0^"> < 19 . H.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. MiillerKirsten.. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ .12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. .Q . t For this purpose *We do this in the manner of the original derivation of R. W.+4)!ll_.( " + 2)2] ~ h2 h. (19. Manvelyan.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq.l ) 19.1 for the evaluation of a typical case): hb + 0(hw). MiillerKirsten and N. + 0(/. VahediFaridi [10] (this paper contains several misprints which we correct here). + i).C + l) 3 (i/ . Spector [257].^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). J. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) . Then di 2 1 t" + 2(^1) + • • • . H. J .446 CHAPTER 19. W.5 D e r i v a t i o n of t h e S ..55). Liang and Yunbo Zhang [189]. J. Aly.55 ' Example 19. H.
56) where by Eqs. 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.3a) and (19. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0.19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. p. 22.58) 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 considering the propagation of this wavefront.. Magnus and F. we have W r = ip + ut — 7r/4' . Then y reg = = r rV2M®(z. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. I / + 2 2 l\TT +E ^ i\ E j=i.59) which tends to zero with r. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.57) ^ 1 . In the case of the repulsive singular potential here. . W . (19. (19. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves).j7L0 P2i(2j)exp i[v + 2j + 2 2 .h) 1/2 w) (19. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. and then the continuation of this to infinity. Oberhettinger [181].5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo.
178) to be convergent for  cosh z\ > 1. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z. This means that the origin of coordinates acts • > as a sink. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. AV . h) through the entire range of !&{z).A1"—2 z '+"2 A~~ n or rA .3b) ln • 7T Q"*4' r fg ° Vk (19./i). We require therefore > > the continuation of Mi. Singular Potentials so that when t — oo : r — 0. 4%r g >0 4r^r ' rg See W. '(z.#.3a) and (19.448 CHAPTER 19.A14A. Now from Eqs. '2"4 . but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.(2j)<gH • (19.60) Using the above asymptotic expressions for the Hankel functions.j^Q p2.h) can be shown (cf.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. p./l ~ 2 ~ 2z + " 4 > 0. 2h and the square of this expression implies 2 2 ~ ' A. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. Oberhettinger [181]. Meixner and Schafke [193]. . (19. 17. p. (19. The series M^ (z. Magnus and F.j¥=o Ki(2j)exp(Tmj) (19.h) = exp(Mr^)il4 4 ) (z. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).h) oo j 2 — rV flp^ + ^fc* £ i=l j=i.V + § 9 2 > 4hh* .62) Also with Eq. 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.61) In fact.
1. This interchange in the > solution My interchanges y reg with y+. Meixner and Schafke [193]. A suitable set is the pair of fundamental solutions Me±v defined by (19. i.ijt/4 Rez . Thus we choose . i Im z z = In r/r0+i7t/4 .rl)(r2 .e. Since the real part of z changes sign at r = ro. and in order to maintain this symmetry. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. (19.3b) and (19. 449 This condition. These solutions converge uniformly for all finite complex values of z (cf.1 The domains of solutions. 19.2 The Potential 1/r4 — Case of Small h2 Hence (r2 . the domain r_ < r < r+ — as illustrated in Fig. we have to assign different signs to the imaginary part of z in the two distant regions.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig.63) Thus there is a gap between the two regions of validity — i. (r — r+)(r + r+)(r — r_)(r + r_) > 0. (19.49).e.r2_) > 0 with r£ = rg(2 ± >/3). . p. is satisfied only for r < r_ < r__ or r_ < r + < r. (19.38) or (19. which has to be bridged by using another set of solutions. Originally we chose the sign of m / 4 as in Eqs.19.3a). z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. 19. r 7r for 0 < r < ro.62). 130).
and we determine these coefficients from this equality and the additional equality obtained from the derivatives.64) there to another combination of solutions Me±v by setting: rl/2[aMey{z) = rl0/2[a'Meu(z) a^{rl'2Mev{z)} dr dr + ^Me_. At r = ro the real part of z vanishes and then switches from negative to positive. i. i.z=—in/4. Thus. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}.450 CHAPTER 19. to r^2[AM^ + BM^\ A.l/2Me_i/(T. h) = Me^u(—z. (19. h) (as one can check or look up in Meixner and Schafke [193].)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. these relations can be reexpressed for one and the same point z = —iir/A: rl/2[aMe„(z) rl/2[a'Meu(z) + + /3Mev(z)]z=_in/A p'Me^(z)]z=_i7r/4. . (3) (r)] aA[rl/2Mei/(r)]+/3i:[r. p.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. (19. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19.v{z)\ .65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z. 131).(2)]z=w/4 + 0Me.[ r l/2 M. (3 = a .B^0. Singular Potentials Then starting from the region r ~ 0. as just explained.66) a = (3 .^)] dr z=in/4 dr (19. at this junction we require also the imaginary part of z to change sign. Hence we match the right hand sides of Eqs.v(z)\ —m/4) + ^[r^Me. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me.e.
3b) we infer that (z = l n r + const.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri. A and B. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. (19.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)].) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ . (19.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19.68) From Eqs.68) the derivatives with respect to r by this relation. Then.71) We now determine the coefficients a. and we are left with relations expressible only in terms of z. dr BM^(r)]. (3.65) and (19. (19. (19. we use the notation W\f.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}. (19. (19. Thus we obtain from Eqs. ^ ] = ^ + ^ ^ . (19. = aMev(z) + 0Mev(z). (19. (19. the nonderivative parts (from the first term on the right of Eq.70) by Me'_u (the prime meaning derivative). and the second of Eqs. (19.69) Thus if we replace in the derivative relations of Eqs.65).70) by MeU and subtracting . multiplying the first of Eqs.69)) cancel out in view of the nonderivative relations. with the replacements of Eq.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.3a) and (19.g]:=f(z)*&g(z)WZ) dz dz Thus.19. for instance. (19.64) the relations M&Hz) jz[M^\z)] and from Eqs. (19.
75) With these relations and Eq. we obtain the first of the following two equations.A] = W[Mi3\z). h)/M^\o.. we use the following circuit relation which can be derived like Eq.M^] J™aV. p.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.77) av I 2ism VK \ ctv av ^Thus. h). . 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.(0.^ ^ a „ Q _ . W[M^\Mlf)]W[Meu.74) Moreover..Mevy W[M?\Mev] W[Mev.* « * r r z H . (19.\ \ ( av 2iwKav ^ v . Me. 7T W[Mev.73) We now use Eq.57) and (19. (19.j] given in Meixner and Schafke [193] (p. = IT <*„. using Eqs. Meixner and Schafke [193]. (19. i. TV W[Mev.e. M^4)] + W[MJ>3).Me^} (19.4] = . B = J 2 L _ e .71) we obtain in a similar way A B = = W\MP\ Me. and the second is obtained similarly: W[M?\Mev\ W[Mev.M^] W[Mev.76) .g. Singular Potentials the second resulting equation from the first resulting equation. (19.u}W[Me. (19.46). the leading terms of the respective cylindrical functions. 7T [1. (19. (19.] W[Mev.^ 47 9? MJ>4)] MeV] [3. (19.47) we obtain (av defined by Eq. av{h) = Me.3] = [1. we obtain immediately W[Z.Mev]' P l ' From Eqs.46)) W[Mev. i. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.u.58) and considering large values of \z\. W[MJ?\ M^] W[Mev.48) (or cf.452 CHAPTER 19. or obtainable by substituting e.M^} W[Me„. .Mev] = = = av.e.M^]= IT e~iv*av. and Wronskians W[M^\MIJ)] = [i..4] = . 7T (19. 170/171).
83) The Smatrix element is therefore given by ifc*"* . (19.46) together with the relation Mev(z.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). (19.(l)le^}Pi(cos 9).h) = Me_u(z. in Chapter 16).2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z. (19.85a) .l) ' We note already here that with the substitution R = ei7r7. (19./ ( .19.g. eii/7r/2 (19.h). Then f' . as we recall from earlier considerations (e. The superposition of an incoming plane wave and an outgoing radial wave is written and reexpressed in terms of partial waves for r — oo » with z = rcos 6 as: 1 tCT" "1 OO e *« + /(*)!_ ~ 5 > ' c < f c P . (19./) = (l)1/2R2^2i^)ei^)U^.f c . Z ) e ^ ] .80) 2^fc (1981) where \7r/ i t sin VK /(*.78) we obtain Ay reg « _ . B=tfD'"".h) + BM^\z. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. Z ) e ._ [ / ( f c . f c r .79) where the second expression follows from Eq.(ite™)" 1 { 8 J ' f(k. (19.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).
Thus in terms of the (3) variable z. R 2i sin TXV.58)) in the limit r — 0.2 The repulsive potential and the various waves. we can define respectively as amplitudes of the incident wave. 19.56). (19. h) (see Eqs.Smatrix in the form s i n 7T7 Si = CHAPTER 19. R — — = 2i sin ivy. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u).z—>oo 2r \l/2 i(is+l/2)ir/2 2hir cosh z r> \ „2ih cosh z RU + i [Re —2i/icosh z R Thus.56) to (19.454 we can rewrite the . (19. we can write this (multiplied > by 2ismvTT.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.58)) 1/2 2r e~ 2hir cosh z. Singular Potentials r(i+k) sin 7r(7 + u) (19.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. (19. (19. cf. and recalling that yreg is proportional to a function Mc> {z. 9?. 19. (19. Fig.2. and with the left hand side following from Eqs. We can rewrite Eq.86) . V(r)=g2/r Fig.
89) . in which v is complex. This implies basically the evaluation of the expression R of Eq. unitarity is preserved. therefore we cite it from Meixner and Schafke [193]. and then exploit it for the evaluation of the quantity R. unity minus reflection probability = transmission probability. By inserting in Eq. (105) to (108). Gubser and A.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.e.88) where the factor Me„(0. S.84) of the ^matrix.( i J e ^ ) " 1 ! 2 = R2 + ^ .e.28)) and R = ey real. (^)l4lu{h2)Jir(hez)J±u+i+r(hez).2COS2VTT = (R. 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.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. Manvelyan. i.19. Eq. M^l)(0. however not here. The expansion is: 4 r ( / i 2 ) M S ( z . Hashimoto [124] and R.2. Jv+2r. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. but in the 10dimensional string theory context." \R~Ji\2 1  tO _  2 i s i n VK\2 [  .88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence.h) is (cf. 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).79).„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* . H. (19.h) The function Mo (0. (19. (19. i. " S e e S. this expansion is given as 1 oo MJ> Hz. 180). 178). The latter is what happens in the S'wave case of the attractive potential. W.e. Eqs.Q. MullerKirsten..e™*/R\2 ' ' . (p. (19. Liang and Yunbo Zhang [189]. i. h) serves the use of the same coefficients as in the other expansions. J. (19. / i ) = Y.In Meixner and Schafke [193] (p. J.** 19.6 E v a l u a t i o n of t h e S .±)2 + (2sinj/7r) 2 . i.e. l=—oo 1 1 lite*"* . (19. if R becomes a pure phase factor and eil/7r a real exponential.m a t r i x Our next task is to evaluate the expression (19.
(19./>) = Co> 2 ) £ 7 " = — OO j ± > .55).h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).3 9 ^ .94) .49).91) Inserting here the coefficients given in (19.e.g./i) with the help of Eq. Here in the case of the repulsive potential. As a matter of introduction we recall the definition of the coefficients with Eq. + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . e.g. h) in many different ways. Eq. /n ^ .2_4)(l/2_9)iv2y) + <AA. r = 2 as a check).456 so that in particular for z = 0 oo CHAPTER 19. setting r = 0 in Eq. (19.49) from which we obtain for z = 0: Me„(0.6 2 ) m 6 (^_i)3(z. Singular Potentials 4r (h )M$(0. Next we evaluate M^1}(0. (19. The formula (19.90). ^ (19. r = 0 and 2. (19. h2 is complex but v2 is real (cf.28)).90) and choose r = 0 (one can choose e. (19. /=—oo (19. We require the power expansion of the Bessel function of the first kind which we obtain from Tables of Functions as I)EH^(T)Thus.. u»^.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. (19. by allocating different values to r.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0. i. we obtain M&(0. one obtains *.
IV 2 2 2(v =F 3i/ .96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy.9) V2 1 fh ±v 1+ 2 + • (19. ..] 2 \v i + {v2 .. In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.111) / / i 5 4 3 1+ 4i/ 2 + +' (z.1) 2u(4u + 15i/ .l) 5 (19. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).19.7) F " (i/±l)2(^^l)(i/24)V2.93) and the coefficient expansions (19./I) h u .12i/2 + 64u .2 _ ^ 4 ( ^ 2 _ 4)2 (19.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .95) These expansions imply for the quantity R: I/! 2v ri1 i . one obtains M£!(O. 4(z/ T H ^ 3 ..96) With this explicit expansion we can evaluate the S'matrix.2^ 2 ± 59^ . From expansion (19. _2_(h\2 .4)(i/ .4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' .1 ) 2 ( I / + 1 ) ( V 2 .2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.32i/ . + # "+" ( I / . We observe that this approximation involves v and h4 and hence is real.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .
7 CHAPTER 19. In this case R — R* and so 1/i?2 ~ Ofji4). (19.84): Si ST 1 (11/R2)(11/R*2) .100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.2. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case.103) Then SiSt (1+5) i .R*2 ^ * #2 \ 11 tf2 R* (19.^ i 4 2+^ # 1 / •. This is the case of real . (19.98) (cf. and sin 2TT/3 = ft/. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.101) where / is complex and g is real. _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).101) is > zero. (19.9 / .28)).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/? }]"' (19.102) (19. Then cos 2TT/? = 1 . Eq.104) We now consider the case of a — 0 implying that g of Eq. 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.458 19. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3.28) that v is real. (19.eiun/R2)(eiu*n ~ (19. (19. (e iun e~iv*^/R*2)' (19. where n is an integer and a and (3 are real and of 0(h4) We have with Eq. + ( i + s w (E 4 .
f W . At = 2i[l + 0(h2)}. Manvelyan.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].1)\} 2 h \2 iv 1 + Au {y .Q. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .SiSf » 4 sin7rz/ R (19.19. Hashimoto [124].107a) In this result* we now have to insert the expansion of the parameter v given by Eq.e. (19.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. 1 29/ ^ (1 .105) sin7r/3 (19.^ ) ~ 2 2 459 1+ 9 / • & ( ! . and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19. In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the threedimensional case we are considering here.28). We can also evaluate now the amplitudes (19.106) and evidently violates the unitarity of S. Hence with Eq.6C .61n/i} + ~9~ 0(h8) (19. J.102) we obtain Sis.3.3b)) h2 is real for E = k2 > 0. J. Gubser and A.577. = 1 . (19.61n2 . 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 . Liang and Yunbo Zhang [189] (there Eq. H. "This is the result effectively contained in R. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). Here in t h e case of the attractive potential (as remarked after Eq. Magnus and F. (19.107c) Here h2 is actually Vh4.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . p.107b) = Ah2[l + 0(h4)}. . S. ( I l l ) ) in agreement with a result in S. W. With t h e help of Eq. MullerKirsten. i.l ) 2 2 \) +0{h«) (19.97) this can be written A 4TT2 [v\{v . Oberhettinger [181]. (19.
Tamaryan. (19.1. i.3.h) 10. we again make use of the replacement (17. 19. S.460 CHAPTER 19.3.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z .109) ^We follow here mainly D. H.e.3 19. Park. Miiller and N.H. h) for large h and upwards I = 0.3 The function q(l.a }tp = 0. 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.4 Fig. Aly.h) (19.108) For the large values of h2 that we wish to consider here. Singular Potentials 19.26) and hence set a1 = 2hl + 2hq + A{q. J.$ q(l. completely different solutions. W. W. Wannier [278]. 2 2 a = [l + 2 (19. K. J. N. where the part of interest here is based on H.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5matrix obtained above by using the asymptotic expansions for large values of h2. H. 2. MiillerKirsten and Jianzu Zhang [226]. We are again concerned with the associated Mathieu equation given by Eq. . VahediFaridi [8] and G. This is therefore a highly interesting case which found its complete solution only recently. H.
19.h*y_(l.110) a where y+(7r) is the solution evaluated at z = TV.21b).h*y (19. We begin. 19. we defined the Floquet exponent v and observed with Eq. which therefore cancels out or can be taken to be 1. (17.2 T h e F l o q u e t e x p o n e n t for large h2 In our treatment of the periodic Mathieu equation in Chapter 17. . y'_(0) = 1. The replacement (19. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v.3.§ a W[y+.5) that this is given by the relation cos iris = ^ M . (17. J4(0)=0. See also Eq. (19. With the boundary conditions y+(0) = l.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively.e.63).111) we have W[y+.19.y} where VF[y+.e. which is even about z = 0.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. (19. (17.yJ\ — 1 and cos TTV . with an essential mathematical step.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation.l + 2y+(l.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. The behaviour of q as a function of h for some values of / is shown in Fig. (19. and a is its normalization constant. however. y(0) = 0.3. i. i. We also found the boundary condition (17. to which we have to return in this subsection.21b). y± oc A{z)e2h sm * ± A(z)e2n sm z . Therefore the second term on the right is nonzero. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.113) 'We are considering nonintegral values of v.
B. This paper contains the normalization constants of large/i 2 periodic Mathieu functions.114) a a Our first step is therefore the determination of the normalization constants No. J. W.A(z)e2h sin 2 sin z sin z (19. i. 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.112) from evaluation of Eqs.e.114) at z = 7r/2. (17. 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 . We now obtain the expressions needed in Eq. K [ (z)}2hsinz (19.462 CHAPTER 19. (17.39 2 14 /i 2 + (19. 1 1 Or see R.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0). Dingle and H.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. which is \/2 at z = 0. in the second expression 2h is obtained in addition from differentiation of the exponential factor). (17.46) 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 . B. See R. Miiller [73]. W.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. (19. J. .118a) "We emphasize again: For integral values of v the normalization constants are different. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. (19.43) we see that z = 7r/2 implies W(TT/2) = 0.117) a a Referring back to Eq. Miiller [73]. (19. and hence with Eqs. (19.32). Dingle and H.
(17.119b) 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].117) are known from the matching of solutions in Eqs. Tamaryan.87 2 14 /i 2 [ i ( g .119a) This formula — derived and rediscovered in Park.118b) The factors a.i ( g +3)]! 1 24/i {2hl>2).62a) and (17.3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .1Q11QM cos irv + 1 ~ •==— e .62b) (see also the explicit calculation in Example 17.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . a in Eq.\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)]. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. . (19.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. (19. (19.! ) ] ! [ . (19.h . p. 210.2g + 3) 27/i ( 3 g 2 .! ) ] ! [ . With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .19.3).112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .
—h. (19.3. —z). V— 2in (19. of course.z). h.122) that given one solution ip(q. z). z) are TP(q. . h. (19.3. h. the expression (19.122) Correspondingly the various solutions ip(q.h)iP(q. tp(q. we can obtain the linearly independent one either as ip(—q.h. h.4.z) i/>(q.e. Vcosh z \ 1 .. z) = ip(q. for h — oo).h.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. —z). z) can then be written as .z) = A(q. h. —h. .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/\ (19.108) and set ip{q.h.464 CHAPTER 19.h) =?==. Then straightforward integration yields > inq 4 „.2. 17. With the solutions as they stand. dA 1. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.z) or as tp(q.h. h(q. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ .z)exp[±2hismh z\.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.109) remaining unchanged. .3.? s m h z) Vcosh z (19.h.h. „ 1 A „ —A cosh z——H . h. each with a specific asymptotic behaviour. We define these solutions in terms of the function Ke(q. h.z). z) = ip(—q.123) We again make the important observation by looking at Eqs. i = 1. it is still a solution but not with the symmetry property ip(q.2 we insert the expression (19. In the following we require solutions H e ^ ( z ) .120) to (19.109) into the equation (19.3 Following the procedure of Sec.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.h).120) The resulting equation for the function A(q.
as we saw earlier. It follows t h a t S e t t i n g s i n h 77 = t a n 0 .126) For large values of the argument 2/icosh 2.3 The Potential 1/r4 — Case of Large h2 465 Instead.<5±1) (19 125)  (the expression on the far right in leading order for h2 large). we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr.h2). 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]./h). / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo. 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 . (z.19. (0. in performing this cycle of replacements the function picks up a factor. i. >• * F $ 0 * e".WJ " C o n v e r t i n g t o n o t a t i o n of W a n n i e r [278].e.45). . : 7r/2. [ \ .127) sin z T /'1 M TT\ — <>{ 4 . We observed that these satisfy the relation (19. (44) of W a n n i e r . 2j<.e. av{h2) = ^(0J ^ Mi i . In order to be able to obtain the Smatrix.fi2) can be reexpressed in terms of Hankel functions. the Bessel functions contained in the expansion of the associated Mathieu function M{.e. (19. Me„(z. (19.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P .125) is t h u s seen t o b e i d e n t i c a l w i t h t h e p r o p o r t i o n ality factor in E q .^ s i n h 7?)]g. * ** = > rojKe( "./i 2 ) M (19. w h e r e Iv := I drj\f a + 2k2 cosh 2*7" J0 I n s e r t i n g t h e e x p r e s s i o n for a. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. i. achieved with the help of the Floquet solutions Me±v(z. . we h a v e 77 = 0—> 6 = 0. o u r h2 is W a n n i e r ' s — k2 a n d o u r a 2 is his a w h i c h in t e r m s of o u r p a r a m e t e r q b e c o m e s a = 2k2 + 2iqk + A / 8 . i. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. This is.h2) = OvMPfrh2). One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO.
z): Re(2\z.g.q.h. (19. HeW(z. 3te»0./i).q. (19. h.q.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.130b) r_0 rll2exp\g/r (ig) ' 19. (19. In this way we obtain in the domain .h) = e(z) exp ihez ./i) Be^(z.h) He (z.h) Be^(z.q.466 CHAPTER 19.130a) Re^(z.eW(z.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.130b). 3te»0.q.h).q.h) (3) = = = = Ke(q.vr .q.z).h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.3.130a).g.4 T h e c o n n e c t i o n formulas We now return to Eq. (19.q. He (z. Singular Potentials Note that the sine part is nonleading! Retaining only the dominant term of this expansion we have (with z — 2h cosh z) > Me±l/(z.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T . KeW(z.h). exp[—ikr — in/4\ kr Re{2)(z. (2) (19. (19.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H.
3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z.g./i)./i 2 ) = Me_j.li 2 ) Oil. Substituting the Eqs. h) = sin 7r(7 + v) He (1) (z. q.130b)). (19.i ^ / 2 H e ( l ) (z> ?) (19. (19.h) is asymptotically small (cf. (19. we see that the solutions (compare with Eqs. (19. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e . — sin 777 H e ( 3 ) ( 2 .h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. Eqs. (19.132) These relations are now valid over the entire range of z. (19.h2) Me„(z. z) are proportional there.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. q.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.q. and thus can be matched in this region. g .« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e . h) = Ke(q. h.130a). these are exponentially increasing there. z) and He (3) (z. t With Eq.132) into Eqs. q. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .h).h) sin irv = sin7r(7 + ^)He (3) (2.(z. and He(*> exponentially decreasing. (19.135) sin irv 'This means.131) where the second relation was obtained by changing the sign of v in the first.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. .g.q. we obtain in the domain tfcz <C 0 the relations: Me./i) — sin 7T7 Re^(z. h.g.134) Considering now the first of relations (19. Changing the sign of z.(z.q. h) = Ke(q.129)) He (2) (z. / i ) + sin7r(7^)He(4)(2.133) In a similar way one obtains the connection formulas — sin nv HeW(z.19.134) in the region where the function He^(z. h) — sin 7T7 He^(z. (19.h2) a. q./i).
137) We see that this expression agrees with that of Eq.119a) or (19.133)) — sin TXV He^4' (z.135).119b) as „4/i 1 + COS TXV ~ rrl ( g . (19. (19.l ) (l„—ikr e From this we can deduce that Si (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV .l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr .119a) and (19. which in our case is the solution He^4^.m a t r i x From Eq. where 8i is the phase shift.1 .Sin TXV. Thus here the S'matrix is defined by (using Eq.135). (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". Wannier [278]. (60) in the work of G. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( .( .l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. cos TXV + sin TXV. Wannier did not have the relation (19.5 D e r i v a t i o n of t h e ^ .135) is the equation corresponding to Eq. of course. The quantity 7 is now to be determined from Eq. (19.119b).e. sin TXJ.(19. (19.468 CHAPTER 19. and therefore remarks after his Eq. is only the dominant term for large h2} 19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. H. Squaring Eq.e'i7ri 1.139) ''Equation (19. q.3.133) we can now deduce the 5matrix Si — e2t51. .136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs. Singular Potentials The factor on the left.85b) obtained earlier. (19. (19. i. The latter is defined by the following larger behaviour of the solution chosen at r = 0.
e.8h COS From Eq. 19. the real part of v must be an integer.141) COS TTU irvj — i s i n WR s i n h TTUJ. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. .140) From this we obtain the absorptivity A(l.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h. (19. as a comparison with Eq. i. for the attractive potential.139) could actually be neglected) implying  cos m>\ = cosh nvi.§ 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.19.28) shows.109). since cos TVU is large. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. h) of the Zth partial wave. (19.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. (19.4 The absorptivity of the S wave (attractive case). 9/2 (91) with the approximation q ~ h obtainable from Eq.e. A{l. i.e.138) we obtain.137) together with (19. imaginary part uj (so that 1 on the left hand side of Eq. T h i s is r e a l for UR a n integer. 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. (19. This is different from the behaviour for small values of \h2\. we can approximate the equation for q ~ h (i.
Lew [128]. In essence. N. Rosendorf [225]. D. "G.107b) and (19. Esposito [88]. Partial or other aspects of the weak coupling (i. N. Handelsman. H. S. VazquezPoritz [61].^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently. A. 19. Treiman [271]. H." as well as other aspects. Paliov and S. 11 J. The diminishing fluctuations of A(l. Bertocchi. B. small h2) case have been considered by some other authors. D. Cvetic. Y. Lii. YuanBen [285]. Eden [46]. see also M. Limic [177]. Jones [78].** 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. This is sketched schematically in Fig.4 (there are tiny fluctuations in the rapid approach to 1). Fubini and G. J. C. L. one may expect corresponding results for other singular power potentials. Furlan [20]. Cvetic.139) 2Tr(16h)q /32\ f e Thus with Eqs. in further contexts beyond those already mentioned. such as phase shifts. I r a n [61] and M. 19. Pao and J. Tiktopoulos and S. Challifour and R. **R.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . W Some of these sources can be helpful in further investigations.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. A. Dombey and R. Apart from the papers already referred to. A. Masson [191]. and G. it seems it is not possible to guess from the above the result for the absorptivity of a singular potential with an arbitrary negative integral power. (19. With the exploitation of perturbation solutions of both the periodic Mathieu equation and its associated hyperbolic form for both weak and strong coupling (more precisely h2) together with corresponding expansions of the Floquet exponent it was possible to go as far as the explicit calculation of the S'matrix and the absorptivity.e. Pope and T. **A. N. H. (19. Apparently this is one of the very rare cases which permits such complete treatment.142) we can see the behaviour of the S'wave absorptivity as a function of h. .k) in the approach to unity are too small to become evident here. F. S . in fact. Lii and J.P.470 CHAPTER 19. and. E.4 . Shabad [248]. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi .
J. W. Dingle [71]. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. 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. Muller [74]. also convergent expansions are known and a lot of literature exists on the equation.g.* In fact. This is indeed a remarkable connection. f R . *R. the asymptotic solutions in different domains all have the same coefficients. Wu [18].* However. We concentrate in this chapter on the large order behaviour of asymptotic *C. Dingle and H. 471 . and how this is related to the level splitting — in fact we shall see this in both cases of the cosine potential and that of the anharmonic oscillator. B. We shall also see explicitly. 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.1 Introductory Remarks The subject of the large order behaviour of perturbation expansions — meaning the study of the late terms of the asymptotic expansion of some function with a view to extracting information about the exact properties of the function — received wide publicity with the publication of the anharmonic oscillator studies of Bender and Wu. T. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. Bender and T. e. [19]. B. M.
20.2 and below. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. 5. r for p = 4. 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.2.6. 8. Since the method is also applicable to convergent expansions. 8. Presumably this has never been done so far. 19. 10 12 14 16 18 20 Fig.1 The function ur = pr jr\ vs.1 and cases in Sec. Fig.7. This approach leads . 20. as evident from Example 17.6.2.8.2 The function ur = r\/pr vs.472 CHAPTER 20.
as we discussed earlier in Chapter 8. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E. T V lim h2N 2 n EQ?) .2) or.3) n=0 Thus for any given value of g . 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. even if each En is known.J2 2n h E (20. for instance. and the expansion is said to be Borel summable.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). For an authoritative discussion of these aspects we refer to Dingle [70]. The behaviour of such terms is illustrated in Figs. The formal perturbation expansion of a quantity E(g ) .1 and 20.1 Introductory Remarks 473 automatically to asymptotic series. this term represents the size of the error.e. 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 traditional method of using an asymptotic expansion is to truncate the series at the least term. 20. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20.e. in the power expansion of the exponential function). and the integral E(g2) is called the Borel sum. In some cases the answer is affirmative. equivalently for g = l/h.20. (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. . i. 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. It can be seen from these.2. i. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power.
n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. 1 we obtain oo F(z)£(*)" = — .4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. In any case.6)). We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . In fact.4). Q. (20. . it is important to know the large order behaviour of the asymptotic series so that the question of its (exact or only approximate) Borel summability can be decided.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. In this ideal case the integral of Eq. . El . that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). (20. (20. Large Order Behaviour of Perturbation Expansions Eq.e. n=0 (20. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0.474 CHAPTER 20. . . many asymptotic series do not even alternate in sign (as in the case of the series in Eq. Thus if CO F(z) = aiS(z) + J2 anzn. (and E0 = 0). 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. J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . n=0 (20. In general this is not so easy.1)? Asymptotic expansions originate from integrating a series (i.5) then with the integral representation of the gamma function F(z) = (z — 1)!.n _ En+\ n! En+1 ^ (!)"«!. . . (20.
20..2. (20. i.. ? When ?s E ^ 0.e. (20. or can we determine Ar independently? Fig. S... M.3 The potential V(r) gr». § where 7(ff) = To exp  . if the potential is V(r) = +grN. Popov and V.3 and g < 0.11) V. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis). and E represents an eigenenergy. Weinberg [228].g.. . This is the case if the real potential has a hump of some kind. we have a quantum mechanical problem where probability can leak away.9) The question is therefore: How can we determine 9 E(h).10) is considered in detail in this work.2. Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) I jQO Ar = 7T JO tir~l%E{ti)dti. (20.1 Introductory Remarks 475 except for possible subtractions.10) r as shown in Fig.N = l. for < ^ 0.. N = l. 20... 20. The potential (20.. e. if the potential or a boundary condition is complex such that the problem is nonselfadjoint. $s E can be calculated by the WKB procedure./ \p\dr\.
(20. S. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. the BreitWigner formula (10. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. as was (and still is) not the case for t h e equation 1 V . Popov and Weinberg. (20. 20. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available.476 CHAPTER 20.1: Application to the Yukawa Potential Evaluate 9 £ = —7/2 for the Yukawa potential V(r) = g— r / \ r and insert the result into Eq. Weinberg [83].73) in the W K B or semiclassical approximation. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations.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. M. V. (11.11). . Example 20.^ The logarithmic factor in Er is a novel feature of this case. Here n is the principal quantum number of the Coulomb problem (see Eq. Thus in the following we apply these methods to our typical examples. the cosine potential being t h e most completely investigated will also be seen in the following. as we expect on the basis of Eq. However. the eigenvalue expansions of the cosine potential and of anharmonic oscillators.76) (see also the comment on the comparison with the Kepler problem before Eq. Eletsky. L.11) is. (11. of course. 8.9) for Ar. Popov and V. (14.1. Solution: The solution can be found in a paper of Eletsky. Equation (20. An interesting application is provided by Example 20. Large Order Behaviour of Perturbation Expansions where the frequency of particle bounces against the potential barrier is 70 ~ ^classical/27r and w c i assica i = 2ir/Tkepier oc 1/n 3 . 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.113)). the perturbation method of Sec.134)).2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion.
2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.58) and (17.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf. i. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf.3 ) P i _ 1 ( g . (17.13) In the special case j = i the boundary conditions stated after Eq.60).Bq.l ) +2[(q + Aj)2 + l + A]Pl. . (17. (20. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.14) Thus these coefficients can be obtained once the coefficients Pi(q. (20.e. 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.12) (20.Cq. 3)? .j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.26) and (17.M2i+1 for large i.j) of functions ipq+4j.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g . (17.l(q. Eqs. Eqs. obey the recurrence relation jPi(q. The recurrence relation (20. h).J) = (g + 4 j . are given by i . j .j + 1). or equivalently of the quantity A. M2i = 2j2 J[Pi(q. W.55)) X = E(q. (cf. (17.j) are known. B. (17.h) = 2h2 + 2hq+j. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const.13) represents effectively a partial difference equation. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods.20. Miiller [74].j) a n d M2i can be deduced.34)) that the coefficients Pi(q.l ) ( g + 4 j . % = Aq.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.2 A = Mx + ~M2 + 7^WM3 + •• • . Dingle and H. Eqs. J.
91 l)]![i(g3)]!} 2 { [ .l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.l (20.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.15b) The zth term in the expansion of the eigenvalues A of the Schrodinger equation with cosine potential (Mathieu equation) is therefore given by ith term i\ (8h) il 1 2g2 + 3 2i )[{[\{q 7.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign.478 CHAPTER 20. Using this relation and the duplication formula lw('\W±^.18.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i . see also below) .*.•91 {[l(? + i)]![(? + 3)]!}2 (20. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions. . Hence exact Borel summation is not possible. also written lCj. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . (20. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.16a) (20.
Stone and J. . we emphasize a few points *M. J. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions.3 Cosine Potential: Complex Eigenvalues 479 20. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. which means that the potential is there approximated by an infinitely high harmonic oscillator well with levels enumerated by the quantum number qo or (equivalently) n. We retain the previous boundary condition at one minimum. remains unchanged.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/?. W. but with different boundary conditions. MiillerKirsten [200]. the difference q — q$. Reeve [262]. 2 . Previously. when this is g0 = 2n + l . f H . 20.20.* However. At the risk of repeating what we explained earlier. \ 1 \ \ \ \ \ 0. 20. .5 \ \ / \A Fig.2 solutions obtained earlier. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. n = 0 . The potential we consider here then has the shape shown in Fig. .4 The cosine potential shifted by TT/2. but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.4.e.3. . i.3 20. in the calculation of the level splitting as a result of tunneling. this difference was found to be real. l . A means to achieve this was found by Stone and Reeve. .5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. we follow here a formulation in the context of our earlier considerations of the Mathieu equation.
which means replacing z by z±ir/2.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. Here we choose t h e second way. We can distort the potential there or alternatively impose a different behaviour there on the wave function. (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. Setting w = 2h1/2z. as we found these for this potential.480 CHAPTER 20. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. 20. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . but t h a t with complex argument there. i.20a) As stated. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. a boundary condition which at the face looks somewhat abstract. and allows its imaginary part to tend to infinity. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20.21) <u> + ( i « .2h2 cos 2z]ip{z) = 0.4.Ah2z2 • • • }^{z) = 0. 8 the equation can be approximated by d2ip(w) (20. Thus one chooses not the solution with real argument at the minimum. X + 2h2 = 2hq+—. Now we change this situation at the minimum on the right.T K ^ a D_i_{q+1){±iw). To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . 20.e.4 (around z — 0) to the one on the right (around z ~ ir). (20. Previously this situation was the same at the neighbouring minimum. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). 20. but achieves exactly what we want. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. Effectively we consider the tunneling from the left well in Fig. we shift the argument by n/2. .4.
so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. that at z = 7r. .e. 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. by a different method. .20. . 3 .e.4. 20. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. 2. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. we see that ips ~ cos \j2qhz for z ~ 0. . We consider here explicitly only the case of the ground state around z = 0. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. A (which we can loosely dub solutions of WKBtype). the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}.24) Considering t m O i n Eq.> 0 for z > ±ioo. such that this deviation becomes imaginary. Thus.22). Eq. . 17. whereas here we construct these from oscillatortype solutions. Since the large/i2 solutions valid away from the minima were the solutions of types A. 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 = 1.63). 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. 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. A derivation is given subsequently. (20. i. (20.3 Cosine Potential: Complex Eigenvalues 481 As explained above.23) . 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. however. we naturally constructed the even and odd solutions with these (cf.
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.3.27) differ by an angle IT.w) = e T Dv{w) 2TT n») AvViDv^iw). Oberhettinger [181]. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. We see this as follows. Magnus and F. (20. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. and then evaluation at u = 0.29) Equation (20. . Magnus and F.g.30) 'See e. From Tables of Special Functions'. we know from the solutions of types B and C in Sec.g. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. (20.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing. W. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. 17. 2. (20.28) Extracting Du{—w) we have A.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. Thus the only nonvanishing contribution comes from differentiation of Y[—v).•w) + 2ir T{u) <"+!)% Dvi(iw).h(2h1/2 J £2=0 (20. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. W.27) requires the differentiation of this expression with respect to the index v. However. Oberhettinger [181]. tp(v). p.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. The arguments of the functions appearing in Eq.482 CHAPTER 20.24) are linearly independent around z = 0 for q ^ qo.25) qo) E2 0: (20. (20. §See e.(. this can be handled in a few lines. 92. p.
Thus calculations for qo ^ 1 are not given by M.4.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. and "Prom W. . Oberhettinger [181]. . .27) exponentially decreasing contributions.33) = IV2~TT.29). "See e.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 . (20. 5 7 7 . (20. Differentiation of this equation and subsequently setting v — 0 yields. . we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. i.)e" w /4 2 l ^Dx(iw) (20. dropping from the derivative part of Eq. Oberhettinger [181].//n and ij>{y) = C . with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. 92.. 3 . including the derivative of DE2ufl(2h1/2z)). pp.g. W. For u — 0 one obtains the result (20. Reeve [262]. Stone and J. This result is » not easily generalized to <jo > 1. (20. which are valid around z ~ 7r/2. Magnus and F.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.31). Magnus and F.29) v=0 (in + In w)D0(w) + v2ire (m + lnu.31) dT(u) r*(v) dv 1 i/>(v) ^ o l.e.32) which we can use here since all other problems have been resolved: d D {w) dv v (20. p.e. we obtain 4>s D0(2h^z) D0(2h1/2. r(i/) We now return to Eq. i/^0 483 T(v) It follows that d dv [T(u)\ (20. We now go to the WKBtype solutions. those of type A.20.
92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] . with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).— 721 .e. Recalling that we have shifted the potential by 7r/2.34) and (20.sm I We now want to match this solution to that of Eq.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. 2 c2e2hehz UJ2 .484 CHAPTER 20. by there replacing z by z + ir/2.e. We observe that in the matching domain (i. i. Thus in the present case and with q ~ qo = 1 we obtain „2ft. where we can then match these to the solution ips. i.38) in its limiting domain towards 2 = 0.41) 2TT . y COS c\ J (20.34).39) Now. that in approaching z = ir. near z = n the boundary condition (20. (20.36) and consider this at the other end of its domain of validity.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h.35) We construct with constants c\.37) Comparing Eqs. (20.40) From Magnus and Oberhettinger [181] (p.e. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.36) cos  .. Thus the WKBlike solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^Tsinf 2/i hz2 _z_ 2 (20. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.36) V'WKB cie~2hehz' 9h  . Then V'WKB becomes (with cos z ~ — 1 + z' /2. cos z/2 ~ (20. To this end we set z' = z — ix. We now return to the WKBlike solution (20. x . smf (20.40). (20.cos z — 2/icos z ipA^ Tcos  and i n A —.29) and (17. (20. (17. Eqs. (20. we obtain the required solutions from the earlier ones.
(20.39) and (20.38) can be written TAWKB (20. with c\ = exp(—2h) from above. (20.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. ^'mliw <2a47) For qo = 1 we recover the result (20. J Hence Eq. 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. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l .43) Comparing now Eqs.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. (20.e.34) we obtain E2 = i%\ = ±ih22Qe~8h. .2/.e. (20.^.43) we obtain. (20. 1 2 4 1 2 8h ^^2 2 (20.1/V)_le^_ Z With this result Eq. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.45) Comparing this result with tps of Eq.46) Repeating these calculations for the general case (i. (20.44) = DQ{2h / z) ± (2h7r) / e.46). We derive the general result in the next subsection using our own method. i. The general result stated by Stone and Reeve [262] is therefore presumably guessed.
2 CHAPTER 20. Wiedemann [4].20a).e.*. MiillerKirsten and A. •_(:)*<>.e. We have to impose two sets of boundary conditions.z. (20. Achuthan. i. and expanding cos 2z around the point z = — TT/2. On the other side of the barrier at z = —TT/2 we consider different boundary conditions. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. W. 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. i.()#0.5. •+(=)* V(f) 0. Eq.3. 20.5 The cosine potential from — TT/2 to n/2. (20. At the minimum to the right of the barrier shown in Fig.5. 20. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ . at z = TT/2. this equation may be approximated there as d2ip ~dz^ + E + 2h2 .) + 7T V> = o. 20.46) we consider a half period of the cosine potential as illustrated in Fig. H. i.48) These conditions provide the required quantum number qo = 2n + 1.e. V(z) . J.E curre it \ j.Ah2 ( z + . Fig. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq.486 20. Considering the original equation. (20. . the result is formula (334) there.
(20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. For particles or probability to leak through the potential barrier from the trough at z = —ir/2 in the direction of negative values of z where we choose the potential to be zero. (20.50a) 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. For the continuation we use again the WKB formalism.51b) 2i(«1)[i(9l)]'.h.r/2 ~ e i(2hq) V 2 . Proceeding as in our previous cases and substituting the expression (20.h. we have IPA(Z) „2hsinz Aq{z) + O il>A(q. we demand that ^(*)L<. three pairs of solutions of types named A. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish.21)).20. B and C.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .z)=tpA(q.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. as shown earlier.e.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. (20.z). (20.21) into the Mathieu equation we obtain. i. Recapitulating these from Chapter 17.
53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. ^ W H .488 CHAPTER 20.48) and proceeding as in Chapter 17.55) Extending these solutions to the domain around the minimum at z = 7r/2.^ i . Finally the third pair of solutions is Cq[w(z)}+0 Cq[w(z)}+0 Cq[w(z)} where Cq[w(z)} = 22(9+1) •((73) 1 h1/2 1 hW> (20. we obtain 1>±W = \ a a (20.57) where go = 1) 3. we obtain our first transcendental equation. (20. i. 5. (20. Here the upper sign refers to the even states and the lower to the odd states. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.56) Applying the boundary conditions (20.e. and the second around z = 7r/2. We return to this condition after derivation of the . 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.qo .52b) The first solution is valid asymptotically around z = —vr/2.=F2 2\1/2{28h2y/4e~Ah l + O (20. the relation q . \vw).52a) = Cq[w(z)}.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. (20.
58) Ignoring the correction A / 8 . C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. These then provide complex exponentials like (20. .67). formula 280. (20. I^A(Z) overlap parts of the domains of validity of these solutions. i. p. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. as we shall see with the result of Eq. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum. Some nontrivial manipulations are again required to uncover the generation of important factorials.61) [Ah2 cos2 z2hq]1/2dz. Friedman [40].01. Jz+ ' T In principle the integral could be evaluated exactly — see P.49) and we can apply our boundary condition. 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.[ [Ah cos z . Byrd and M. we see that the turning points at z± are given by 2hq .e. Our procedure now is to match these to exponential WKB solutions. D.20.2hq] / dz) We know that the solutions I/JA{Z). r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . F. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier. and we obtain the proportionality by appropriate expansion.Ah2 cos2 z ip = 0.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum.Ah2 cos2 z ~ 0. Thereafter we extend the latter across the turning point down to periodic WKB solutions.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 . Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (20.
490 CHAPTER 20.» _ i y v . using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).64) .62): L: = In ~ and hence L = ln In * f 2(z + f) {q/2hyn + + (* I In + 1/2 1/2 2hj 2 In In 2h 4(1 + sin z) (g/2/i)V2 cos z' 2(z + f) (qphy/^z+Z) !sin2(f + f) ( 9 /2fc) /22sin(§ + f)cos(§ + ^2 ' 1 4 = ln 4tan(f + f (9/2MV2 (20. (20. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20.62) as follows. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(.sin2 z = l . We have for the following expression contained in (20. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. we write cos2 z = l .63) We handle the logarithmic terms in Eq.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.59) 2h COS Z+  ~ 0.cos2 ( z+^) = (z+ .) +0 z+ IT Inserting this into the integral I. (20. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.
we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. (20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .50b). 1/2 (20.+ qj V2 4 A.68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated.20.63) and (20.62).50a) and (20. . on using in the second step Eqs.66) The WKB solution therefore becomes (approximating the denominator of (20.60) by (4h2 cos2 z)1!4). 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.64) into Eq.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. . but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . (20. We ignore this here.3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 . (4^2)i/4 [ i( g _3)] ! V/ l (20. ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  .
50)). 1 2 1 ±2 .492 CHAPTER 20. (14.67) and (20.. S1H (2/ KZ ) / (z f 2 (2^).e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.l^l !T ^^* T2 7 r ( ^^ /. (14.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { . (20.z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). i.49). io(8/*)g/V^i6ft2)V4[±(g3)]! .KW4^8 4 .2 [ ? ) Imposing the boundary condition (20.™) . Large Order Behaviour of Perturbation Expansions again using the Stirling formula. Eqs.i ( 2 / i g ) 1 / 2 ( z + . we obtain l(8/i)g/ 4 e. We now return to the even and odd wave functions defined by Eq. we obtain T _ ( ± ) = 0. i 27r(8^/ 4 2/t /4 ' r(±) .49). where _ 1 + [±) *) + £ (20. Thus with IPA(Z) and TPA(Z) taken from Eqs.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.?/ 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.69) l(8M9/4e2^ \2 [ i ( g .z)}].2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .T 7 77 r )V2 /4(2 .+) + B^) HAh^/\ir. (20.1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .U ^K^.55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+.
(20. we obtain E(q.h)+2h(qq0) (16h)i°+le'Sh E(qQ. 20. Ex = 2h2 + 2hq0 + A 1 (<? .72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20.72). h) of Eq.20.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 .57) and (20.3.e.47). (17.25)) E = A = Ei + E2.h)+i%Ego. Eqs. Combining these equations by replacing (28/i2)<?/4 in Eq.3 ) I = }(»« !7rl/22I(. (20.74) This result agrees with that of Stone and Reeve [262]. (20. In both cases we wrote (cf. i.h) + i l + O '4{[^ol)]!}2 E(q0. (20. in fact.h) = = E(q0.3 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. We now have the two equations which have to be satisfied.75) .21) expanded about q = qo.73) Inserting this expression into A = E(q.qo) El Ah' (20. Equation (20. for q an integer both sides are imaginary.72) We observe that with the square root as on the right hand side both sides are complex.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ . who derived it for qo = 1 and guessed the form for general values of qo.26).72) is our second transcendental equation. (20. the result (20. (20.57) by the left hand side of Eq.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e. Eqs.h) + (qq0)( BE — 90 E{qQ. (20.
77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut.74b). cf. Achuthan. Wiedemann [4] and Example 20. 25Eqo=8h[ 2\1/2 . The same formula can be derived in the case of the double well potential.+  .76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level.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. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. Bogomol'nyi and V. we can recalculate with the help of Eq. A = .e. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. i. MiillerKirsten and A.2 below. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. Eq. W. .494 CHAPTER 20. B. A.80) *P. J. Setting ng AEqo:=2iSfEqo. (17. H. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. 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.W [£(901)]! < (20.I g  ± l ( 2 „.) {I6h)™/2e4h 1 + . (20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. f E . 20. 7 8 ) i=l V .* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. its tunneling properties and the large order behaviour. (20. by others.3. Fateyev [26].9) the coefficients Ai of its perturbation expansion.
19) for the large values of i under consideration here.j) satisfy the recurrence relation (20.83) One now sets ftH = ( 2 % P * ( 9 . f H .81) We observe that this result agrees with the result of Eq.l ) ( < ? + 4 j . i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. J.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.74). 20. (20. Using again n = (go .3 ) P i _ 1 ( c ? .13) out again: jPiiQj) = (g + 4 j . For convenience we write Eq. (20. so that an exact Borel summation is not possible. MullerKirsten [200].. (20. B. We also observe that these terms do not alternate in sign. The integral is recognized to be of the type of the integral representation of the gamma function. a = J + \<1> (2084) . J). there is nothing special about this case. Dingle [74].20.1).j + 1).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. the integration is seen to give (20. *R. (20.j) + {q + 4j + l)(q + 4j + 3)Pi1(q. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. W. and the method can be applied in many other cases. j .l ) +2[(q + 4j)2 + l + A}Pi_1(q.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo .14).13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. But as Dingle remarks.
27h pi(a) a = 4j(4j + 2 g . (20.1/2(z) (a1)! (20. nx pi_i(a + l).86) one can convince oneself that Eq. g < 4 j . formula 13. Pi(a) i. the recurrence relation (20. 0 4j(4j + 2g) pi_i(a .e.85) Considering now the generating function of the perturbation coefficients.ll{z)= L . 4j(4j + 2g + 4) + / . (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .h) 2+ 8h a p(a.z)WaA/2(z) Setting V.1) + 2 P i _i(a) + Pi_i(a + 1).(a) oc . (q + 4j) 2 ~ 4j(4j + 2g). Abramowitz and I.l/2l Wa. e.87) Taking terms of 0(h~l+1) of this equation. which are solutions of the differential equation known as Whittaker's equation. Whittaker functions WKjfl.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1. we obtain (2a . (20.g. z2 (20. (20.4 ) ^ . i.h).K + ^ ] L + K^)WK.83) can be written j . A. 1  1 / 2 (4 (20.85).l. Now.1) + 2 (a — 1) a .^(Z). we regain Eq. d2WKh dz2 + 1 4 K z zM2n WKtll = 0.31.496 CHAPTER 20. (20.89) Replacing here K by a and \x by 1/2. i.90) See M.h) +p(a + l. Taking dominant terms. .h) = ^2pi(a) with p.91) + Wa+lil/2(z) = (1 .85) implies p(a . a. (20.e. (20. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.e. Stegun [1]. p(a.a R .4.
(20.84)) {8h)q/2( Ah for j < 0.(« 1) 2 ]. In the present case this means the replacements q — —q. Magnus and F .j — —j. 89. h —> —h.e.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. (20. > > ^See W.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.86) have to be chosen such that p(a.(« .i/2(z). Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).g(h) are functions of h which in view of Eq. (20.h) ~/(fe) ' (a1)! . h) is expressed in descending powers of z = 8/i. v . i.(« .i/2(z) + Va+lA/2(z) = 497 2+ Va. Dingle discovered the solutions^ p(a. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f .87).93) (a1)! ' where f(h). ^Observe that if one solution of Whittaker's equation (20.94) Hence to insure that p(a.\)V . (20. Oberhettinger [181]. p.88) is known. > (20.Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. such that p(a. g(h)(20.92) Comparing Eq.4 Cosine Potential: Another Method of Calculation we obtain Vai. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. we have to remove a factor (recalling a = j + (7/2 of Eq. a (20.h) = \~\Pi(&) with pi oc h l.20. .. h) is expressed in descending powers of h.92) with Eq.
_ i s.94) here we have a series in descending powers of h. W. Dingle and H. j — —j.l ) ] ! 11 R. . Comparing > > > this result now with Eq.j _ g / 2  1 / 2 ( . g — —q.498 CHAPTER 20. Correspondingly we set for j > 0 / 1 A pla = j + q. (20.96) and a similar expression for j < 0 with /i — —/i. ^ .83) and is given by 1 1 .1/2(8^) ( .95a) Inserting (20. h\ oc (8h)q/2e~4h ( _ . we obtain the result of Dingle: (20. This expression can be obtained from the recurrence relation (20.+ i f or 3 < 0 (20. (20. J. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl.8 / t ) for j > 0.). n = 0. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 .. 2. Muller [74].1..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. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2. (20. i). B.95b) Then for j > 0 we obtain from Eq..94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1.
l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. (20. M. Miiller [74]. A special case of the formula can be found in I.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. Dingle and H.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20.14).e.0) = 1 as desired (although not really enforcible for large values of i). Ryzhik [122]. is M2l = 2j2j[Pi(<l. Thus ° 9 (i + \q) 2«(i+Igl)! p . S.3)]!2 2 '+^. l{Q.97c) We assume that in a leading approximation we can cancel the factors (i + \q). formula (30).+i?_l]!}2 (20. Gradshteyn and I.100) **R. E* J'=I i+M 2 [2i + 9 . 5.{j + \<l). i.157(4).99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.97d) Then the coefficient M<n of the eigenvalue expansion.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20. . of Eq. W.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (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. p.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q. B.e. (20.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q.l ] ! (20.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . J.j) (i + lg)[2i + i ( g .l ] ! {[. (20.20. formula 0. note the misprint there: (r — s) ought to be (r — s)\.
l l ! with the large i behaviour in agreement with the result (20.. n = 0.86). since \{q — l) + \{q — 3) = \q. the result of Bender and Wu [18].1 Anharmonic Oscillators The inverted double well In Chapter 18 we obtained the imaginary part of the eigenenergy E of the Schrodinger equation for the inverted double well potential. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity. .1.e. (18. (18. With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator.101) This formula allows us to cancel all factorials [i. i.e.5. Eq. 20.17)..105 ) the latter because we see from the eigenvalue expansion (18.o i • . i.34) that £ is an expansion in ascending powers of h 6 /c 2 . replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.e. i..102) Thus the result is M m 1 2«Fi + g .. i\ M i large 1 .e. [^ + i ( g .' 1 i • (20.. [19]. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20.500 CHAPTER 20..3 ) ] ! _ 1 Hence M or. Eq.81) for q — qo = 2 n + l .]\. (18.86) and set (cf.5 20. We refer back to Eq.2.4)) £ = ^ = ^+ w and y= ^ > ( 20 .e.i ) ] ! [ ^ +  ( g .Am i . i. i.l.
106) With £= E r=0 Mr (20. .e.(20. (20. i. in terms of K where go = 2ftT + 1. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.l ) r . (18.107) we obtain (i)r r 71" JO yirZ£{y')dy'. the discontinuity across its cut.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. and then the relation (20.20.e. and hence inserting here the imaginary part from Eq.5. (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. The result establishes the series unequivocally as an asymptotic series. 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.108) Proceeding in this way. This can then be compared with the level splitting we obtained for this potential.86).4) of Bender and Wu [19]. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K . We leave this to the following Example. i. in fact as one with Borel summability in view of the factor ( .77) can be verified.5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20.110) This is the result in agreement with the large order formula (4.109) or.
R.^ 20.2: 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. E. J . t p . Cizek. W.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. Achuthan.Q . K. H. Liang 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). *Some of the literature in this field following the work of C. [19] can be traced back from articles in the Proc. J. Harrell. J. J. V. Paldus and H. Cizek. Paldus. + 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. G. Bender and T. S. J. Nakai. J. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. Grecchi. Achuthan. Gram. J. J. 1982) [139]. Its usefulness has also been demonstrated there. Damburg.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. W. " P . from p a t h integrals with instanton methods) and the large order behaviour of the perturbation expansions which establishes the latter clearly as asymptotic expansions. J. Eqs. Solution: For the solution we refer to the literature. Large Order Behaviour of Perturbation Expansions Example 20.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . T. Wiedemann [4]. Wu [18]. Harris. M. of International Workshop on Perturbation Theory at Large Order (Florida. Harrell. . The relation (20. A relation similar t o (20. V. (357) to (371). M. Damburg. Grecchi. Propin. Wiedemann [4].77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. MiillerKirsten and A. Silverstone [52]. S. M. H. MiillerKirsten and A. W. Silverstone [62]. R. J. 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. t The relation (20.502 CHAPTER 20. in t h a t when the quantity on one side is known. E. J. Propin and H. . J. Graffi. MiillerKirsten [164]. it can be used to obtain the other. K. J. S. as in later chapters. R.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. *R.
Probably the most readable introduction. However. unknown). our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. since in such cases the probability of this system choosing a path far away may be considered to be small.Chapter 21 The P a t h Integral Formalism 21. P. Schulman [245]. P. A further wellknown reference is the book of L. S. For the application of path integrals to the hydrogen atom (i. A standard reference is the book of R. can be found in the book of B. Feynman [93]. Hibbs [95]. This formalism is particularly useful for evaluating scattering or transition amplitudes. Felsager [91]. ^ 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. Feynman and A. The path integral is. which explains also difficulties. 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. Chapters 2 and 5. However. *R. Feynman [94]. since this illustrates the most important and most frequent use of path integrals in a wide spectrum of applications ranging from field theory to condensed matter physics.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. Coulomb problem) we refer to existing literature. The formalism is not so useful for boundstate problems (in this case the initial wave function is. 503 .e.* This formalism deals with an ensemble of paths {x(t)} rather than with wave functions and constitutes an alternative to canonical quantization in quantizing a theory. P. however. Kleinert [151]. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. R. in general. 'See in particular H. A brief introduction is also contained in R. We illustrate the method here by rederiving the Rutherford scattering formula with this method.
*/ = U + (n + l)e. (21.tf\ip) . i. (21. (21. (21.tf\x.x).x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.ti)ip(x. x.t).2) where K is a Green's function.e—»0 Sn. *In Dirac's notation we write this equation later (xf. ti) = 5(xf .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.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo. be ip{x. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.1) Let the wave function i/> at time t = ti.ti){x. 2mo dx2 (21. The Feynman Path Integral Formalism 21.x.5) Sn = X / fc=l mo (zfe . =J dx(xf. ti).e. we can compute the action S1. Obviously K(xf. the initial time.tf. U.ti\tp).tf) = / dxK(xf. we wish to know* ip(xf.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo.ti). 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).504 CHAPTER 21.
1 (for convenience we write in the following in the argument of the exponential frequently S instead of S/h): oo poo po> / oo dxTl JS/h dx\ I J—oo dx2 ••• J— c X f~ x n+l Fig.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. the Fresnel integral (13.6a) . The usefulness of the method therefore depends on the fact that some paths are more probable than others. of which one is illustrated in Fig.ti) to its final position at (xf. 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. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation.28)): f dr\ exp ia a) ' (21.t).21. 21. 21.1 A path in (x. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi.tf). as will be shown to be possible below. 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.
dxneiSn/n.m0 .. ie f me\ 2a\a~) 1/2 (21.506 CHAPTER 21.7) 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.s f c _i) + (xk+i . tf > U. 2 + 2a.Xi.tf.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7. for example.t ~Xk+l> + xk+i 2 ) +{xk . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.z f c _i) 2 } } .6b) —c ia o drjr] e x p—V € (21.8) Note t h a t in spite of the n integrations. dx exp z—{a..0 ^7 (x f c + i . (21. J—oo dxneiS/h — • 0.x f c _ i ) 5 \m0 J (21.(x fc+ i ..2 + (x + x f c + 1 .xk_i) dx exp i — { 2 a . We then integrate and finally let /x go to zero. 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.5)): G(xf. One therefore removes the troublesome factor by defining (Sn being the sum in Eq. one has j ^ dr1eifa+i^ /£ = .Zfci) 2 } I +{xk+i x*.xk+i) exp . (21.6c) In the evaluation of these integrals we replace a by a + ifi.37)) dxk exp d(xk . n > 0.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.mQf Xk L ~2e^ ^ .ti) '•= f g m ) (N(€\]n+i dxL.xk+x oo —oo oo .
5).Xi. Different constructions of these families of paths usually lead to different measures.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. (21.1. and so on.21. which all vary from —oo to oo. It is neither a variable of integration in T){x} —> Ylndxn.24).9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. p. the Green's function K of Eq.e.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. However. . A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims.5).26)) that — ignoring the potential — the multiple integral (21. (21. B. 21.g. (21. These integrations are indicated by the horizontal arrows in Fig. G(xf. Felsager [91]. Felsager [91]. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2.ti) for tf > U. 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.Xi. (21. in fact. i.7) for n — 1) to give a time interval. p. 183. B. i. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. and will therefore not be attempted here.tf. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.10) can be looked at as the product of a succession of free particle Green's functions (or propagators).2).tf. we arrive at a path integral without the measure factor N(e) above. (21.Xi.e. nor the limit of integration of some variable Xi.g. We shall see later (with Eq. See the square root in front of the exponential in Eq. as in our discretization in Eq. 11 For discussions see e.tf. it may be noted that with a different philosophy concerning the summation over paths. 183.t In the following we demonstrate that the function constructed in this way is. § See e.U) = / JXi=x(ti) = K(xf.2 The Path Integral and Green's Functions 507 Eq. (21. (21.ti) For tf > ti we now write* G(xf. however prefixed with a new factor. summation over all possible paths.
we have G(xn+2.74)). 21. Then we have on the right the factor G{x+rj).U) imo 2 G(xn+i W)Idr. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.tf + e. the other with that of the action of the free particle SQ.ti) = 1 j^p: f°° cte n +iexp } • fm0. pp.xi. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. we have See. (23. (21. Expanding this in a Taylor series around x. one with the full action S. 187 . so that the piecewise linear paths fill the entire space of paths.Then we can construct a recurrence relation as follows.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.tf.ti). Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals. Each of these linear pieces is completely characterized by its endpoints. . Felsager [91].188.3 in the evaluation of I a specific path integral (see Eqs. the discussion in B.tf + e. Consider a time t = tf + e for position Xf = xn+2. I We employ this method later in Example 23. Then the measure drops out.ti. and one is calculating the Feynman amplitude or kernel relative to that of the free theory.xi.Xi. however. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths.exp 2e n ieV(xn+2) G(xn+2 + V.Xi.508 CHAPTER 21. We have G(xn+2. It is for these reasons that — unless required — the measure is frequently ignored. we obtain the same as before except that there is no factor N(e).72) to (23.1 and allowing n to go to infinity. Thus consider an arbitrary continuous path connecting the endpoints and use the time divison as above but now connect the intermediate points with straight lines.
U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.xi. tf.6a).Xi.Xi. U) = e^G + 0(e 2 ) dtf 1 d2G(x.tf''Xi.ti) N(e) or G(x.t Thus + 0(e2 (21.— p G ( j .tf)= dxG(xf.G(x.xi.x. ti) satisfies the timedependent Schrodinger equation.tf + e.Xi.ti)ip(x. tf + e.13) We integrate with the help of the integrals (21.U + 0(e ). Retaining only terms up to those linear in e. tf.ieV(x) + ri< —G(x.tf.tf + e.tf.21. U) + .ti) = j^rr dnexp I ^rj1 .xi.14) = G(x.tf.tf.U) +••• \ 509 G(x. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.15) We see that G(x. tf.xi.tf.ti) = 1 d2 + V(x) G(x.xi.ti) for tf > ti (21.xi.Xi.Xi. .tf.Xi.xi.ti) .Xi. It follows that ip(xf.16) .tf.tf.Xi.ti) = G(x.tf. U) .ti) +\if(^G{x.Xi.(21.tf + e.ti) 2 —l€ — ^2 + V(x)G(x.6c).2 The Path Integral and Green's Functions so that G(x. 2mo dx2 (21. d and this means d i—G(x.ie 1 d2 2nV0dx^G{X.6b) and (21.tf. (21. Xi. tf.ti)\ .ti) +V(x)G(x.U).tf.xi. we obtain: G(x.
U). as may be verified by inserting this into Eq.U) = G(xf.x.x.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.tf. (21. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. denoted by KQ. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.x.16). Hence our original function K is the same as G.U) = / T){x} eiS\ [fdt \mx2.U)G0.10) is indeed to be identified with the timedependent Green's function.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. (21. now called So.15) is a wave function (cf. The function G which satisfies the timedependent Schrodinger equation (21. We have therefore established that the Feynman integral (21. The result.U)6(tf . (21.8)) Go(xf. is the Fourier transform of the nonrelativistic propagator.tf.3. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function. (21.ti) = = 9{tfti)—G + G5(tfti) .U. Eq. or K(xf.510 CHAPTER 21. 21. and the Green's function KQ\ S0= Here (cf. (21.ti) = 5(xj — x).x. Eq.Xi.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf. K0 = 9(tf . In a problem where V is small it is convenient to calculate K with a perturbation method. as this is also called.17) Differentiating K with respect to tf we obtain —K(xf.tf. 21.1)).20) .tf.
Xi. In Example 21.Xi.ti)9(t).23) We now insert our expression for In into Eq. 2t Go(xf. exp where Xj = xo and Xf = xn+\ and there k = 1).Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.x 0 )" n + 1 (in agreement with Eq.ti) = = G0(x.t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.24) ( s / . f c . (7.22) • \ n/2 .Xi. n + i .3 The Green's Function for Potential and hence Go{xf.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.Xi.tf.ti) Ko(xf.8). .t) K0(x. (21.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x..Xi. we have Go(xf.1 we show t h a t oo /.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.ti)= iX (o. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a. We obtained the same expression previously — see Eq.21./ . (21.tf.z / c .. Then Go(xf.tf.37) — by explicit solution of the differential equation which the Green's function satisfies.25) T h e quantity Ko(x. (21.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21. But t = tf — ti = (n + l)e.tf.tf. (21.tf.21) fe=l and t = (n + l)e = tf — U and h = 1. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.23) for Go and obtain Go(xf. exp 2 (21.Xi.i ) 2 .t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.
xo)2 + (x2 .5) is carried along. (7. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 . (21.ti.22). or else by induction. For n = 1 (see below) h This result can be verified like Eq. Solution: The result can be obtained by cumbersome integration.22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:.x „ + i ) 2 = (x„ + 2 . (21.tn+l•xn.7) by direct integration with ^ ..x „ + i ) 2 = ( x n + 2 Xi + Xi .25) we can rewrite the Feynman path integral or kernel (21.an) Then n I 1/2 (n + 1)A" "I 1/2 fX r ioo ^n + dxn+1eiX{^+v2y2Hx"+2x"+1 l ) 2 } J —OO (n + l)A n «ix„+1ea(Sl»22»lI«^I.)+(1"+2. In our proof of induction we assume the result is correct for n.x ^ 2 + y2 . (21. i. the latter being a Green's function with interaction like that for the harmonic oscillator. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .2y(xn+2 .tn)••• K0(xi.f_^y e.e.1: Evaluation of a path integral Verify Eq.54).xi)2 = 21 xi ^2o(„. If the potential V(x) of Eq._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.26) with n integrations and n + l factors KQ.6a). Then oo 2 / (21.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i..512 CHAPTER 21. as / JXi f r D{x{t)}eiS°lh . Eq. we have instead of 5*0 and KQ expressions S and K.t0) (21. ^ (xi .)!} J — OO . (21. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x0.10) in a product form. Example 21. The Feynman Path Integral Formalism We can now see the group property of the Green's function or propagator by observing that with the result (21.
Q . This quantity is also described as the nonrelativistic free particle propagator. (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.6a).28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.2 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.3.2y(xn+2 . Using Eq. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .x2rriQ 1/2 toij_. it . (n+l)A n n j(a.i) H —3/ . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i .3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). We pass over to momentum space with the help of the following two integrals.i j ) H n +— z 1 .i) . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. e > 0 .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m. (21.25): Ko{xJ) = = (S) e{t)(^X'2 y J . i. The first.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.n+2a. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. ) H — (xn+2Xi) n+2 n+l n+2 y2 .2 y ( a .29) Hence with KQ from Eq.Xi) + ( x n + 2 Xi)2..27) we have (with a = t/2m0) . n + 2 .a . 21. The second integral is an integral representation of the step function. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A . and + 2 2 l n + 2 ( x n + 2 . a dp exp [ipx + iap2] = exp — i(21.2. (21. (21.e.21. A(n + 2) 21.
51)) is fairly selfevident. an expansion in rising powers of the coupling constant (contained in the potential).Q (21. The Feynman Path Integral Formalism Im t \t>0 Rex Fig.2 Contours of integration. The generalization to three space dimensions (required below in Eq.*' where 2mo — r. ( 3 b) ° The expression E+ V — it 2m.e. (21. i. This is the type of expansion most people describe as perturbation theory. Replacing 9(t) by its integral representation (21.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. Thus in first . 21. The first order correction contains one factor of the potential V.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. the second order two factors of V and so on. 21.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£.28) this becomes K0(x.514 CHAPTER 21.
dxi ( m0 V V27rie/ m0 \^He \ exp fe=.^){iV(a.. (21.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.5) a n d (21.^.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.(xfc .. } e °° f l f •/ oo /"oo F ( a .i. \2meJ n+l J.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.^ J_00 V(xk. (21.t) JSo dtV(x.21. x (n+l)/2 \ lim .Xi. dxn exp i ^ —. Eqs.i.t)\. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2. (21.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.xi.ti. .4 Including V to First Order: Perturbation Expansion h=l..tf.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2._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 . .20)) and with elS = exp i S i o / hi dtV(x.i V e / D { a .)} (21.xfc_ OO J— OO J—OO L J. li f f 515 order (cf. .. .32) Since — we recall this for convenience G(xf.Xi.t/.U).33) xG0(xi.t) the first order correction is (G\ ~ G — Go) Gi(xf. dxr.tf.t)] n+l .
They must therefore be deleted.x2.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).U) xV(xi. dx2Ko(xf.ti). t') + • • • .t2) xKo(x2.t)V(x. t) Ju Thus the term containing n times V contains a factor 1/n!.t2)V(x2.x2. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.t)K0(x. . i.516 CHAPTER 21. (21.t) (21.tf.ti)K0(x. This is cancelled by n! different time orderings. (21.ti).tf. i.I dtV(x. Jti Consider the term of second order. Xi.xi.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. = i dt2 / dx2Ki(xf.e. OO /"OO / oo dt / J—oo dxKo(xf.tf.t) ! Jti r*f dt'V(x.x.36) {if •j.xi.t2. / 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.ti) t2)K0(x2. The factors of "V" originate from the series expansion of the exponential exp ftf . (21.i / dtV(x. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti. t2.t) =li dtV(x. in writing down the expansion exp i dtV(x.e.xi. The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.t.
x. (21. These may be formulated in configuration space as below.38) .tf.39) dxK(xf.Xi. . called Feynman rules. Here. 21.39).ti). x"^ X x"2. of course. E) which is given by (in onetime plus onespace dimensions) V(x. For various reasons — such as illustration.ti) = K0(xf.tf.tf.3 Feynman rules and representation of the Green's function.Xi.t)= f dpdEeipxiEtV(p.21.t2 K o (x 2'l2 .t) ) K K K0 'V X K o Fig.x.t. we are concerned with quantum mechanics.ti). U) (21.3 along with the diagramatic representation of Eq. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.ti) i dt = K0(xf.i ] P I dt J dxKn(xf.Xi. The Fourier transform of V(x.Xi.Xi.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.t. (21.tf. Such diagrams in the context of field theory are called Feynman diagrams.tf. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. x r l i 'V(x.t)K0(x.t)V(x.t) is V(p. 21. or in momentum space as mostly in field theory. tf.t)Ko(x.ti) (21.40) The rules for the present case are given in Fig. The diagrams represent mathematical quantities and hence are designed on the basis of rules.t)V(x. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.xi.E).
= l. The Feynman Path Integral Formalism 21.ti).ipf have to be properly normalized.7 = e i k ' x _ i : l i with f i/}*i/jdx.e. to one particle in all of space. for the allowed values of k: kx = 2imx/L.e.518 CHAPTER 21.tf. * 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.ti)il)i(xi. i. i.Xi. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.k = 2wn/L. (21. and this means the transition of a nonrelativistic particle from an initial state 'i' to a final state ' / ' with wave functions ipi (or ipin) and ipf (or if>out). in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. (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. T = duration of time for which potential is switched on. mo V since ip is normalized to one particle in volume V. (3) The transition probability per second is = \A\2 . We recall for convenience from **Note that from the wave function we obtain.41) We have to pass over to three space dimensions. The plane wave function normalized to one over a box of volume V = L3 is. here with E — k 2 /2mo. There the wave functions ipi. ^(x. with p = Kk. and dn = dnxdnydnz. . The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions.t) = . (21.tj)K{xf.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. dn = (L/27r) 3 dk.
# = ^an(i)VnM) n = a^(t) + \a^(t) + .T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.21. H = H0 + \H'. Then H'ki can be taken out of the integrand in Eq.5 Rederivation of the Rutherford Formula 519 Chapter 10 the Schrodinger equation with the subdivision of its Hamiltonian into an unperturbed part and a perturbation part.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. u = ^\H'kfp(k). and with these the following perturbation ansatz: ih^* with an(t) = HV..t) = e~^lT\ r where r = Ixl.117) and Example 10.e.4).44) = u(Ek.— . (21. 47T 1 3 7 (21. Eq. The transition probability per unit time is .= 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. V(x. (21..k. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. (10. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. Proceeding in this manner one obtains Fermi's "golden rule". P(*) = j r (2145) (cf.47) . i. in which T is a large but finite time.44) and f_oo can be replaced by fQ. «= — .
tf..Xi.52) . T a large but finite time.tf. In our previous onedimensional considerations we had with Eq. (21.e.ti).48) where from Eq.44) (here in the scattering problem with initial and final energies equal). x. (21. £.i x (i)V(x.tf).t)V(x.t) = ^ J*ie*<*'*><<*'*>. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.51) Inserting this into Eq.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P .ti).>/>.tfx.yLi.tf)K0(xf. Thus with the ansatz Ko(x. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.520 CHAPTER 21. (x*. (21.39) OO /"00 / dt / OO J —OO dxK0(xf.x. dt dxip*f(xf. t. by threedimensional dxf and so on.tf)ip*(xf.f.^ / . (21.).t. t.xuti) = ^ J°° d p e i p x _ i ^ * .t) (21.x. i. Xj.ti)'ipi(xi. The lowest order approximation of A. We can relate this amplitude to the amplitude ark of Eq.t)K0(x.30a) K0(xf.50) Here x — xj — Xi and t = tf — ti. we obtain A1 = clS(ti tf) / dxf(i)V(xf. ti) = co5(t — U)5(x — Xj) etc.tf.49) i. (21. Ai = dxf / dx. (21. This now has the form of the amplitude (21..f. (21. also called Born approximation.44) by contracting the propagators to instantaneous point interactions. t)K0(x. (21. so that 0(t) can be dropped) K0(xf.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.* / . we have to replace the onedimensional quantities dxf etc.tf.tf)Ki(x.tf)ipi(xf. The threedimensional generalization is (taking tf » T and ti <C —T.
).57) we have A(p.T) = ^JdXJdt%*t2/T2exp i ( k . which is evaluated in Example 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 . k2 ipx + i—t + ikx — i 1 2mo 2mo (21. (21.~ .p ) .T) = x exp L z/ q/2 k2 521 f dx. 2 „ x v~ m o (21.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p.59) .fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .iaT2/8: L With 0 0 • .i ( k 2 . P2 . Then integrating over q and q' we obtain . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .21.k.r) = = y .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.T) = JdtJdx(^\V(x.k.56) (21. (21. (21.t)exp A(p. .54) Now inserting the expression for the potential V(x. £). Then we require the following integral.k. this becomes A(p.p 2.53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.k.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 .58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.
2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . otot contains the factor \A(p. ( 21 . k2 .e.„.63) we have: Probability per unit time = 77777. mo m (21 . <rtot. In our case this has not been done.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .\ 2?7lo / 1 (21. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).< ^ £ \M VTTT2.T2f 8 V/ 2 . / / P 2 _ . (21.P ) 2 + M2]2 4 VvrT2.60) Now.T)2/T.63) We establish this result for T — 1/e in Example 21. Using the result (21. e (21.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. y 2 [ ( k . e 8{k) = lim 7=*—^ (21. 66) o .„ 2.65) But E=*. . i.59). integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T. write down the probability per unit time which is ( 8 \1'2 _a2 16^ . T ) ~ . k . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21. dE = PjV = ^ E ± .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.522 CHAPTER 21.3.^ . { I i _ F T l } .a ( — ) ^ . we obtain .51 — 2 /w2 y ~ —. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . This expression assumes that the integration over time t is normalized to 1. (21." + M ] p) ' "•"• ^2 .64) Inserting the result now into our expression for the total cross section. Using Eq.k.62) We want to consider the limit T — oo.
setting M2 = 0. and inserting the expression thus obtained into the final result of Example 10. (21. 2mo / dft.59) by replacing there q 2 by ( k — p ) 2 .67) and (21. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.cos 9) = 4k 2 sin 2  . Then (x = r) foo „2jj.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 — . (21. This result may also be obtained from lyukawa of Eq.^ . p 2 = k 2 .69) This is the wellknown Rutherford formula.68) Inserting (21.e . giving the potential a coupling constant. /Yukawa = ) Solution: We choose the 2axis along the vector q..21.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) . i.j .70) .M £ M * = / m0kdn = m0^2m0E (k . 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 .67) T h e delta function here implies energy conservation. —Mr 4w = .65) with the cutoff M > 0.^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.= e " f c a ^ o 2v^fa /4a = lim £^o —.4.^ .p ) 2 = 2 k 2 ( l . esfn (21. so t h a t (21.* 9 ' d r e . 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.e. Example 21.j .68) into Eq.e. (21.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .
' i a = lim e —0 pk 2 /e2 e^TT 21. We write the state vector of a particle of threemomentum p in the momentum space representation: p). a simpler method suffices.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . a. We consider again nonrelativistic quantum mechanics. Va we obtain g(k) = Je~k V a = 0. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx.* .^0 a^C 2^/TTO. We have x> = I J ^ l p X p l x ) .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. Direct integration /4a .72) co From (21. however.q).70) and (21.x) (21. (21.75) . (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 . 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 ) . Since dxe" 1 1 1 = « / .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. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. '<*> = .71) with respect to k yields the same as °° 2 ..oo J / )eikx.524 The function required is CHAPTER 21. (yx) = 6(y . (21. co oo dx co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). Differentiation of (21. In configuration space the state of the particle is written x). In the present case.
e.e.t) = +iH\x.t> =iH\x.t). Hence for the infinitesimal transition (x'.£) = ( x V ^ ' . Then ( X  ^ . at initial time ti to its final position xy at time tf. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H.21. t) = (x. H. 4) is to satisfy the Schrodinger equation. Ryder [241].t) = e* H t x)..If '/'(x. (21.82) (21.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). i.t).e.t'x.82): The general Schrodinger equation has the state vector to the right..t)} constitute a "moving reference frame" in the sense that ip{x.i'x.We now introduce the Dirac bra and ket formalism. x.81) Comments on signs in Eq. The time development is given by the Hamilton operator H. (with ft = 1) d/dt\ip) = —iH\tp). (21. The equation 1 = J ^p)(p. (21.0) = e""x).77) (21.160. pp. d/dt\x. i.We define x. Thus (qPp) = p(qp) = p(27r)3<5(q . (21. so thattt d — x. so that \if>t=o)s = H>)H.i) = e iH4 x.t) = —iHip(x. t' — t.t).i) = K(x'.t\.p).79) Feynman's principle is expressed in terms of the propagator or transition function (x'.^ l x ) . which describes the propagation of the particle from its initial position x. as had to be shown.76) These relations imply the completeness relations 1 = J dxx><x. See also L. i.80) (21. i.x. (px) = e~ipx.e. i. t\ipo)s.t'. .6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . we must have d/dt{x.t). M^ip(x.83) (21. ) H = (x</>o)s. Similarly we have a position operator X with Xx) = xx).t\ = iH{x.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s. x' — x infinitesimal. (21.e. Then the states {x. 159 . We consider first an infinitesimal transition.
P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«. Pq) = J 7^3Pp)(pq) = J dppp>*(p .89) Using this and previous relations we obtain for the Hamiltonian (21.V 2 < 5 ( x . "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).e.x).x') + V{x)5(x' . (21.i p '.y<&(p'ix')(x'ivix)<Xp) = f dxe.90) (P'IVIP) = y<&.p').x y(x)e i p . (21. The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P '  e ""'"'' )  P ) < P  X > e.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).85) the amplitude or expectation value or position space representation (x'#x) = ..88) Thus P has the operator representation P= /dxx)V(x. (2187) (21.526 CHAPTER 21.x = V(p .91) . 2mo Since (21. i.* .
the signs are reversed when these operators are applied to the ket vector \x.tj+i\x. tf subdivided into infinitesimal transitions (xj+i. 527 (21. ti to x j .p') • • • j W < 5 ( p ' .p'). We have <x'.92). i. But e'ipx + 0[{t .e.t') \^(27rf5(p 2mo +V(pp' where we used (21.t) (cf.i'x. t). (21.j. (x'.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.21.84). Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.fx.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 '. i. EQ — Hi .p) + i{t .92) We now return to the expression (21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx .tj). (21.82)). We now consider the transition of the particle from Xj.95) when applied to the bra vector (x. for the amplitude of an infinitesimal transition.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .p') + V(p .
.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi . (Xf.ti. i.x 0 ) + p 2 • (x 2 . t n _ i ) t„2><xi.. d x n .xic) by fc — Xfej H(pk.528 CHAPTER 21.tf\Xi. . d x „ _ i . .tj) = / ••• / d x „ _ i d x n _ 2 . x f c ) = ^ . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/. dx. = 1 X X<jPfc We now replace H(pfc. ..i/xj.t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21.tfei) fe=i .e.94) and thus obtain a phase space expression.xn)(tn in_i)]. . x n which the particle passes at times tQ.xk (21. d x i ..99) .ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . x i .x i ) H h pn • ( x n .27) in which we make the replacements x —> eifc. x i ) ( t i ..X n _i)] h) (21. t n  x n _ i ..97) H hH(pn.2. We now insert for each of the infinitesimal intermediate transitions the amplitude (21.i { f l " ( p i . (21.a — —e/2mo and divide b o t h > sides by 2TT. zmo T h e n t h e integration over all p ^ can be performed by using Eq.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) . .tf\Xi.. i. . then corresponds to the sum over all paths.98) tk — *fcl # ( p f c .t 0 ) + H(p2.x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.ti) rx/=x(i/) d x i dx2 .tn respectively. T h e integration over the n—1 intermediate positions.+ V(x f c ). .. d x i ( x n . x exp [ .tixo.e..
fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .e. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.100) over V{p}= L J ^ rwooll V (27T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O. i. > Thus the phase space representation of the pathintegral* (21. Truman [254].98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.. (21.8) as N(e) so t h a t the integral does not vanish for e — 0. (21. i.p)]\. (21.e. Smolyanov.103) is another form of the pathintegral for which the integration with the help of Eq. G. .] ex H ^I (.98).6 Path Integrals in Dirac's Notation 529 Forming products of expressions of this type we see t h a t this relation can be generalized to the threedimensional case. G.tf\xi. to / m0 \ 3 / 2 las. / {xf.101) Then Eq.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.21.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. (21. A.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. Tokarev and A.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21.H(x.
We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.104) where for the particularly simple Lagrangian L under consideration here m0±l . £ /  X J .H(xfe. xfc).ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. this agrees with the corresponding discrepancy in the onedimensional case of Eq. Thus finally we have Feynman's principle: (Xf.e.105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. Consider the following expectation value with primed and doubledprimed states denoting some intermediate states between initial and final states denoted by indices % and / : (xV'l x(<) \x'. . Garrod [106].530 CHAPTER 21.ti) where nl D{x(£)}exp '. We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. pk) = L(xk. i.V(xk) = m0±l .t'}. operator (21. We have now obtained the path integral representation of the matrix element ( X / . t to x.106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). I hi dtL(x.tf\xi. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. t. (21. (21. pfc) = Pfc • xfc . p . £ J ) with no operator in between. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation.tf\xi.8).iJ(x fc . nl (xf. x.8). (21. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq.x) .x) (21.
.77) and thus have 531 *!*') = *M = * 7 ^ I P > « ./ ^ I < « . .5 l^ t 'lp / > e .* * .76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). Thus <xV'xx'. (21.107). ••• 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 (21.21.i p '• x . .108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x". (px). With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'. With the wave functions (21. x fdxG  ^ P <p"x3)<x3e^V) x (21.0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'.t"\x\x'. so that we obtain a cnumber position coordinate which we can shifted across other cnumbers.t"\x5){x5\x'.109) .v ' x ' and thus {x".t'}. (21.* * f dp = j (2^ p) ^ d .6 Path Integrals in Dirac's Notation For the verification we use Eq.
i ( t " . = x"6{x" .532 CHAPTER 21.£ n xx n _i.i(t" . (21.i t " ( x "  #  x 5 ) ( x V ) +.85) (xVxx'.£/xxj. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • .t') (xV'xx'. t0). x / / e .t') dp .• dx1 (x n .x') .£'> = x"[«5(x" .113) Inserting for each of the infinitesimal intermediate transitions the amplitude .t'){x"\H\x') = X + •••} (21. ] .tf — tn and x$ = XQ.i{t" . (21.x') .*/<x"x5)(x5tfx'> + .tj+i\x\xj.tj) at equal time intervals e. .t n _i) x ( x n _ i . Inserting here the integral representation (21. We now consider the transition of the expectation value of operator x from Xj.£j to Xf.112) where x" is a cnumber. we obtain with (21.x ' ) (2^ i^+vM + D «p(x"x') (21.f ) i f e i p .0 dp x„ / (27T): 1 . (21.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .. ( x " .110) Recalling Eq. . We now proceed from there along parallel lines. This is the expression corresponding to that of Eq. £ n _ixx n _ 2 .iit!' 1') 2mo + vtf) + <5(x"x'). Thus with x^ = xn.tf subdivided into n transitions (xj+i.ti = to as before.111) 1 . (21. tn2) • • • (xi. we obtain (here and in the following the limit n —> oo at the end being understood) (x/. tixx 0 .94) for the case with no operator in between.£j) = / ••• / dx n _idx n _2 .90) we can rewrite this expression as (x".74) of the delta function.t"\x\x'.
y (27F (27F •" (2^(2^)3 exp[.7 Canonical Quantization from Path Integrals (21.{x". .and hence may be put in front of the integrals to yield (xf.x 0 ) + p 2 • (x 2 .. we obtain* (xf. n—>oo *•»• 21. x n _ i x n reduces to the single factor x^.{^( P i.112).t') . (21..xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta. (21. tj  xj1 x». the string of factors xi X2 . Xn _i X„ xi dx.x n _ i ) ] .tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{. x (21..« • > We reconsider therefore the transition amplitude (x". .( x „ .t"\x'.105).3 " n d x f c .t') sider the variation S(x".tf\x. We demonstrate first that the operator representation of P given by Eq. (21.t') (21. U).2 . tj \ xt.t') = and con 5x"Vx„{x". tj) = xf (x/.x 1 )(t 1 Z & ) +H(p2.115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] . . For simplicity we consider the case of one spatial dimension.x2)(t2 ~ h) + • • • + H(p ••• + p „ .89) is consistent with the conjugate expression dL (2L116 * . x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .t"\x'. dxnn _i Xi X2 ...t"\x'.\Xi.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 (21.10) or (21.t"\x'..114).21.2 .t') = (x" + 5x".t"\x'.x .iyxxj..114) x exp [i{pi • (xi .ti) 533 rxf=x(tf) = / • • • / dXidx. dyi l xi x 2 .117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq..
120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t').t') =  6— / 1 V°{x}t . jt/ 37 ( £T. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. x) Jt Jt dL. with N = N(t" . The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. For the purpose of enabling some formulations below. Hence w « = (t).119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq.t').118) — = (21 5x"(x".t"\x'. dx dt \ dx _ (21. (21. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. " (t).tt).x). V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x.122a) = 17) 5x"Vx/.118) JI(x)/h .t'): 1 fx" 6(x". Then. t".t"\x'. (21.(x".534 CHAPTER 21. .l&cdi.10) and set D{x} = D°{x}/JV(i/ . and we demand the validity of the equation of motion.i/(x)/fi ) j ^ ) dtL(x. <9L' dt —dx + — 5 x ox ox f . we separate the metric factors from T){x} in Eq. Thus in this case of classical mechanics dL_ d fdV. dL r /•*" . (21.t i_(dL_ x't' h\dx J t„ (21.
1>) = ^{x". / Jxi V°{x} / Jx' V°{x}.125c) Corresponding to our relation (21. § The evolution of the system is expressed by the Feynman functional integral (xf. ^2\x'.1 (A / ).U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x.125b) and the measure property (3) N~\A + A') = A^.t'\ = l.t'). S.t')(x'. F. 6 (21. x' (21.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization. I(x) = / JA dtL(x. Streater and A.x). 535 (21. (21. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. .7 Canonical Quantization from Path Integrals or JL(x»y\x.122b) By comparison with Eq. Wightman [264].115) in three space dimensions.1 (A)iV..21.126) R.124) where with A = tf — ti. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.125a) (2) the completeness relation holds.Xf = x(tf).tf\xi. (21.t"\p(t")\x'. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). (21.
(21.t') \ i ox" i ox"J or in commutator form [P.127)) (x". tn lie correspondingly in consecutive order between U and tf.t"\x'.ti) f ' V0{x}xlX2eu^lh.tf\xi.t')=x"(x".t"\x'. (21.128) where tf is later than ti and t\.t') (21.ti) = x' 'Y^{xf.tf\T(Xl • • • xn)\xi.t\x'. . Thus (xf.114) (xf. (21.t') + x"^J^{x".131) = ih(x"..127) We can now write down an expression for a timeordered product of operators. Then the relation follows with the help of the properties (1).x](x.t'). (21.536 CHAPTER 21. (2) and (3) above.ti) = —!— p V0{x}xx • • • xneu^h.t>). We have (cf.t'){x\t'\x2\xi.t').130) ~i5x7jX"{x".133) = ih{x".tf\x(ti)\xi. We consider the case n = 2. Thus if T denotes such ordering. We can rewrite this expression as the commutator relation (ITU*" x"^](x". Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21..t"\X'. we have the relation (xf.132) The generalization to higher spatial dimensions proceeds along parallel lines.ti). (21.t"\x'. (21.tf\xi\x'..t"\x'.t') = ih(x.t"\x'.t').129) = ^ y With Eq..tf\T(xix2)\xi. Eq.t\x'.ti) = Xi(xf. (21.t"\x"\x'.122b) we can deduce the nontrivial canonical commutation relation arising here. The Feynman Path Integral Formalism The time t is in general a time between ti and tf. (21. But for t = ti we have (and equivalently for t = tf) as we can see from Eq.
as we have seen in Chapter 8. i. Consider e. i. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e.e. such as the fine structure constant of quantum electrodynamics. Besides. z = a + i/3 with a — oo or f3 —> oo or both. i. W = e x p ( l / z ) .g. in the context of quantum mechanics are not convergent. although the first few terms indicate a convergent behaviour. see e. N.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. Expansions of this type arise. nonzero radius of convergence in the domain of the coupling parameter. In physical applications such cases are rare. > 537 . so that successive contributions to the first approximation do not invalidate this approximation. 102.e. E.g. Watson [283]. but asymptotic.e. if a function is expanded in the neighbourhood of an essential singularity* or if an expansion of the integrand of an integral representation is used (i. If the expansion is mathematically well defined. p.Chapter 22 Classical Field Configurations 22. T. Whittaker and G. Such an expansion parameter is usually known to be or assumed to be small in some sense. the series converges. An essential singularity is really an ambiguity rather than an infinity. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. It was therefore natural to search for alternative methods of expansion. it has some finite. divergent. in the sense of rigorous convergence tests.e.g. for sufficiently large r. Such expansions are strictly speaking.g. The vast majority of perturbation expansions discussed e. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. integrated over) beyond its radius of convergence.
e. Many of these aspects are interesting by themselves .h2.g. 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.h°. although this terminology is not precise. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. i. cnumber.) l/h2. and in general this change is accompanied by certain constraints. There. i. On the contrary.e. The study of expansions of this type has turned out to be extremely fruitful..h3.e.h. The classical. 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. and the corresponding expansion is called the "loop expansion". Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. the consideration of conditions which insure the existence of Green's functions of this new type of expansion procedure (which again leads to asymptotic expansions) and so on. The fundamental extension of the method of canonical quantization (i.. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0). In searching for other means of expansion. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". A further challenging task was the development of methods of quantization of theories which incorporate classical. the latter's topological properties if any. This type of expansion is decsribed as "semiclassicar. One therefore faces the problem of quantizing a system which is subject to constraints. and has led to insights which previously seemed unimaginable. However. such > a series begins with a contribution proportional to (e.. 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.538 CHAPTER 22. In the following we consider various typical examples. These methods are often described as methods of collective coordinates.. such as soliton theory. cnumber first approximations. higher order corrections are of orders ±. nature of the dominant approximation does not imply that this describes classical motion. A method which achieves this is in particular Feynman's path integral procedure. The procedure requires a change of variables from the original ones to collective and fluctuation variables (in analogy to centre of mass and relative coordinates).
A 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). or even a solution to a modified field equation (e.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. may like to consult the article of D. 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>.g.2 The Constant Classical Field 539 and in any case deserve detailed study. modified by going to imaginary time). Tong [272]. a static (time independent but spacedependent) solution of the classical equations of motion. vortices and kinks. Thus references to field theory will be exceptional and hopefully do not irritate the reader. . t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. V{</>) = m§0V + A0(<^)2. d^] = d^d^ . 22. This means theories with field densities which are therefore infinitedimensional.V(4>). it is important for a better understanding of the considerations which follow. 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. We shall not be concerned so much with the case of a constant first approximation. monopoles.22.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking".1) ' T h e reader who wants a highly advanced overview of instantons. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. (22. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). to keep this case in mind. However. 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.
m l + \04>*4> )$ = <).£). of the equation of motion? For these we must have all derivatives of <j) zero. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0.S£/<90(x. V ) .3) See e. 22. 22. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. V[<t>] m 2>o o Fig. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. 433. § (22. p. i. cnumber) solutions <j)(x) = 4>c = const. and hence here [d^ + m20}(l) = Xo(f (22.1. . 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).e.e.1 Different potentials for different signs of m§. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. £)) is bounded from below we must have Ao > 0. B. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0.540 CHAPTER 22.2) We ask: Are there classical (i. The EulerLagrange equation is d„ dC d(d. Classical Field Configurations where d^ — (—d/d(ct). Felsager [91].g. as illustrated in Fig.
2 The spontaneously chosen phase. the only such solution is the case we wish to consider.22. H = —C = V. 22. w A iP 3ml 5* "XT" (22. (22. The Hamiltonian density H is defined by the Legendre transform H[<f>. mg < 0. not the solution x(t)). We can convince ourselves that <fic is the field associated with a state of minimized energy. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution.2 The Constant Classical Field For Ao > 0. It is important to observe that <bc does not possess the rotational symmetry of the Lagrangian density C or of the EulerLagrange equation in the plane of complex fields <f>.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.ir]=7r<fiC.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. we have 541 0. Fig. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected.</>2)plane. But for Ao > 0. 22.e. TT = —J.5) . and so ft[<M=m6:(^* 1 + Ao(0*0)2.e. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. i. aq> For (b = const. TTIQ > 0. with 0 = 0. like the solution x(t) of the simple Newton equation mx(i) — —V'(x) violates the invariance of this equation under time translations t —> t + 5t (i. the equation has this invariance.
the one for (3 = 0. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. (22. and partly by climbing up the parabolic wall on either side. and we wish to investigate the behaviour of the field cf> in its neighbourhood.7a) (22.7b) In Eq.e. &H/d(fi = 0) if [m20 + ^>]</> = 0. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi.tp2. in a transverse direction. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 .3ml = 2ml = ml + TA 0 A 2 = 0.e. (22. 22.542 CHAPTER 22.7b) the constant on the left and the coefficient of A on the right of Eq. V?) (22.7a) vanish on account of Eq.e.g. >° for m 0 < °> .3). . e.e. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. i. as indicated in Fig.3) obtained above. i.2. Suppose we choose one such configuration. (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).2) and separating real and imaginary parts. 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. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 . along the trough of V (which we can call a longitudinal direction).Then m\ m\ = ml + A0A2 = m2. Classical Field Configurations the density Ti is minimized (i. we set cf>(x) = <f)c + ri(x). i. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. (22.6) Inserting this into Eq. This is the condition (22.
the existence of individual perturbation contributions). We can see the problem here by looking at Eq. ^ = h eip / \ ^2 l . tangential to the classical path. (22. irrespective of the question whether the perturbation series as such converges or not. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(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.7b). In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. has to be removed in order to allow a well defined perturbation procedure (i. There the wave function with an associated vanishing eigenvalue is called a "zero mode". which. though not identical phenomenon. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. of course. (3=0 Fig. i. 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.e. The elimination of the . t') = d(r .r')S(t .t'). Goldstone's theorem applies to fully relativistic field theories.e. 22. 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. r'.22. t. It is known from there that the vanishing mass of the photon together with the transversality of the electromagnetic field implies that only two of the four components of the fourvector potential A^ are independent. whereas ipi(x) is y ' " " " " " ^ . 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 £.3 The spontaneously chosen phase seen from above. Thus the Green's function is similar to that in electrodynamics. then a massless boson exists.
Later we wish to develop a perturbation series about a classical configuration of $.e. To insure that the energy (see below) is positive definite we require V[$] > 0. Zimmerschied and H.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). C. F. Soliani [237].£) in one spatial dimension. t Potentials of this type have been discussed by M.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. 22.* Potentials of higher polynomial order in $ than $ 4 can also be considered.t In particular the sextic potential can be considered along parallel lines. however. 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. W. equates to zero the coefficient of this wouldbedivergence. A. N.544 CHAPTER 22. Behera and A. Khare [17]. i. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. we shall need only at a later stage. 7711 = — 1 ( 1 6 Minkowski manifold). see e. *S. . instead we shall consider static and timevarying solutions in important models of one spatial dimension.g.d^} = ~d^d^V[^) = ^2^(^j V[$]. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. spinless field $(x. The Lagrangian density is taken to be C[^. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. Rebbi and G. Lohe [179]. and we choose the metric ?7oo = + 1 . The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. (22. 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.* We consider the theory (later: quantum theory) of a real. MullerKirsten [291]. We make one more assumption concerning V which. J.
i.12) where the prime denotes differentiation with respect to $ .l}.t) has an explicit timedependence it is a Heisenbergpicture operator.11) (22. ^ ^ * + V'[*]=0. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0.13a) We write such fields <j>(x).13b) We are not interested in just any solution of this equation. Since $(x. (22. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static.§ Unstable classical configurations are also important and will be considered in later chapters. (22. The field $(x. (22. From Eq.«{x) = V'(<p).12) we see that they satisfy the Newtonianlike equation <j. (22. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. 1].22. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. The first such condition to be imposed is that the energy of the solution of interest must be finite.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.e. (22. such that J > — — — = 0.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series. The other condition which we impose (and study in detail) is that of stability. at least in the present case. — m$ 1 — cos Im 1 I ^ $ (22. i. For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>.e.g] = \ v m For instance.l]. . In a quantum theory the field $ is an operator defined on a space of states. the socalled ^theory = ^V[g$. t) is an observable (which in general — in field theory as compared with quantum mechanics — does not have to be hermitian although we assume this in the present example).9) with quartic potential is defined by ^V\g$\ = ^V\g*. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions.
<9$ i. We observe here already that the integrand of the integral in Eq.2\dx 2 ' 1 fd& + V[$].17) Since we have to integrate over all space. (22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^.7r) where IT is the momentum conjugate to $ defined by dC ? r = .14) where the overdot implies differentiation with respect to time t. n = & .e. i. 22. (22. the classical solution will therefore have to be such that this integral is finite.e. (22.15) The zerozero component is equivalent to the Hamiltonian density H.4 The wellknown soliton of $4theory. d.T = <I>.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.T^ = 0. dC • Too = r?oo£ + =r$ = C + vr$ = H. (22.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. HI tanh m(x>k) 9 Fig. .546 CHAPTER 22.c = <£z + .
In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . — XQ). (22. O±m(xxo) (22. < = — tan [e >  ° ] 9 2 rrm/g — Fig. 4m . (22.777. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2.4. . In fact.10) and (22. m ( x .20) (22. 22. See J.x n ). in the case of the first.18) Here XQ is an integration constant.21) This sineGordon configuration is depicted in Fig.3 Soliton Theories in One Spatial Dimension 547 As we shall see below.13b) for the specific potentials (22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . the $*theory. it is not difficult to find classical cnumber solutions 4>c to Eq.5 The soliton of sineGordon theory.e. "The name sineGordon is derived from the analogous KleinGordon equation or theory. Rubinstein [240]. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n".5.18) is depicted in Fig.11). > . 22. 22.1 . Again XQ is an integration constant.22. i. (22.
<^>(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 .^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). In the case of the sineGordon theory defined by the potential (22. and hence 4/71 9<t>{x.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 .18) and (22.9 2 4>{x) = i. ± 1[tanh.o) =m7r.e. Classical Field Configurations The solutions (22.548 CHAPTER 22. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ. With we obtain g(/>(a. or \(<P'? = V(4>)+ const.13b) which we can write d \W? = !"(•».1 "»JtKxo) 9 1 . 4m In tan tan 5 l/e±m(*xo)tan^£^)U.* ( m' m 2 r)] ^2" .20) are obtained from Eq. Here the constant is zero for V(<p) and <j>' zero at x = ± o o . .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. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . (22. Inserting the potential (22.
however. The relation (22. this does not require the action to be minimized.4 Stability of Classical Configurations The concept of stability can be complex.17).22) and that its second variational derivative be positive semidefinite. However. the discussion of Bogomol'nyi equations) in a simpler way.e. instead of doing this now. A state of stability is therefore associated with minimized action and/or energy. the curves rising monotonically from negative to positive values are known as kinks. 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. 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. (22. (22. we obtain these expressions later (cf.19). i.17) implies dx Jdx dx n£)' + ™ ( ' • # \ dx J S(f>(y) \dx and with partial integration we obtain f dx d fd<f>\ dx\dx J  5V(</>) 5(x y) + 5(j)(x) Tx5{x~y) . Suppose now that we demand stability in the sense of minimized energy. Jackiw [141]. (22. for instance. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks".22. 22. SE((f>) 5cb(y) 0. Then we have to demand that the functional derivative of the energy E(<p) be zero. (22. . The expressions (22. in the onedimensional Ising model.23) See for instance R. the others as antikinks.21) for the energies can also be obtained from Eq.
in fact ip0(x) const. the integrated contribution vanishes. if we apply the generator of spatial translations d/dx to the classical equation. Since it arises from the violation of translational invariance by 4>c. and the EulerLagrange equation retains this property. (22. In fact. i.e. = 0. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.e. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. The original Lagrangian was.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite.e.13b) retains the invariance although the solution (f>c does not. i. of course. if 4>c{x) is a solution. Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). Thus the vanishing of the first variational derivative yields again Eq. The Newtonlike equation (22.550 CHAPTER 22. Thus for every translational shift "a" we obtain a new solution </>c in much the same way as we obtained new solutions by a change of phase in the Higgs model which we discussed before.25). (22. dx2 + V"{4>) (22. dx (22. a very general phenomenon which can be established as follows.A. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. we see that the latter has one eigenfunction. (22. The eigenfunction 4>'c{x) is for this reason called a "zero mode".28) . Eq. This is. namely (p'c.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. (22. with eigenvalue w2.a / 0 is not the same. its eigenvalues w\ > 0.13b).27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration. (22. in particular the invariance under spatial translations. written down in Lorentzinvariant form. i.26) Comparing this equation with Eq. the function 4>c(x + a). it is also called a "translational mode". A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. in fact.13b) with classical solution 4>c. Classical Field Configurations Since y ^ ±00.
17) we obtain oo dx[V(4>c) + const. (22. (22. i. 0 = S'[UC] = S'[gUc}. = 0. it is an expression like (22.32) = V'(</>).27). (22. (22. We can choose the potential V such that the constant is zero. this integral has to be finite. As stated above. We return to our considerations of stability. We see therefore that the occurrence of these zero modes is a very general phenomenon. lim [V(<f>c) + const.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. (22.] = 0 or x—>±oo lim V{(/)c) = 0 for const.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. in fact. Thus TUC is a zero mode.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.e. and we have seen that Eqs.31) Inserting this into Eq.22. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero. and that S[UC] is finite (or equivalently the energy).20) are such configurations in two particular models.]. 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.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. In the onedimensional case under discussion we can rewrite Eq.18) and (22. (22. we are interested in classical cnumber field configurations of finite energy.33) x—>±oo . (22. i.e. Then the vanishing of the first functional variation implies that 5S = 0. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). / oo Thus for E(4>c) to be finite. (22.
In the case of the 3>4theory we can define a conserved current k^ by (22. oo J—oo y The topological quantum number q is defined by 0. . (22. n = 0 . the potentials have degenerate minima (i. (22.1). 22.35) and it is seen that in this case q can be ± 1 or 0.34) jfc" = e^dv(l since d^k^ = 0. (22. » In the <fr4theory defined by Eq. . In the sineGordon theory > • defined by Eq. 22. the classical field (bc must approach one minimum for x — . ± 2 . ± l . .6 The minima of V in $4theory at x — ±oo. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. .6. Thus the classical energy or vacuum configuration is not unique.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. Classical Field Configurations V Fig. For a configuration to have finite energy.552 CHAPTER 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. V{4>) = 0 at different values of </>).11) we must have correspondingly for x — ±oo: » 771 9 2vrn.c o and another for x — oo.e. in view of the antisymmetry of eM^. . This number is defined like a charge in field theory by the spatial integral of the time component of a current. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—.
. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition.7T (22. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . Again we see that E(\<f>c) is finite only for A = ± 1 . Tfl \ (22. continuous values. ± l ± 2 . For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. In the case of the sineGordon theory the minima of the potential (22. We can see this as follows. and we call this "topological stability.37) for the onedimensional case here. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. in general. termed topological. 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.7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo).cos(27rnA)].22.38) 4 V(\<t>c) >m ~2[l .11) are given by ( so that for x — ±oo: > — J.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. and exp(ix(0))<pc reduces to 4>C when x(#) = 0. one may ask how topology comes into the picture. zero is not a meaningful topological number. .A2)2 + 0 except for A2 = 1. . First. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. n = 0 . . Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. A property is.
428.31) to (22. 22. 425 .** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents).40) (22. Eqs.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. (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).17)) f 00 > 0. i. Eq. (c) the time component k° is a spatial divergence. pp.13b) or ^ ' ( * ) ] 2 = V(</>).9 the foregoing arguments in a higher dimensional context. . 22. J. (22.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. (For a broader view we consider briefly in Sec. if oo / dxy/2V{<l>)<f>'{x).g. 00 (22. (d) the topological quantum number iV arises in a completely classical context.e.43) **See e. Thus with one space dimension higher there. 00 (22.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. (22. the energy is minimized. W. the factor A = exp[ix(#)] varies over a complete circle). Classical Field Configurations involves the difference An = rioo — n. We observe (a) the topological current k^ of Eq. its integral is nonzero only on account of nontrivial boundary conditions. MiillerKirsten [215].^ = N. (22.41) dxy/2V{$)4>'{x). The inequality is saturated. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. H.554 CHAPTER 22.
(22. n E .40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi.44) This first order equation which saturates the inequality (22.19). (22. Here (V(c/)) > 0.n > 1. . the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. Bogomol'nyi [27]. In the sineGordon theory we can consider. 555 (22. In the case of the <fr4theory we obtain 9 9^£ 3m? <t>{c°)=rn/9 <f>(oo)=m/g m 3 / j \ 4 m 3 9 2 2 9V 37 3^' which is the expression given by Eq.22. for instance.39).^ Clearly it solves the second order equation or (22. m 9 V V / and with x = g4>/m this becomes P IA\ CTnin(0) = = /^ J /^ V ™3 f2W A I A • 2X n—7T I ax\/2ll — cos x) = n—7T / ax* Asm — Z 3Jo 2ir r 9 Jo V 2 2m 2 cos — n — n^[2(l)+2] 2 ~ m3 8m 3 = n—rrin agreement with Eq.40)) this occurs when 4>'(x) T V2V(4>) = 0. Eq. y/V(ft) > 0) oo =2nnm/g r<p=2imm/ g / dxy/2V{<f>)</>'(x) oo /•rj)=2Trm/g = / J<j>=0 d</>y/2V(<f>) = = n Jd>=0 rt=Mmlg n / J<p=o d<f>y/2V(<l>) 2 / ^\ — # W 2 1cos— . We discuss this in the following Example. (22. 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.5 Bogomol'nyi Equations and Bounds We observe that (cf. B.21).
g.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. rotate) only in the (y.46) **See e. Classical Field Configurations Example 22.ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. Assume that the pendulums can move (i. Assuming that the pendulums can move (i.g. xn+\ — xn = a) 00 a —> dx. z)plane and only such that their elastic strings remain taut.t) 4>{xn. rotate) only in the (y.7 The pendulum analogy. Initially the pendulums are suspended freely in the gravitational field. from the yaxis as shown in Fig.e. We can now write down an expression for the energy of such a system of (say) n pendulums. . V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. At equidistant points xn = na along this string we attach strings with pendulum bobs. so that oo > u. all being identical. H. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22. z)plane and only such that their elastic strings remain taut the position of any one of the pendulums is determined completely by an angle <p(xn. Solution: Consider a string along the xaxis as described. Construct an expression for the energy of such a classical system and its Lagrangian.t)\2 +mgr{\ .7.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.e. 22.t) measured e. Goldstein [114]. Chapter XI.cos</>(x„. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) . t)). ka —> K.2 V dt + k[(f>(xn+i.556 CHAPTER 22.t) <j>(xn.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 . } n= — oo roc —> / L J . Fig. with mass m and connected with short strings to neighbouring bobs. 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. 22.
6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. The equation is called stability equation or equation of small fluctuations. If some w\ were negative. (22. Hence we set ${x.50) This equation is identical with Eq.47) into (22.49) would imply an exponential growth in the future t > 0 or in the past t < 0. wk would be imaginary and so the factor exp(—iwkt) in rj(x. It is instructive to obtain this equation by yet another method.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. t) of Eq. This number. (22. ' Equating coefficients of exp(—iw k t).22.t) = ° (22'48) (22. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. (22.13b) we have for small fluctuations rj(x.47) where n(x. Inserting (22. (22. We can understand what stability means in the present case.7 linking the pendulum bobs). which represents an expansion in terms of normal modes. (22. (22.49) exp(iwkt)ipk(x).25) obtained earlier.t) is a small . and so would invalidate the procedure which assumes that rj(x.t) = <l>c(x) + r}(x.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22.t). Eq.13a) and (22. t) is the fluctuation (field). is therefore also called "winding numbed. 22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. 22. the topological quantum number. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x).e.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): {w With the ansatz r)(x. of the continuous curve in Fig.
and need 9 = tan" 1 j e±m^Xo) 1. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists.50) has the general form of a timeindependent onedimensional Schrodinger equation. . i. For x — ±oo the quan> tity acting as "potential". cos46» = 2cos2(2(9) . If that is necessary. In fact.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ .e.52) tan 6 := e±m^xo). V"(<b) = 2m2 + 6g2(f>2. This means that the stability is not universal but only local in a certain sense. the zero mode. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22.558 CHAPTER 22. (22. Thus for stability w\ must not be negative. we obtain V"(<t>c) . i. we shall see later that the fluctuations n(x. Classical Field Configurations fluctuation.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. Inserting (j)c of (22.A \ Inserting <f)c of Eq. V"((/)c(x)). In fact.e.1. 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£^)*.20). approaches a constant value which is nonzero in both the <54theory and the sineGordon theory. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction.18).2 m 2 + 6m 2 tanh 2 m(x . The equation of small fluctuations (22.
e. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable. We can investigate the spectrum as follows.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.* For such a potential the spectrum can be b o t h discrete and continuous.53) = m(x — (22.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 . Those of the other case are complex and periodic at infinity.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. ^+(0) d 0. Consider the equation fd2 dz2 . which satisfy the boundary conditions V>(o) = o.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. We can therefore construct solutions of definite parity.ijj^ respectively. ip+. even and odd solutions.55) The differential operator is even in z. — XQ)] cosh[±m(x — XQ)]: 1 . i. A  *(* + !) cosh 2 z tp(z) = 0.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . .2 (22. (22. (22. and as a consequence a limiting form of the small fluctuation equation of all basic potentials.22. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.
Z)2. (22.e. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± . i.58) = (1 + sinh2 z)~l/2x(z).( 2 n + l . i.61) Discrete eigenvalues are obtained when the series (22.Z ) 2 . i.Z ) + ^(A + Z2) = 0.e. and in the odd case n+\)U+\l)+\(\ i.4 n ( n .Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. .( 2 n . + l2) = 0.e. Classical Field Configurations We now set (with ip either ip+ or ip) tp(z) = (cosh z)~lx(z) and change to the variable £ = sinh2 z. when in the even case n ( n .I) .560 CHAPTER 22. 2n < Z. A * An = .60) terminate after a finite number of terms.57) Then the equation for x{z) 1S (22. A ^ An = .60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.l2 = . (22.e.59) Setting (22.
2 . °2 i.1.. m i.N<1. We can see the continuous spectrum of Eq.6 The Small Fluctuation Equation 561 and the function ip.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0. w2 = Am2.50) we obtain with z = m(x — xo) tp(z) = 0.e.e.e. (22.. (22..65a) _dz2 cosh 2 z l=2. (22.2Z = . m N<2.( 2 .10) and (22.\=Kl2 From (22.63) we see that the continuum starts at A= 2 2^ = 0.j < I . N = 0. (a) ^theory: the equation Inserting (22.is normalizable provided 2(n + .64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.62) where even (odd) iV are associated with even (odd) eigenfunctions. 3m 2 .22.u2)—^V 2 dio and ip(z) = elkzV{to).2.62) we see that the equation possesses discrete eigenvalues w —2 .54) by setting X = k2 Then 1 . d w = tanhz.51) into (22. i. We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 . for N = 0. (22. We can now derive the spectra of the stability equation for the potentials (22. (22.11)..Ny. the continuum is w > Am2.1: w2 = 0. From (22.
.50) we obtain with z = m(x — XQ) the equation (22. . so that its derivative is nowhere zero (except at ±oo). This property can.e.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. It is now particularly interesting to look at their wave functions.65b) dz2 cosh^J. Thus either spectrum contains the expected eigenvalue zero. 22. In the $ 4 theory we saw that (cf. . of course.8 and we see it has the shape of a typical ground state wave function. N<1. We know that 4>c is a monotonic function. .8 The zero mode as typical ground state. %(x) Fig.562 CHAPTER 22. m d . The zero mode is depicted in Fig.. d . Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. .27)) . and — as expected — this eigenfunction is even. Thus 4>'c(x) has no node. (22. Classical Field Configurations (b) SineGordon theory: Inserting (22.66) We see that this is a nonvanishing function for any finite value of x. i. for N = 0: to2 = 0 and the continuum starts at w2 = m2.53) into (22. 22.=1A=^_. . m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. also be deduced from d(j)c/dx. the zero modes. Eq.
i.70) . (22.e.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.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ). . — XQ) OC Hence the conclusion is similar to that in the previous case. Similarly in the sine—Gordon theory we write so t h a t 4 /oo * " / > & • ) W171 dx oo cosh m(x — XQ) TYl = 4 — [ t a n h m O r . (22.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. .6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to .22.3 .= E(<f>c). d . ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. Finally we consider the normalization of the zero modes. without quantum corrections). 1 o rrr tanh x — .t a n h x .68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i.oc (22.z 0 )]oo = 8 . Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1. (22. . .e.
> 4>a{x) = 4>(ax).2d. we consider E(<P) which we also write as (22. P(</>) = f ddxV{(f)) > 0. for if such configurations exist in a realistic theory with three spatial dimensions. J As a first example we consider the expression (22. it is important to understand their physical implications.564 CHAPTER 22.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. Olive [113]. A. One can have different types of scale transformations — see e. ^For basic aspects of scale transformations see e. f G . H. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). (2.72) with T{$) = I' ddx]{V(j))2 > 0. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. . Affleck [5] who uses in his Eq. Goddard and D. *See e. (22.T(4>).71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . I. *The considerations of this section extend in part beyond quantum mechanics and into field theory.e. Derrick [68]. They may help. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. Kastrup [145]. I. Classical Field Configurations 22. i.4) <j>(x) —• a<f>(ax). however. ddx {V4>f + v{4>) (22. P.g.41) for the energy of the classical configuration but now with respect to a ddimensional position space.g.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . to underline the fundamentally different nature of topological and nontopological finite energy configurations.g. H.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. This question has important consequences.
. i.80) . (22.e. Consider first the Lagrangian density £[&M = \{d^)(d^) . we set 0 dE(<pa) da J a = i (22. Next we consider more complicated cases. the second derivative is positive) only for d = 1..e. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).79) In order to emphasize that (f> — (p^ in any direction.78) For the energy of a static configuration.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.73) we see that this implies (2 . d = 1 minimizes the energy.V(<P) (22. i. (22.\ eMo(22. we write • = lim J r .76) we see that E(<f>) can be minimized (i.d)T(<f>) = dP{4>). (22. since d2E(<pa) da2 2(2d)T(<f>).e. (22.75) Since both T and P are > 0 and d is integral.. x = (x1. to be finite the field <P ~^ 4>oo £ Mo for x — oo. a=l (22.. 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}.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}.71). 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.xd). Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a.x2.74) From (22. In fact.22. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension.
= —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . S1. i. .1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. — 1).566 CHAPTER 22. are the same. These observations are summarized in Table 22. i. therefore. and. Then each point at infinity is characterized by a different value of the polar angle 9. If we go to two spatial dimensions (d = 2).e. in fact. —1. We see. x = 00. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. as is the case for the constant solutions. Thus in this case S^1 is a disconnected set consisting of two points. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. (ptx = <f>oo. one set can be mapped into the other. is a connected region. i. —00 or equivalently S = + 1 .81) and (since d = 1) S = ± 1 . —00.1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const.e. that we have a set of disconnected classical vacua (</>oo><^oo.38).and sineGordon theories we assumed $ to be real. 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). Table 22.1.e. 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 . of • course. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00.
i. This will be "nontrivial" in the above sense only if 0oo is independent of 9.83) . the field must map position space into field space at infinity. Similarly.7 Existence of Finite Energy Solutions Suppose. for instance. $(r) : Sl . by allowing <fi to have three components in some "internal" (isospin) space. Thus. (22. i.2. if <p is real. to have topologically distinct classical configurations. We see. i.e. therefore.3 (^real) v{4>) = / (22. if cj> were real and position space twodimensional. 9) would depend only on r = x and hence the problem would be effectively onedimensional. i. .e. 00.1 . cj)(r. On the other hand. assuming invariance of C under the transformations of the group SO(3) in internal space. Thus.g. VMx)) with (0(oo) = linvxx) (p(r. that if the problem is to be topologically nontrivial. If position space were onedimensional. we have w e ca {Mx)}> and 1 3 1. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. we would have only two points at infinity.22.e. This can be achieved e. i. phase transformations 6Q ^> 9Q + 59.e. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}.e. the one point S = + 1 . * n reach any other <^(oo) by simple phase shifts.e. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.82) Now Mo := \ 4>(oc) ^24>2 = a2\.• M0. map into. i. —00 or S = 1 .
invariance under replacements $ — —$ in $ 4 theory.V{4>a].e.568 CHAPTER 22. Nielsen and P. Example 22. i. sphere. The symmetry of the potential (and Lagrangian) — i. . so that details of these models are irrelevant. A) = T(4>. Olesen [220].Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3. L. $ — > > $ exp{i Xlo=i Pa. through each point at infinity). if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line.„ where G r = 9 M £ . depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories.e.. + i ( D " f t W ) a . going once around S%~1 (i. 11 H. $ — $exp(ia) in a U(l) symmetric theory. A) + T{A) + P{<j>). $ — $ + 27 > • T in sineGordon theory.e. i. B. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. Georgi and S. Glashow [108]. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1.</>] = ~G^Gai. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). circle. the subsequent discussion requires only the general form of the Lagrangians. . the appropriate rotation..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) £(</>. we cover <S^_1 n times.3. 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.2. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved. it is effectively a surface (and so a continuous set) S2. (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. the winding number is n.e. % . Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space.a M J . a = 1.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". If.
84) and investigate the existence of finite energy classical field configurations..A). J 4 ) + A 4 . (22. (22. (22. T(A) = f ddxV(4>) > 0.d)T(A) . l j ddxGijGai] > 0.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. and so B ( ^ A . A). A) + (4 .d T ( ^ .<^[x) transforms like a vector. = 0. i.84) in this way one finds that the quantities involved transform T ( ^ .d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ).Ax) dX Since T((f>. 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd .e.7 Existence of Finite Energy Solutions where T{<j>. we have {^/»)A^ (22. Mx') = E with a scale transformation x — x' = Ax. A. Thus we set dE(<fix. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.i T(AX) = X4~dT(A). (2 .d T ( A ) + AdP(</>). A A ) = A2dT(^.T{A). negative.22. (22. <0 >• <£A(X) = <KAx).89) P{4>) are positive semidefinite and not all zero.84) is positive semidefinite and so must be finite by itself. (22.e. ^ A ) = A 2 . >• A„ x ( x ) = AAM(Ax). P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>.86) (22.dP(4>) = 0. Solution: Each contribution in (22.d)T(4>. some contribution must be .87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. We now consider scale transformations with A £ R. i. (i)^ 2d T Considering the contributions of (22. > Since T>IJ.87a) C?M„(x) — \2G^{\x).
(c) For d = 3: All three terms must be present as in the GeorgiGlashow model. (22. T^{<t>x) = \4dT(</>). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. . 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. (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). 22. of course.89). ft V .570 CHAPTER 22. Tchrakian and H. Landau [111]. MiillerKirsten [268]. (22. Thus such a contribution could counterbalance T{4>. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x).g.90) and we would have to add to Eq.t) such that  ^  2 describes the density of Cooper pairs. (e) For d > 5: In this case it is not possible to stabilize Eq. We then have a situation of the potential like that described in the context **T.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. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. a > 0.e. (22. H. In the following we consider a related theory. D. a = a(T) = a T —T c .89). Ginzburg and L. is temperature dependent and changes sign at the critical temperature Tc. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. A) or P{4>) for d = 3. W. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. (22. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. This case is. R. J. Skyrme [253]. 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. L. One can thus define a scalar field or wave function ty(x. H.89) the contribution (4 — d)T(4>). exemplified by soliton and sine—Gordon theories. 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.** e.
94) to check the dimension) 1 = I—a. 22. ^(00) = a (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.93) J' (22.e.2) £(^. but for T < Tc it has a double well shape. i. Variation as before. 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 < 0.94) 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 . Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material.e. (22. leads in a onedimensional case (considering this briefly) to the equation ^  = aV + /?^ 3 axz Imposing the boundary conditions (in the domain 0 < x < 00) ^(0) = 0. Clearly the length (see Eq.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. assuming there is no superconducting material in this domain).2): For T > Tc the overall potential has a single well (minimum). from x < 0) means — as is familiar from electrodynamics — replacing V by V . setting S£. (22.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. Sec. Example 22.22.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. Then the static energy density becomes (cf. i.(tp) = 0. The parameter 7 serves to put the minima of the double well at V = 0.
e. (22. The magnetic field on the other hand.96) ipH.g. (22. In the Maxwell equation See e. H. 22.97) we can rewrite j s as m \ h In macroscopic electrodynamics the magnetic field strength H is (in SI units) defined by H := — B . W. the order parameter \ip\ attains (approximately) its asymptotic value. h [^DV'V'DV]. The density of the Cooper or super current is given by the typical expression of a current. J. p.572 no superconductor CHAPTER 22. 22.98) where M is the magnetization resulting from atomic currents.M. i.e. 450. i. as indicated in Fig. 2mi Setting < / > \P\e vp B = ViA. We can see this as follows. 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 ). yJ—a/[3. i. . exhibits a completely different behaviour — in fact. V x M = j s . (22. MiillerKirsten [215].9 Behaviour of ip and B in the superconductor.9.e. Classical Field Configurations superconductor (x>0) Fig.
div grad" and V • B = 0. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.99) Inserting (22.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter. those of the socalled type I and those of type II.M i. Using the relation "curl curl = grad div . K i ™*.5 to 10~6 cm). this becomes A B = no V x j 5 . Hence 0 = V x H = V x ( — B . V x B = [i0V x M = /lois. (22. and V x (V x B) = / J 0 V x j s .8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. (22. In the onedimensional case we have B{x) = B(0)ex/XL. There are two types of superconductors. Approximating p by its equilibrium value..L B = O.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B.e. These vortices therefore carry magnetic flux which is necessarily quantized . i. m xi A B .22. This is zero in the present case. Also dD/dt = 0 in a static situation like the one here. the normal state of the metal begins to reappear in thin vortices and not uniformly thoughout the metal as in the case of superconductors of type I.e.
the space consists of two diametrically opposite points on the circle. A.. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or.g) twodimensional space as being bounded by a circle at r = oo.36). i. *A. (22. (22. of two points on a line.9) ^e^ '>\<poo\. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. Schaposnik [69] after their equation (3. Classical Field Configurations (see below). Abrikosov [2]. i. De Gennes [67]. In the two models of Example 22. J.e.102) In the onedimensional soliton case of Eqs. A. Finiteness of E requires \<t>\ = \<fioo\ on this circle.e. 'See the remark in H. . 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). G. The existence of these vortices has been confirmed experimentally.37).101) (p(r. See also P.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. in fact a periodic function of the polar angle with period 27r. de Vega and F.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. We can think of (e. Thus the asymptotic field </>(r. we must have that lim V ( H ) = 0. where exp(ix(0)) is an arbitrary phase factor. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. 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.574 CHAPTER 22. In the NielsenOlesen model defined in Example 22.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. in other words. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM). (22.9). 22.
States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig.X ( 0 = O)].2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. Fig. X is called the domain of / and y its range.22. 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 map A : S1 > S1 defined on S1 C R2 by (22. 0 < 0 < 2TT. i.1. commutative). Thus smooth deformations *See D.9 Introduction to Homotopy Classes 575 By a map or mapping / : X —> Y of a space X into a space y we mean a singlevalued continuous function from X to 1". The two theories we considered above in Example 22. W. defines the orientation of this line element.102) is called the exponential map of S1 onto S1.1). The 1sphere is the unit circle in the space of all complex numbers. Naturally an integer cannot change continuously. Suppose <j>.* 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. abelian (i. We may define this as « = ^[X(0 = 2 T T ) . . Sl := {zeC\\z\ = 1}. (22.10 States of winding number 0. Then a continuous function of (j> on [0.e. (22. topological group with the usual multiplication as group operation. 22.118)).10.e.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. Finkelstein and C. Misner [96]. 22. In mathematical language S1 is therefore a compact.
e.105) is described as "pure gauge" since it is determined entirely by the phase x(#).576 CHAPTER 22. we can write 1 d h J ee rde w At h r^oo .. (22. the winding number must be a constant of the motion. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below).e. i. i. r^oo 1 dx(9) n r o0 (22.g. we required that lim r—>oo w o.104) so that d2x hJ Since A has direction e#. one hole. The gauge field required in the asymptotic limit (22. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example). i.i'ldx(0) 9A Thus for (22.104) to be finite we must demand that g . no hole.) remain the same under homeomorphic transformations and are therefore called topological invariants. The winding numbers n are such topological invariants. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero. if it has one twist as above or a hole. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. For that reason n is called a "topological invariant?.e.. We observe that since time evolution is continuous. Thus two homeomorphic spaces have equivalent topological structures. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite.. connectivity.g.105) the corrections falling off faster than 1/r. The different kinds of connectivity of these topological spaces (e. elements in the same class have the same topological characteristic. In other . finite.
(22. (22.e. Of course.9):=t(f>£\9) + (lt)(f)W(9) with te[0.<b rd9—4^ = n. However. i. It follows that the field Fij.e. the vector potential A. A field configuration with one loop (or twist). the higher order contributions in (22. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. with n = 1. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22.110) . for B / 0 . the group space of (the phase transformations of) U(l) continuously. Therefore. cannot be pure gauge everywhere. it is determined by a gradient (l/r)d/dd. into S i .9 Introduction to Homotopy Classes words.108) map the spatial circle S^.105) insure that B ^ O .22. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6).106) J g j r d d g Thus the magnetic flux is quantized. the gauge field. However.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. cannot be deformed continuously into one with two loops (n = 2).109) A W(0) = e^ W. going once around S^. An example is the mapping H(t.l]. is also zero at r = oo. i. we can consider transformations which describe continuous transformations of a configuration with fixed n. we have B = V x A = curlgradx = 0. 9){^ and hence the functions (22. gauge transformation of a static abelian field is A > A' = A + Vx 577 and we know that the The field which is determined entirely by V x is therefore termed "pure gauge".e. if A = V x . Ffiu. we can go n times around S^. (22. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . i. the number of flux quanta 2irh/g being the winding number n.107) The fields <f>(r.
the unit interval I = [0.g.e. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8). i. e. f T h e standard texts on homotopy theory are N. Dittrich and M. Reuter [77]. H(1. But this type of combination is excluded sind H(t.9) = <fiW(9) and H(l. Steenrod [261] and S. A superposition of classical configurations with different winding numbers does not minimize the energy. 9) is an equivalence relation. Thomas [270]. H. time cannot make a field jump from one homotopy class (see below) to another. i.e.e.111) In principle an expression like (22. We can consider (22. In fact. Two curves which can be related to each other in this way.1].3: Homotopy an equivalence relation Verify that the homotopy map H(t. Classical Field Configurations which is continuous in the parameter t and is such that H(0. these two maps (fin from S^.9) is called a homotopy (emphasis on the second syllable).e. it is clear that this is also a homotopy. We verify these in the following example. — S\ > (22.lQSl^S^ with H(0. Example 22. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9).0) = $\6). i.T. H. A brief introduction can be found in the very readable book of W. 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.9) = ^°\0).112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. The relation (22. The map H(t. i. the phase factors of (f)(9).111) as a map from the space described by 9.113) . n.110) could be written down for configurations with different winding numbers. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. n' ^ n. i.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9). S%. 8) = cfinl\9). (b) symmetry and (c) transitivity. to the internal group space of £7(1). and the space of t. (b) symmetry and (c) transitivity. (fin (9) in the above example. Since the passage of time is continuous. Solution: As stated. (22. (22.578 CHAPTER 22. we have to show that the map satisfies the three properties of (a) identity. Hu [134].e. like <pn (0).8) itself must be a configuration with definite winding number.
0) = 4>W (0). . which therefore demonstrates that the homotopy relationship is transitive. .* 22. i. then </>W ~ <t>{0)• We define an H(t.ffi(O.0) for for 0<t<. negative or zero). U(l)). is such that H(0.t. at t = 1/2 both i?i and H 2 give c/^1'. Homotopy classes are sometimes also called Chern—Pontryagin classes. a class being a set of equivalent maps. Then H(t. We now define the function H ( f 0 ff . 0). so that H(t. since and tf (1. 0) = <£{2).9) = <f>(0\6) and H(l.0). being the continuous map connecting </>(°) with <j>(1'>.e. i<t<l. we see that of the set of these maps some are equivalent and others are not.9) H(0. 9) = <f>W (0). 0) = tf2(l.9) = . 0) = ^ (0) and 6(1. On the other hand H(0. i? is a homotopy connecting </>(°) with <^>(2'.0) = </>(0) is continuous there. as they are called.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. (c) We can demonstrate the property of transitivity as follows. (b) Next we check that if 0<°) ~ <p^. <f>W . \ M)\ H!(2t. 9) := H(l .22. 9) connects <pn (0) with itself continuously.e. Hence the set of all such maps breaks up into equivalence classes or homotopy classes.10 The Fundamental Homotopy Group The set of disjoint equivalence classes in the case considered previously is denoted by the symbol Here the subscript indicates the dimensionality of the onedimensional sphere S1. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). 0). In the above case of 7ri[[/(l)] we know from our earlier considerations that every equivalence class is characterized by a specific value of the winding number. so that H is a homotopy connecting <jyl>{0) with <f>^>(9). Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. Moreover. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'. 9) = H(l.e) tf2(2il. Returning to the maps \(x) : Si ^ S1^ (i. these groups are then called homotopy groups (cf. below). Hence <f>(0^ ~ <p 2 ).e. ' . 9) = H(0. is isomorphic to) the set of integers Z (positive. <^(2). (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent.
In the diagrams^ of Fig.115) Varying t € [0.g. 52.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. to make the result plausible. S 2 the surface of a sphere in a threedimensional Euclidean space.11 The trivial n = 0 map allowing shrinkage to a point. 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. p.11 the circle drawn with the continuous line is the circle S%. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. Sketch of idea demonstrating that nilS1] = Z.580 CHAPTER 22. T < 9 < 2TT. In order to really appreciate how topology comes into the picture here one should understand (e. Fig. which is to be mapped into another circle Si indicated by dashed lines. 22.e. characterized by winding numbers n are for instance (Xn(#) being a continuous function modulo 27r): (a) For n = 0: Xo(0) = 0 for all 9.1] 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. T (22. Rajaraman [235]. 22.) why 7i"i[S2] = 0. and then introduce some of the related mathematics. 22. How can we see this? We first sketch the idea. but TTI[S2] = 0: S 1 is a circle. Appropriate maps. .
i.s.e.s.h. i.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). of (22. to the case of winding number zero.117) in ±.22. We make another observation at this point. 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.h. 22. For winding number n we have the mapping (i.e. . Coleman [55]. 2 the maps are wound around the dashed circle. (22. The maps for these cases are illustrated in Fig. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n.116) In fact from this we obtain (see below) 2  n and this means r.12.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. and hence the winding number. 22. d (22.12 The n = 1 and n = 2 maps not allowing shrinkage to a point. (22.117).117) 2^ d6X(x)\1(x).118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. One can prove that the expression (22.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. In the cases n = 1. the transformations^ (the result may be looked at as 8n of expression (22. is invariant under infinitesimal (continuous) deformations. 0 < X(P) < 2vr.e.
. i.582 CHAPTER 22. 1 1 See N. In fact. Classical Field Configurations Under this transformation n changes by 5n Now. 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. the cyclic group with two elements (since 50(3) is doubly connected). whereas 7ri(50(3)) = Z2 (i. the identity class (the group space S2 is simply connected. only one homotopy class. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. Steenrod [261]. an integer modulo 2) i. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W).8. Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations.e. there is a theorem which says thatH TTg(Sn) = 0 for q < n. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided.e.e. 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. Thus there is only the trivial map into a point. 15. Sec.
e. finite Euclidean time) classical configurations (which are nontopological). and in particular the exponentially small level splittings of the double well potential. We consider such a theory now with action S((p) depending on the real scalar field 583 . An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies.Chapter 23 P a t h Integrals and Instantons 23. our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. in such a way that the factors appearing in the result are exhibited in a transparent way. different from literature. 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!). 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. Since we have not yet introduced periodic (i. 23.
23.1)) and hence E0 = mh/y/2. i. Path Integrals and Instantons S(<f>) / dt \i>2 . "For comparison with literature.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. Gildener and A.e. 2 2 4 ™ = ~.2.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O .1 The double well'potential. Chapter 6) 1 2 . (23. Fig. ! " ^ j = ^ 2 .2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. The Hamiltonian of the simple harmonic oscillator (cf. 23.1. (23.v{4>) <?VV m4 __ 1 .* /•tf CHAPTER 23. and ^2(0) : m* . note that here the mass of the equivalent classical particle is taken to be mo = 1. in particular E.584 <)>{t). Patrascioiu [110] and Chapter 18. n = 0.2) Fig.
</. We can develop a perturbation theory around either of the minima. 23._(0) = 0. as we did in Chapter 18 — but as we saw.2 The wave functions of the lowest states. i. W(0) = 0.2 — and we can deduce from this the boundary conditions.x) = x2 V(x). Fig. but there is tunneling between the two minima if the central hump is not infinitely high. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig.e. must have definite parity as in a physical > situation.g.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. e.23. This tunneling affects the eigenvalues. this alone does not yield the level splitting. and in the above 4> corresponds to x. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering).In mechanics of a particle of mass 1 with ~ position coordinate x(t). 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> ~ * ±</>. 23. The states of the system here must be even or odd under x — ±x. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . we have the Lagrangian L(x.
We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. But quantum mechanically it can tunnel through the hump from one well to the other. classically forbidden domains then become "classically" accessible.£'(0) = 0.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.  . The transition amplitude for this. (23.e.7.6) d*l dr 2 . Euclidean time T = it.e. i. i. dr We observe that Eq.4) J <bi =—(An rTf (ITLE = I JT.<h. dLE = 0.e.2. (23. (23.2 m 2 ( l m z (23. Equation (23. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.tf\ . that given by Eq. i.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21). In the semiclassical path integral procedure that we want to use here.3) for tf — oo.9) . r . d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. is finite if we go to imaginary. Then (0o. SE. (23. dr m+™ (23.5) and iS = dLE d dr[d(d(f)/dT)} i.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.2 V'(<l>) = .7) so that (23. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.3).U) = J'D{<f>}exp[iS(<l>)/h] (23.0o.586 CHAPTER 23. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.8) V(4>) + const.e. For small h we can attempt a stationary phase method. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time.
Fig.23. Thus we know the solution from there which is the kink solution (22. is depicted in Fig. 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. we see that the equation is the same as the classical equation for a static soliton (i. . and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest.3 The instanton path. Its trajectory. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. the classical path. from fa — —rn/g. this means the total energy of the particle in Euclidean motion is zero. and hence kinetic energy gained is equal to potential energy lost.e. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".e.2 Instantons and AntiInstantons 587 In fact.10) where TO is an integration constant. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p). in the present case (T — To) = m tanh m(r — TO). If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below).e. i. 23.3. i.18). — (23. 23.
thus indicating the nonperturbative nature of the expression.' . Tf = oo.4). but with the same Euclidean action (23.5). Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = .13) We observe that this expression is identical with the energy of the static soliton of the 1 + 1 dimensional soliton theory which we considered in Chapter 22. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. This antiinstanton configuration rises monotonically from right to left as shown in Fig. Path Integrals and Instantons The Euclidean action of the instanton solution (23. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.13).12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.10) is given by (23. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.8) we obtain T f SE dr[2V((j)c) + const. (23. 23.14) But this is not yet enough.13) into Eq. the constant must be zero.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1. Inserting the result (23.r 0 ) ' (23. (23.4. The configuration (23. We also observe that the expression is singular for g2 —> 0. The classical equation (23.].588 CHAPTER 23.7"/! ).13). we obtain lim f+ OO.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r .Tj) oc exp 4 m3 m^2 (23.7) admits another configuration which is the negative of (23.11) For this quantity to be finite for TJ = —oo.^ . (23.TJ T fro. Using Eq.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.
e.TO) + ^ c ( r . #(r) = </>C(T .15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). in the case of (23. 23.15b) since we require configurations from vacuum to vacuum. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.23.5.15b) has the shape shown in Fig.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig. 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. . as we can see e.5. b u t if TQ. 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.g. 23. / > Looking at the instanton (23.r0). (23. 0c(r) = &(T). The sum (23.15b) a solution of the classical equation? Is it a topological configuration — if so.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ.g.4 The antiinstanton.15b) is not exactly a solution.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. Clearly we are here interested in configurations of type (23. from the sketch or from the variation of SE. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.TO far apart and T'Q 3> TQ. 23.15b) In the case of (23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . (23. Questions which arise at this point are: Is the nonmonotonic configuration (23.15a) we have a simple superposition. and an antiinstanton localized at (say) TQ.
23.T 0 ) } ] .5 Superposition of an instanton and a widely separated antiinstanton. at oo the variation is <<> = 0). of course.15a) is a configuration which starts from 4> = —i^/g and ends there.e x p { . 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 .2 m ( T o . #(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 . Ignoring exponentially small contributions the Euclidean action of the configuration (23. since (23. g cosh m(r — TQ) (23. 5^ In order to answer the second question we observe that (23. vanishes as T\ —> ±oo (of course.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. Finally.15a) is not exactly a classical solution. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. T\] with LECIT = 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. In fact. we see that in the domain [—oo.16) .CHAPTER 23.15a) can be argued to be zero. it also does not exactly stationarize the Euclidean action. Path Integrals and Instantons '.
23.2 Instantons
and
AntiInstantons
591
and looking at Eq. (23.13) a n d ignoring exponentially small contributions, we obtain (note t h e same values of t h e limits)
<Koo)=(/>(oo) SE dA
Jd> oo)
#A*T
For t h e configuration (23.15b) we have with a similar argument
fTi SE \Joo roo JTx \ / /
rTi
( [ * • • • + [°° •••)LEdr=
( f
\Joo
1
...
f
i— Ti 1
...)LE
J
= 0.
Joo
As a further example we consider a configuration consisting (approximately) of two widely separated instantons (or kinks) and one antiinstanton (or antikink) localized at TQ , TQ and TQ respectively, § # ( r ) = ^ T  T o ^ T  T f M T i  T )
0(r  T I ) ^ ( T  r^VCTa  r)
+^(rr2)^c(rr(S3))0(T3r) with  o o = T 0 < r^ 1} < 2 \ < r^ 2) < T 2 < r ^ < T 3 = oo.
(23.17a)
Fig. 23.6 Two widely separated instantons and one antiinstanton. Again t h e configuration is nonmonotonic and as in t h e previous case one can show t h a t it is only approximately (with exponentially small deviations) a solution of the classical equation. In order t o answer t h e question concerning the associated topological charge of </>(3)(r) we recall t h a t this is determined by t h e boundary conditions of the configuration; thus in t h e present case t h e configuration belongs to t h e instanton or kink sector with topological charge
3
r (3) r
0
( E.g. in the overall range of r = 2T we can choose the locations TQ i ) : 2T/3.
2 T / 3 , T ^ 2 ) = 0 and
592
CHAPTER 23. Path Integrals and Instantons
+1 (like the single instanton or kink). Thus for each of the four charge sectors (corresponding to the two vacua and the kink and antikink sectors) we have a multitude of approximate solutions^ of the classical equation as illustrated in Fig. 23.7 in which (a) and (b) depict the two constant vacuum sectors and (b) and (c) the instanton and antiinstanton sectors respectively.
(a) (b) (c) (d)
Fig. 23.7 Classical solutions in their topological sectors. In evaluating later the Feynman amplitude about such configurations we use the completeness relation of states. Thus in the case of </>c (r) we write
oo poo
/
oo
d<t>2 /
J —oo
d<j>1(<t>f,Tf\(j>2,T2){<l>2,T2\(f>l,Ti)(<l>l,T1\<l)i,Ti).
(23.17b) Evidently the evaluation of this amplitude requires two intermediate integrations.
23.3
T h e Level Difference
The nonrelativistic Schrodinger wave function ip of a discrete eigenvalue problem for a potential V satisfies a homogeneous integral equation which we can write ip(x',t')= dxK(x',t';x,t)tp(x,t), t' > t. (23.18) Here K(x',t';x,t) is a Green's function satisfying the differential equation DK(x', t'; x, t) = iS(x'  x)5(t'  t) with differential operator d Id
2
(23.19)
(23.20)
" B . Felsager [91], p. 143, therefore calls these solutions "quasistationary configurations". Further discussion may be found there.
23.3 The Level Difference
593
(One can convince oneself by differentiation that the integral equation is equivalent to a timedependent Schrodinger equation). In Eq. (23.20) V0(x) = V(x)  SV(x), V = V0 + SV,
where 5V{x) is the deviation of the exact potential V from an approximation VQ. (Since we cannot — in general — solve a problem exactly, we have to resort to some approximation; thus we have to distinguish between the exact problem and the approximate problem which we can solve exactly; the Green's function is constructed from the eigenfunctions of the exactly solvable problem). The Green's function K(x', t'; x, t) can be expanded in terms of the eigenfunctions T/4 (x) of the Schrodinger equation for the potential VQ(X), i.e. (to be verified below)
K(x',t';x,t) =
We have (cf. Chapter 7)
^Y,f 
dE
'•i0)*(x')^0)(x)exp{iE(tlt)} EE&0)
(23.21)
" 1 d2  V0(x>) 2mo dxf2
^<?V) =  i « V )
so that DK(x',t';x,t)
•JL
dt1
V0(x') +
d2 2m 0 dx'2 EE, dE 2^
(0)
^h(x/)^n0)(x)exV[iE(t't)]
n
J
iE(t't)
=
i6(xx')5(t'
t),
(23.22)
as claimed (note that the first delta function results from the completeness relation of the eigenfunctions of the exactly solvable problem). We can go one step further and integrate out the independence of Eq. (23.21). Replacing En in the denominator by (En —ie),e>0,so that the integrand of (23.21) has a simple pole at E = En —ie, and integrating along the contour shown in Fig. 23.8, Eq. (23.21) becomes (with the help of Cauchy's residue theorem) H K(x', t';x,t) = Y, Tp{n}* (x')^n0) (x) exp[i£4°) (*'  t)}9(t'  t).
"This formula is discussed, for instance, in R. J. Crewther, D. Olive and S. Sciuto [60].
(23.23)
594
CHAPTER 23. Path Integrals and Instantons
This is our earlier expression (7.9).
Im E t  t>0
ReE
Fig. 23.8 Integration contour. We now recall from Eqs. (21.10), (21.17), that the Green's function K can be written as a path integral, i.e. that K(xf, tf;xi,U) = J cD{x}eiS/H{tf  ti). (23.24)
Evidently we want to relate Eq. (23.23) to Eq. (23.24) in the case of our onedimensional $ 4 theory. In the derivation of Eq. (23.23) we distinguished between the exact problem with potential V and an approximate problem with potential VQ which can be solved exactly. In an actual application such a subdivision can be a matter of convenience. Thus we could try to approximate V by taking as VQ the harmonic oscillator part of the potential and 5V oc <5 4 then the set {T/4 } would be the set of oscillator eigenfunctions ?>; which is wellknown. Alternatively, we could be interested in a completely different problem, e.g. one with Vo given by Eq. (23.1) and 8V oc <^6. In this case ipn would be the exact eigenfunctions of the Hamiltonian for the (/>4potential (23.1). It is shown for instance in scattering theory that the exact transition amplitude is characterized by poles at the exact eigenvalues (e.g. in the case of the hydrogen atom all discrete eigenvalues are simple poles of the scattering amplitude). Thus if we want to relate (23.24) to the path integral representation of the transition amplitude for the (/)4potential (23.1), we must take in (23.24) the exact (real) eigenfunctions ipn{x) and eigenvalues En for the potential (23.1) which, of course, we do not know. We proceed as follows. From the above equations we obtain K(fi,t';4>,t) = Y,M<i>')Tpn(<t>)eMiEn(tt')/h}
n
= f %>{<!>} exp[iS/h].
•*
(23.25)
23.3 The Level Difference
595
This equivalence of the standard decomposition as a sum on the one hand, and on the other hand with the Feynman integral on the right is the reason for describing the Feynman amplitude also as kernel. We now consider the left hand side of this expression for Euclidean time r = it. Since r' > r, the expression is dominated by the lowest and next to lowest eigenvalues E+,E associated with the ground and first excited state wave functions ip+(<fi) (even) and i/j(<fi) (odd) respectively. Thus the left hand side can then be approximated by (with h = 1)
K(<//,T';0,T)
~
^+(cf>')^+(4>)eME+(T'T)} +^_(0')^(<£) exp[E(r'  r)].
(23.26)
We are interested in the transition with oo and Moreover, V>+W>o) = V+(<fo), so that K(<f>',T';4>,r) ~ {^ + (</>o)} 2 exp[E + (T'T)] {</>_(0o)} 2 exp[£_(T'T)]. We define AE, E0 by** AE == EE+ so that E+ = E0Then
K{<J)Q,T'\<J)Q,T) <t>Q
(23.27)
^(0o) =  ^  (  0 o ) ,
(23.28)
(23.29)
(level splitting),
E 0 =  ( E + + E_),
(23.30)
AE,
E_ = E0 + AE.
(23.31)
~
{V'+(0o)}2exp
 ^  l A ^ ( r '  r ) (r'  r . (23.32)
{V>(<fo)}2exp
 \E0 + AE\
Here the wave functions at the minimum position (fro are comparable to (i.e. approximately given by) the corresponding ground state eigenf unctions of the harmonic oscillator and so (apart from overall normalization) (cf. Chapter 6) tp±{4>) oc < exp 1
(<£ + <
±exp
{<!><t>o?
= ±ip±{<f>) (23.33)
**Here AJE denotes the splitting of the originally degenerate oscillator level. This is therefore twice the deviation of a split level from the originally degenerate oscillator level.
596
CHAPTER 23. Path Integrals and Instantons
with E0 = mh/y/2 ( as we noted at the beginning of this chapter). Hence we can set 1 ± e _2 *° V#o)(23.34) l^±(0o)l = e2*§ ± I Then
K{(J)Q,T'\(J)Q,T)
« =
{V(^o)}2exp[JE;o(r'r)]eA^')/2eAE(^)/2 2{V(^o)}2exp[JB0(r,r)/^]sinh^:AJB(T,r), (23.35a)
where we reinserted h for dimensional completeness (observe that when AE(T' — T) = AEAT ~ h and small, the hyperbolic sine may be approximated by its argument and one returns to (23.25) for Euclidean time). The relation (23.35a) assumes fixed endpoints 4>o, i.e.
d<j>" j #'(V</>X</>" V
foMW.rW
+ 0o)<^'^>
(23.35b)
= <^0o) (</>0,r'  0o, r) (cp0\ip).
v '
Clearly without the delta functions the end points are not fixed. (This situation will be required in Chapter 26 in the case of certain bounce configurations). We now want to deduce the explicit expression obtained with the path integral formula and then extract AE by comparison. In this comparison we require r' — r to be a large Euclidean time interval (—T, T) with T — oo. It is clear from Eq. (23.35a) that we cannot put T = co rightaway; > thus we also require the study of a gentle approach of T to infinity, which necessarily makes the calculation more involved. We also note that since the above expression assumes quantization (discrete eigenvalues), the comparable formula must also involve quantization. Thus we must go beyond the purely classical contribution, i.e. we have to consider field fluctuations about the classical configuration.
23.4
23.4.1
Field Fluctuations
The fluctuation equation
We consider fluctuations n(r) about the instanton (or correspondingly static kink) configuration (pc: </>(T) = MT) + V(T). (23.36)
23.4 Field Fluctuations
597
Of course, the level splitting AE (to be extracted later) must be based on a symmetric treatment of instanton and antiinstanton. Allowing for fluctuations which are not tangential to the path one necessarily introduces more degrees of freedom — in much the same way as in allowing relative motion together with collective motion. Thus the consideration of fluctuations leads to the consideration of a larger (in fact infinite) number of degrees of freedom, i.e. to the continuum. We naturally impose on the fluctuations the boundary conditions 77(T = ± T )  T  O O = 0. (23.37) Clearly we have to expand the action into (23.5) we obtain
SE((J))
about
4>C{T).
Inserting (23.36)
«« />[(K
(ITLE = /
i(g+ £) + m + „ /«/>[K* + £,
dr
oo
1 / d0c \ ~ m"* 2 2 V dr J ' 2g
2
m
2J,2
1 2J,4 re+2*
+
WlA{t) *»**w+tf*
(23.38)
,2J2 J +Sg*<%rjz + 2<?2</>cr?3 +  5 2 T ? 4
Here the first pair of curly brackets {• • • } yields as before the contribution 4 m3
SE(<PC) = V~2
3?
(cf. Eq. (23.13)). In the second pair of curly brackets in Eq. (23.38) we use the identity
s:Am  MW.
dr where we used Eq. (23.37). Hence
—oo •2,
oo
dr2
(23.39)
dr2
•V(j),
+r,2{^924>l ~ m2} + 2</V0C + g2rf
(23.40)
598
CHAPTER 23. Path Integrals and Instantons
The term linear in 7 vanishes as a result of Eq. (23.7). Hence 7 SEW and SE(<j>) = = Here
oo
= SE(<l>c) + SSE(ri)
(23.41a)
JdrLE^c
+ v) + (23.41b)
SE{ci>c)+lJdTr]\T)L'E{4>)W
dr
/
00
\ it) ^^ ~ ^
jvi^eMSE^/h]
T){<p} eMSE(4>c)/h]
+
+2ffW+ .9W (23.42)
\
Hence, since the Jacobian from </>(T) to T?(T) of Eq. (23.36) is unity (with iS = —SE and subscript 1 in the following indicating that the single instanton case is considered),
I
exp[6SE(v)/fi] (23.43)
exp
4 m6 Jv{r)}exp[8SE(ri)/h}, 3g^h,
where we used Eq. (23.13). We consider the approximation in which we restrict ourselves in Eq. (23.42) to terms quadratic in n. This approximation is called the Gaussian or '•'•oneloop''' approximation.* Then
oo r 1 / j„ \ 2
dT
/ oo oo
(23J.0) f 00
\{tr) +'W*S»'>
2\dTj
dT
i/^y
2
+ rf m 2 {3tanh J m(T  r 0 )  1}
(23.44)
1
(7?, Mr?),
where M is an operator.t We set (observe that r acts only as a parameter which parametrizes the classical path; the dynamical — real timedependent — quantity is £n) V(r) = ^2^nrjn(T),
71=0
(23.45)
"The word "loop" refers to internal integrations which this nonclassical contribution requires. ^The factor 1/2 is extracted so that the fluctuation equation assumes the form (23.54b).
23.4 Field Fluctuations
599
where {r/ra(r)} is at every point r of the classical path a complete set of orthonormal eigenfunctions of M with eigenvalues {w^}: Mrtn(T) = wlrin(r), dTrjn{r)r]m{T) = p25nm.
(23.46)
Here we have in mind p 2 = 1. However, for ease of comparison with the literature* we drag the parameter p2 along. Then
(T?, Mr,) (2 = 5) E [°° dr(^Vn, M^Vm) = E l & l W 
(23.47)
Formally, if we insert (23.36) and (23.44) into h of (23.43), we obtain h = I 0 exp with 4 m3' (23.48a)
I0 = Jv{r,} exp
where
2^M??)
(23
^7%
fc=0
exp
n
(23.48b) D{77} = Urn FT dr](rk)
(23.49) (in our earlier discussion (cf. Sec. 21.6): 77(T^) — %, and r?(r) is given by • Eq. (23.45)). With v(Tk) = Y^nVniTk)
n
we have dr](rk) = E d CnVn(r k ),
n
(23.50)
and (the determinant being the Jacobian of this transformation) (23.51)
fc=0 fc=0
Hence I\ = exp 4 m3 3 ^
d
K^)/(n^) e [4? le " 12 ^
•fc=0
(23.52a)
•"•Note that a different normalization of {rj n } as in E. Gildener and A. Patrascioiu [110] (they have in our notation p2 = h/fj,2,fj,2 = 2m2) introduces additional factors on the right hand side of Eq. (23.47) and hence changes some intermediate quantities.
600
CHAPTER 23. Path Integrals and Instantons
Now,
2 2 d^e w P e/2
J
2TT
9 9'
(23.52b)
—c
so that the integral in Eq. (23.52a) can be evaluated formally to give I\ = exp 4 m3 3 ^
•*(%>) fl V C
d
k=Q
2TT
1/2
(23.52c)
provided no w^ is zero. We now show that this condition is not satisfied, i.e. we have one eigenvalue which is zero corresponding to the zero mode associated with the violation of translation invariance by the instanton configuration. This is the case if the zero mode, like all fluctuation modes, is normalized over the infinite time interval, not — however — if a finite interval of T from — T to T is considered, in which case the associated eigenvalue is nonzero (as will be shown below) and Eq. (23.52c) is well defined. Hence first of all we show that the differential operator M is the differential operator of the stability equation or small fluctuation equation. The operator M was defined by (23.44), i.e.
i r<x>
^{V,MV)
2
dr nMrj = \ drr/Mr]
OO oo
\fJTn
'dr]\
K£) + 7 ^ > (23.53) £ + ""<*•> V,
dr
J —(
where in the last step we performed a partial integration and used Eq. (23.37), i.e. n(dtoo) = 0:
dr]
°°
f°°
,
d2
From Eqs. (23.44) and (23.53) we obtain
M v"(4>c
= 6g2(f>c2 — 2m 2 = 6m 2 tanh 2 m(r — TQ) — 2m2 (23.54a) 4m 2 — 6m 2 sech 2 m(r — To).
Thus Eq. (23.46) becomes d2 1 (23.54b)
23.4 Field
Fluctuations
601
We have considered the spectrum {w^} and the eigenfunctions {r]n} of this equation previously; cf. Sec. 22.6. There we found t h a t the spectrum consists of two discrete states (one being the zero mode) and a continuum. In Eq. (23.45) we have to sum over all of these states. We also observed earlier t h a t the existence of the zero mode d4>c/dr follows by differentiation of the classical equation, i.e.
A.
dT2
dr'
:[V
(</>)], implying
dr2
V'itc
dr'
°'
Thus we know t h a t the set of eigenfunctions { ^ ( T ) } of M includes one eigenfunction T)Q{T) with eigenvalue zero. Before we can evaluate the integral (23.52a) we therefore have to find a way to circumvent this difficulty. First, however, we consider in Example 23.1 a gentle approach to t h e eigenvalue zero of the zero mode by demanding the eigenfunction ipo (r) to vanish not at r = oo but at a large Euclidean time T, as remarked earlier. § Some part of the evaluation of the p a t h integral will then be performed at a large but finite Euclidean time T, and it will be seen t h a t this large T dependence permits a clean passage to infinity. Example 23.1: Eigenvalues for finite range normalization
Derive the particular nonvanishing eigenvalue of the fluctuation equation which replaces the zero eigenvalue of the zero mode if the latter is required to vanish not at infinity but at a large but finite Euclidean time T. Solution: The operator of the small fluctuation equation is given by Eq. (23.54b). We change to the variable z = m{r — TO) and consider the following set of two eigenvalue problems: (a) Our ultimate consideration is concerned with the equation ( — j + 6sech 2 z  n 2 J ^0(2) = 0 with However, we consider now the related equation (b) ( —.r+6sech22n2)ipo(z) V dzz J = ^^(z) m2 with ^ n ( ± m T ) = 0, n 2 = 4 and Vo(±°°) = 0.
and our aim is to obtain ui 2 . We proceed as follows. The unnormalized solution of case (a) is (cf. Eq. (23.12)) dt/>c m 2 1 ipo[z) = —— = =—, z = m(T T0). orr g cosh z (Note that this is the zero mode which normalized to p 2 , i.e. / drr/^ = p 2 , is 770 = p(d(pc/dT)/^/SE, where SE is the action (23.13)). We multiply the equation of case (a) by i>o{z) and the equation of case (b) by ipoiz), subtract one equation from the other and integrate from —mT to mT (i.e. over a total T—range of 2T). Then
fmT
JmT
, f . dHo
V
j d 2 Vo\
dz2
t5g fmT
mz
, . r
dz2
j
J mT
This corresponds to a box normalization. trascioiu [110].
See also Appendix A of E. Gildener and A. Pa
602
CHAPTER 23. Path Integrals and Instantons
On the left hand side we perform a partial integration; on the right hand side we can replace (in the dominant approximation) the integral by that integrated from — oo to oo, and since this is the integral over twice the kinetic energy (and hence S B ( ^ I C ) of Eq. (23.13)) we obtain (remembering the change of variable from T to z) , dipo Vo — dz dj)o\ Wo . i mT dz J . diPo
dz
mT
™2 39 2 '
(23.55a)
In order to be able to evaluate the left hand side we consider W K B approximations of the solutions of the equations of cases (a) and (b) above. The W K B solution of case (a) is (c being a normalization constant and the ZQ occurring here a turning point)
^°( z )
=
~TJl e x p ( ~ /
dz 1/2
v J,
v(z)=:
 6sech 2 z
and
V(ZQ) = 0.
Thus for z —• oo we have v1/2{z) — n and tpo(z) —> 0, as required. The corresponding WKB > solution of case (b) which satisfies its required boundary condition, i.e. vanishing at z = mT, is similarly seen to be
*>(*> =
^
exp
Z**1/2)">(* L
/ dzv1/2\ + exp (  1 I
f,V2 dzv
exp
["dzv1/2
J zn
where v = n 2 — 6sech 2 z — (u> 2 /m 2 ). Differentiating the last expression we obtain d ipo(z) = —cv1'4 exp ( dz and so dz But (see ipo(z) above)
•>Po\ z=mT ^ CV 1/4 ' exp rmT
dzv1/2\
exp ( /
,f, dzv1 / 2
1
4'o\z=mT —  2 c 6 1 / 4 exp
rmT
/ Jz0
dzv1!2
dz
/ J z0
dzv1/2).
Hence (with n2 ~ n 2 — (uig/m2)
for T —> oo) and analogously —^o(—mT) ~ 2 — i>o(—mT). dz dz
dz
•4>o(mT) ~ 2 — ipoljnT), dz
It follows that the left hand side of Eq. (23.55a) is
4>o
But
^0(2)
:
mT dj>o mT ~ 2ip0—ip0 dz dz mT mT 4m 2
g cosh z
^
z
for z —> ±00,
so that 8m* e = F 2 i — ^ 0 ( 2 ) — =F — =F2V>o(z) for z —> ±00. dz g Hence ip0(±mT) — i>0(±mT) az = 2 ^ 0 ( ± m T ) — ^ o ( ± m T ) ~ q=4( — dz \ g
2\ 2 4mT
23.4 Field Fluctuations
Equation (23.55a) therefore becomes
603
if,
\
2 c
4mT „
^0 4m4
/
m2 Sg2
We thus obtain the result
wl z, 9 6 m 2 e  4 m T ,
wo ~ voe~2rnT,
vo • 4\/6«
(23.55b)
We observe that this eigenvalue is positive; the configuration associated with the zero mode is approximated by a classically stable configuration. For T — co the eigenvalue vanishes. Note that > here we used a Euclidean time interval of total length IT. This boundary perturbation method may be repeated for higher eigenstates which are then found to possess Tdependent eigenvalues.
23.4.2
Evaluation of t h e functional integral
It is clear from Eqs. (23.52a) and (23.52c) that the infinite product they contain provides a particular difficulty. It is the evaluation of this quantity that we are concerned with in this section. The highly nontrivial method was devised by Dashen, Hasslacher and Neveu* with reference to other sources. ^ We set Jo := det
drj(Tm
din
/(nIW)^',
^
dr
d£k exp
?£i« n\
2
2 2 Wnp
(23.56)
One introduces the following mapping or socalled Volterra transformation for large but finite T at the lower limit
y(T)
=
V
(
T
)

/
Z(~T)=V(T)
T
^ 0,
(23.57)
where N(T) is the zero mode, i.e.
N{T) =
=
HL
1
.
(23.58)
g cosh m(T — r 0 )
One can then show
fvu
*R. F. Dashen, T Integration in I. M. Gel'fand and and W. T. Martin
— see Examples 23.2, 23.3 below — that Vz } Dry exp 1 \1/2
2ith)
lJTAr2(T)]
1/2 T
[N(T)N(T)\
dr
1/2
/
TN*(T)\
(23.59)
B. Hasslacher and A. Neveu [63]. functional spaces and its application in quantum physics is lucidly described in A. M. Yaglom [107]. In particular this paper utilizes results of R. H. Cameron [43], [44] which are required in our context below.
604
CHAPTER 23. Path Integrals and Instantons
Observe that this expression is independent of the normalization of the zero mode N(T). Also observe that the transformation^ (23.57) transforms effectively to a free particle Lagrangian. After evaluation in Example 23.2 of the functional determinant occurring here, the integration requires the introduction of a Lagrange multiplier which enforces the endpoint conditions, i.e. boundary conditions, on the fluctuation rj{r) and hence on 0(T) itself; this latter calculation is done in Example 23.3. The evaluation of IQ with the above formula is then straightforward. We have for T — oo >
m
V K
'
'
1 g2 cosh4 mT
m 24e
4mT
and
T
dr
i
71 m4 m
rmT
/
JmT mT sinh 4z
32
dz cosh z
m°
+
sinh 2z
3z
mT mT
+—
ill
m 32
5
AmT
(
We observe how the largeT dependence cancels out in In and obtain the simple result In = det dr)(rm)
d£n
allfc
n
2n
w
1/2
m \ 1/2
kP2
(23.60a)
Note the variable T at the upper limit of Eq. (23.57) which implies that the transformation is a Volterra integral equation. This is an important point and should be compared with a Fredholm integral equation which has constants at both limits of integration. Consider a differential equation with second order differential operator M and Green's function K(x,x'), e.g. (M  X)u(x) = f(x), MK(x,x') = 5(xx'), a < x < b,
and with boundary conditions B\ [u] = 0, B2M = 0. The solution u(x) of the differential equation can be written u(x) = h(x) + \f
J a,
K(x,x')u(x')dx'
with
h(x) = f
J a
K(x,x')f(x')dx'.
For f(x) 7^ 0 this integral equation is called an inhomogeneous Fredholm integral equation; for f(x) = 0 the equation is called a homogeneous Fredholm integral equation. If one has an integral equation where K(x,x') is a function of either of the following types, K(x,x') = k(x,x')0(xx') or K(x,x') = k(x,x')0(x'  x),
the inhomogeneous Fredholm equation becomes a Volterra equation, i.e.
rx rb rb
u(x) = h(x) + A /
J a
k(x,x')u(x')dx'
or u(x) = h{x) + A /
J x
k(x,x')u(x')dx'.
In the book of B. Felsager [91] this transformation can be found in his Eq. (5.42), p. 190.
23.4 Field Fluctuations
605
Inserting this result into Eq. (23.52c) we obtain for the kernel h the simple expression ' m\ ' 4 m3 (23.60b) irh) 3 hg2 We see that this result is nowhere near to allowing a comparison with our earlier result Eq. (23.35a) for the determination of AE. The reason is that the zero mode has not really been removed, so that all fluctuations, i.e. perturbations, are still present. This has to be changed and is achieved with the FaddeevPopov constraint insertion that we deal with in the next subsection. Example 23.2: Evaluation of the functional determinant
Apply the shift transformation (23.57) to the fluctuation integral (23.43) and evaluate the functional determinant occurring in Eq. (23.59), i.e.
Thj
Solution: We first derive the inverse of the transformation (23.57) by differentiating this which yields
JV(r)
We can rewrite this equation as
Z(T) N(T)
f,{r)
N(T)
N{r)
N2(T)
T)(T)
:
d ( r](r) dr\N(T)
Thus
?j(r) r
N(T)
i= r
J^
T
dr '
Z(T) N(T)
' hand side of this equation yields
When multiplied by N(T),
r) Jj partial integration of the right
r)JV(r) I" K ' ' JT
V(T)
dr'—(—^—)z<T'), dr'\N{T')) y h N(r)£ ,
dr' N(T') 'V \Z{T').
r,{r)
=
z(r) + f(r),
f(r) =
(23.61)
For further manipulations we observe that with Z2(T) (obtained by squaring the expression of the first equation above) we have by differentiating out the following expression:
\ i, M^
dr~
2/ ^NM
JV(r) r,2(r)V"(0c)
,
N
, ,JV(T)
N(T)
, , ,N2M
'N2(T)'
so that since N(r) — V"{tj>c)N{r) = 0 (by definition N(T) being the zero mode)
V2{T)
+ V"(0c)r/2(r) = z2(r) +
^[^(r)
Mr)'
N(T)
606
CHAPTER 23. Path Integrals and Instantons
The derivative term will from now on be ignored since when inserted into the action integral and integrated it yields zero on account of the vanishing of the fluctuation r)(r) at the endpoints T = r 0 =  T , T. Thus 5S,
i
r1
dr
+ v"(<pcW
£
dr
z2(r) 2 v
;
The functional integral to be evaluated is To of Eq. (23.48b) which excludes the classical factor. Thus
h
/^ } exp[/; /T {i(J) 2 + i^(, c )}
jv{z}
D{V} Dr]
f.A\i2{T)]
D{z},
Dz Dr] Dr] Dz
T>z Dr]
Our next step is the evaluation of the functional determinant. Prom Eq. (23.57) we deduce
dr]{r") ) 7
^L=S(TT")+
(T
K(T,T')5(T'
T")CLT'
with
K(T,T')
.
JT
N{r'Y
(23.62)
where we allow for a general r dependence in the Volterra kernel as well. Discretizing the Volterra equation (23.57), i.e. Z(T) = r,(r) + fT K(T,r')r,(T')dT',
or more generally an integral equation of Fredholm type, by setting 5 = Ti+i — T;, i = 0 , 1 , . . . , n with TO = —T,rn — T, we can write the equation
z r
{ n)
= ^2
finiVin)
+
5
5Z
K T
( n>Ti)v(.Ti)
(23.63a)
We observe that K(rn,Ti) = 0 for % > n. The discretization here is a very delicate point. A different and convenient — but not unique — discretization (discussed after Eq. (23.66)) is to rewrite the equation as
n n ..
z T
( n)
=
^2 Snir}(Ti) + 5^2
n r n
K T T
( ^ i)~{v(n)
+
v(ni)}
cn—i
] T SnrtTi)
+  £
K(r,Ti)f,(Ti)
+ ~Y,
K{r,Ti+1)r]{n).
(23.63b)
Equation (23.63a) can be rewritten in the matrix form of a set of simultaneous linear equations: / z(n) \
2(T2)
/
i + i^Cn.Ti)
5K(T2,TI) 5K(T3,T1)
o
1 + <5K(T 2 , SK(T3,T2)
T2)
z( 3)
T
1 + SK(T3,T3)
0 1 +
0 0 0 0
6K(T„,T„)
V{T3)
V 2(r„) /
V
*See also I. M. Gel'fand and A. M. Yaglom [107], after their Eq. (3.4).
H.23. we obtain exactly the nonvanishing terms on the right of Eq. T. the value of the determinant is half (or arithmetic mean value) along the diagonal. Cameron and W..T')dT' (23. We observe that if we discretize Eq.n) i=l N lim exp £*(T.n) o K(TJ. (23. 2! .63a). r i ) + . .exp • /_:T JV(T')" dr' = e x p [ . (23. This case has been considered in the literature^ where it is shown that if the Volterra transformation is considered as a Fredholm transformation whose kernel vanishes on one side of the discontinuity or diagonal. whereas the determinant arises only once in the original determinant Dn. Thus D„ = l + 5 ^ K ( r i .64).TI) 0 K(T2. X(TI. the kernel K(r.N{T) Finally we add the following remarks concerning the discretization (23.e.. .Ti) i=l : exp \ £ K(T. 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. as in Eq. i. are summed over all values from 1 to n.T2) I exp I K(n.T1)dT1 + J dn J dr2.67) i=l v ) = lim exp V In ( 1 + ' L K(T. Thus in our case one of the kernels is zero and we have to decide what we do when.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .a { l n i V ( r ) . [44].j.64) Inserting the explicit expression for the kernel given in Eq.66) •Dr) .63b).TI) K(T2. 192.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.68) D above. (23.lnJV(T)}] N(T) N(T) (23. p. n )d? (23. Felsager [91]. where a f 1 See R.n) K(Tj.. Our final step is to relate the VolterraFredholm determinant to the required functional determinant. (23. r ' ) is to be evaluated at r = T ' . This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). ^This is the method explained by B.7 Y.64) as above and consider the determinant of the expression. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. so that (with T *T) 1/2 •Dz N(T) (23.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. Thus the factor a inserted above has to be taken as 1/2. = 1 52 K(n. Martin [43].64) D (23.n) z=l x (23.. In the limit S —* 0 . 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. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — 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.
59) the endpoint conditions »)(±T) = 0 and evaluate the integral. I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23.608 CHAPTER 23.Z ( T ). N(T') . and hence may be ignored.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23. In = dzi dz2 • • • dzn e x p . so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. (23. rn+i = T.22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l .Z i . I2 and using induction one obtains (cf.( . (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). Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T).70) and this now requires also integration over the endpoint coordinate z(T). (23.73) . f(T) = N{T) J^ dr'^^z(r') (23.e—*0 In.2o) 2 rc^O. Evaluating Ii. In every integration / dzi the contribution €7.^ . Eq.59)) + f(T)] = — f dz(T) f°° dae^W+f™.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. With partial integration (in the second step) we obtain (since (cf.l + £7t) «i (23.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. (21.(T) + /_ dr — .71) Hence (with h = 1) l = j .72) The endpoint integration dz(T) = dz(rn+i) can be combined with this. T + N(T) / T' = T JT dr' N(T') . represents a constant translation.— .57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . Eq. Solution: From Eq. Briefly. Path Integrals and Instantons Example 23. n J=.N(T) . (23. (23. .0.
(21. ^E. . 11 L. (23. Popov [89].59) and agrees with that in the literature. £n = ~2 dTT](T)7]n(l (23. / — / / We started off with V(T) = ^CnVniT).3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode. 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. / d£o <9r0 5[F(TQ.23..10) and (21. Eq.4. D.£Q) — aTo]dTod£o.75) or more generally a threedimensional surface integration j dau by the volume integration f cftx through the relation dF(x) 6[F(x) — otTo]d x. (3.4 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.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. Ignoring the remaining metric factor with the reasoning explained > between Eqs.e. This result verifies Eq. i. 1/2 T W 2 (r)J (23.74) where we reinserted h. N." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const.^ 23. Gildener and A. Patrascioiu [110]. a = const.12). (23. J co § dr Recall that / f ^ doce^ = yfVfp. j / dTT]n(T)r]m(l r . Faddeev and V.11).
e.81) the rdependence is integrated out and hence the integral does not provide the function F(TO. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). In the integral of Eq. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).610 this means that CHAPTER 23. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. i.e.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). i /*oo (23.81) 2 / dr»7o(r)0c(r) = .e.80) i r or •i /*oo dr7?o(r)0(r) = 0.83) **The collective coordinate in the present context is also called "instanton time".75) enforces the collective coordinate ro to be given by F(ro. The delta function 8[F(TQ.£Q) — aro] contained in Eq. i. Path Integrals and Instantons _^ ^ w .e. this constraint implies that we go to the subspace which is orthogonal to the zero mode. (23. demand validity of) the constraint £o = 0. The number £o is the component of the fluctuation rj(r). We consider the following integral / ( r 0 ) := . / dTm{r)n{T). where L = in + cn.77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration.76) we set 4>c{r) = ^CnVnir). if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .75) requires. in the direction of the zero mode.^n{r). We establish the function ^(To>£o) with the following reasoning. (23.£o)/a. (23. (23. at r = ro. We now impose (i. . Then cn = ~2 / dr(pc(T)rjn(T).^O) with dependence on ro which the replacement (23.g.79) (23. (23.78) <t>ij) = Y. (23.J drm(r)Mr + T0) (23. Thus. In analogy to (23.
87) l/M) (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T). .y/S^ro] (23.£O) OTQ = = £ " ..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 ) . 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) . (23.Zo)y/SETQ We now define A FP '• =0 • (23. jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.75) 1 = AFP f dT06[{F{rQ.88) ro0 Then in Eq. so t h a t with F(TQ. 158.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.^) .23.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). (23.89) Here the factor App is called the FaddeevPopov determinant.<t>c{r)} .91) " F o r more discussion of this point and related aspects see also B.85) PJ W i t h Eq. . (23./ drri0(T)<pc(T) + PJ (23.y/SET0 = 0. p. Felsager [91]. We have AFP = /F(T0.84) (23.86) = Jdrrjoir^Mr F(TO. We evaluate this quantity as follows in the leading approximation.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.(23.
the "static energy" invariant. i.93) where AT = TT — 00. i.e. The factor p is a normalization factor which we > retain here only for ease of comparison with literature and may otherwise simply be put equal to one. 770(7").77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. tangential to (J)C(T)) leaves this and hence the square of this quantity. P the relative sign being of no significance. /(TO) ^/SE~T0} V~SET0 c d£.e./ dT0AFP. Inserting (23. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. if the coefficient of 770(70) vanishes.e. ~ There is a quick way to obtain the result (23. 0 d r (23. .75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 .91) into Eq. if ^ + d£o = 0. Thus the result is that the integration over the tangential or zero mode component £0 is replaced by an integration over the collective coordinate 7 o multiplied by the FaddeevPopov determinant. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). Consider.93).90). P J AT (23.e.92) We find therefore that the value of the FaddeevPopov determinant is given by the square root of twice the classical kinetic energy or simply of SE^ The relation (23. =~ (23. using the relation (23. i. (23. 7 Thus there is no fluctuation along the direction of the zero mode in the sum.0AFP8 4o H P . in d(j)c(To) + dr/fa).94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals.612 CHAPTER 23.
(23.96) does not yet allow a comparison with (23.93).52a) and replace in this J d£o by the expression (23.96) lim .60a) in not involving the zero mode contributions. 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. We determine I'Q by observing t h a t this differs from IQ of Eq.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.98) . We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton.23. Since w\ is for finite T given by WQ of Eq. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.97) is a constant still to be evaluated.4.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.4 Field 23. (23.43)).A T A F P i " o e x p T—>oo p where I'o n^O. We observe t h a t the result (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.60a) for I0 and Eq. (23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. (23.55b).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. 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.
This means we are concerned with configurations like* 2n+l d(r .a n t i .2 m T e x p 2 irgh ' T»oo (23.r). Patrascioiu [110].99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. In view of their normalization of the normal modes. in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells. i. the approximation is then known as the "dilute gas approximation".TiMiy+'Mr «>.101) 2T > oo (cf. (23. Gildener and A.32). they have to introduce the inverse of that factor on the far right of their expression (3. ^See E. their Eq.5 I n s t a n t o n .614 CHAPTER 23.4. we drag the factor p2 along for ease of comparison with literature.55b)) w0 = v0e VQ 23. (3. *For the validity of the calculation the instantons or kinks have to be far apart. Gildener and A. . 0c M.e.* Finally we rewrite the expression (23. Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. ~ rKo')e{Tx . (23. Although we choose p2 = 1.20). 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 . Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 .100) We observe that the dependence on the normalization constant p cancels out. the result becomes h lim 2 m r ^ ^ 1 e .102) *For comparison with the parameters of E.100) in such a way that it exhibits the crucial factors of which it is composed.4>C exp h (23.i n s t a n t o n contributions As already mentioned.
36) (/. r0(3) = 2T/3. Consider (with T 0 = T.fo) = Id2 St SE(<Pf.l ) i + 1 in the sum of Eq.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0.r / 3 .r 0 ( 1 ) = .H)2 + . (3) (23. ( 2 3 .106b) SE{<t>l. (23.Tf\(p2.4 Field Fluctuations 615 for n = 1. are varied.4>I) ld2SE SE(c/>i. (23. 2 . . we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. the total Feynman amplitude is given by (^/.Tl}{4>l. 1 0 6 a ) and SE{(J>I.(l>i) (23. 23. this means we have to consider the amplitude (23. (23. The factor ( . T2 = T/3. Since these endpoint configurations.Tl\(j)i. r 0 ( 2 ) = 0.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.<t>2)eSE{<t>2.6.4>l)eSE{4>l.<j>i) .2 T / 3 . 7 i = .104) Considering the case n — 1 of a kink with one additional kinkantikink pair.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f. .4>i.23. T3 = T) d(f>l((pf.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).T2\(f)l.<Pi)2 + • • • . .r/l&.T2}((t)2. Proceeding as before we now consider first in analogy with Eq. (n = 0 implies the single instanton solution).<t>ff + h . Taking these configurations into account. here (f>2.
106b) is to be substituted into the second relation of (23. (23.. ±m/g (23. i.616 CHAPTER 23.2J.108) where the last expression is obtained as follows. d2SE{4>) d(j)2 2m. rn ^2 J2H 9 1m' ±2gcj). The allowed positions of the individual kinks have to be such that their order is maintained.4>i) { 2 " 2d SE . Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method. On the assumption that the kinks of the instantons are widely separated.e.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations. (23.4>i) / d(f>2 e x p {<h . each of which implies a multiplicative factor 7. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f. the . Path Integrals and Instantons where the first expression in Eq.2d2SE ~c 4>i d(f)i e x p SE^f. the dilute gas approximation. of course. each of the three intermediate kernels contained in Eq. (23. i.12) we have JM~°°) i. dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 .(01 .105) can in other respects be treated in analogy with the single instanton amplitude.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. in the case of l2n+i therefore ry2n. i.e. that each applies only to a fraction of the overall Euclidean time interval of r = 2T.e.e.4>f) d4>\ 4>s72 (23. It has to be remembered.3 9 L 3m 2 and 9SE((/)) ± .106a). With our remark after Eq.
112) We now compare this result with expression (23.105) t h a t each of the three amplitudes contributes a factor e x p [ . and in the case depicted in Fig.Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.7 3 ^ — .110) Finally we have to recall from Eq. we now have to integrate over three collective coordinates. Am3 2m J rW ( r ( 0 and this followed by setting aj. Gildener and A.)(3) 1 o A T 3 Ao lim .r.r/^.(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23.^ . T ] with ] T 3 = 1 A » T = A T = 2T.6 the collective (i) coordinate TQ has to be to the right of TQ .4 Field Fluctuations 617 replacement (23.35a) which contains AE in the argument of sinh and obtain for the level splitting A J resulting from consideration of the instanton plus any number of instantonantiinstanton pairs A*£ = 2n7AFP/0^=exp 2TT 1/2 4 m3 m V2 / m \ 1/2 4 v / g ^ exp ( 2TT irh _±rrv^ 2h ^With shifts r. a 2 0.T/0j.e.93) has been applied. then sign reversals TQ i ) . Patrascioiu JT/3 ° ir< >2T/3 1 dr.V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.(*) o ax = 2 T / 3 .a 3 [110].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 .A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23. 23.23. (3) ° JrW: 2T/3 (2T)3 .2 T / 3 this product becomes that given by E. (23. 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) . In the case of (<^f. i.101): (^/.m A .
Taking similarly into account the antiinstanton (since the physical level splitting does not distinguish between an instanton and its antiinstanton).182) of the Schrodinger equation method provided the difference in the mass of the particle (there taken as 1/2. H. J. Gildener and A. Bose and H. 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. The difference between levels. (23. To help the comparison we note that — with indices indicating the authors. whereas the method of the Schrodinger equation in this case is of the same degree of complication as for the ground state. W. We observe one more aspect of our calculation. Finally we realize that the computation of (23. Path Integrals and Instantons 8V2hm5/2 = exp ( 4m3\ . MiillerKirsten [30]. . — the relation between the parameters is ^LMK = ^I^GP = 4 r " 2 a n < i C\MK ~ \^<3P = 92. ^See P. Wiedemann [4].Q. (23.^r . J.W We also see that the level splitting is a nonperturbative expression (with an essential singularity at g2 = 0).With this correspondence A\E of Eq. Patrascioiu [110]. AE. by either replacing A T by 2AT or doubling A*. W. 2 V TT [ i ( g „ _ i ) ] ! V 9 J V 3 9 HJ For an elementary discussion see S. Liang and H.113). GP for those of E. (18. J. W.113) is very complicated. K. here as 1) is also taken into account. J. MiillerKirsten and A. has been obtained without explicit use of canonical commutation relations (i. even without explicit specification of the prefactor. l~2h = 2qo+2\ = m /4m3Vo/2 / —*exp 2 P 4 m3 \ — .113) for qo = 1: A1E „ . Achuthan.182) (with mass = 1 and H = 1) becomes the following which implies twice the result (23. l CHAPTER 23.113) This is (in slightly different notation) the result of Gildener and Patrascioiu [110] for consideration of the instanton plus associated pairs.618 and hence AE A.' 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. We shall see that the path integral applied to excited states is even more complicated in necessitating the evaluation of elliptic integrals.e.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. 23. we obtain AE = 2A* and agreement with the result (18. MiillerKirsten [163]. explicit quantization). . 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.
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. and their interpretation and uses.2) . Consider for instance the temperaturedependent potential yr(0) = i A ( 0 2 . Previously we considered in particular the quartic or symmetric double well potential for a scalar cp.e. in a somewhat different formulation. i. We shall see that these classical configurations are associated with quantum mechanical instability and hence with complex eigenvalues. Since problems of instability are ubiquitous and occur in all areas of microscopic physics their understanding is of paramount importance. however.a 2 ) 2 + c^T2.Chapter 24 P a t h Integrals and Bounces on a Line 24. Potentials of this type play a significant role for instance in models of the early inflationary (expanding) universe (as a result of its cooling). 619 (24. ™=£('^° < 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).1 Introductory Remarks In this chapter we are concerned with nontopological classical configurations on R1.
620 CHAPTER 24. A2 _ a — \„2\ u ) ~r u/\y> 2 (24. We can destroy the reflection symmetry of V((f)) (i.2=a cT /\ Thus there is a critical temperature T c := (X/c)l'2a. a spontaneous symmetry breaking leads to the double well (i.a2)4> + 2cTz 2 0. > / V[<H 'true'vacuum 'false'vacuum Fig. = 4(ACT2 cT 2 ). T = 2\(<f>2 . 24.1 The asymmetric double well. J 4.e. but implies instability for T < Tc.3) Since d2VT{4>) 2 d^ we see that ~ ^ ' > (d2VT(<f>)\ V # V # 2 2 A=o 2 2 = 2(cT 2 . We see that <fi = 0 minimizes VT(</>) for temperatures T > TC.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.e. Path Integrals and Bounces on R1 Here dVT{4>) vanishes for . at the minimum of the potential the case < = 0 changes to the case 0 ^ 0 ) . its symmetry under .
the higher minimum is called a "false vacuum". The Lagrangian density of the theory is given by C = dp*d»*U(*). and (c) determine the lowering of the energy to the true quantum mechanical ground state (i.g. H.6) The EulerLagrange equation is given by • $ + */'($) = 0 (metric + . a linear term as in w^^'A1^'*0V(<l> = m/g) = 0.Y.1 Introductory Remarks 621 exchanges 4> <* — <P) by adding e. M. also in contexts of string theory. Cornalba. the quantum mechanical vacuum). 24. perturbationtheory vacuum). We shall see that: (a) solitons (or equivalently instantons in the onedimensional theory) describe vacuumtovacuum transitions.1 with (24. <44 2) ' This is now an asymmetric double well potential as in Fig. (24. Guth and S. see e. Penedones [59] and references cited there. S. L. . Pi [126].g. Costa and J. tf(*) =^ 1 . and (c) determine iQE of E which results from a nonvanishing probability of tunneling away from the classical vacuum only. We recapitulate briefly a few points from our earlier detailed treatment of (1 + l)dimensional soliton theory. for instance.t In the following we consider first (after some recapitulations) a simple model of a false vacuum.e. which (b) are saddle point configurations of E (or SE). A. V(<j> = m/g) = 2e. False vacua play an important role in models of the inflationary universe* and elsewhere.7) *See e. . the vacuum being understood as the classical. On the other hand we shall also see that (a) bounces describe vacuumtoinfinity transitions. Since the true vacuum minimizes the potential.g. 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). (24.24. (b) they are minimum configurations of the energy E (instantons of the Euclidean action SE).^J >o.) . T False vacua and bounces occur.5) We see that one minimum of V now lies higher than the other.
12) Since <fic(x) solves Eq. The energy of this classical configuration is obtained by evaluating the Hamiltonian for (f) replaced by <j>c. We have seen previously that the same equation is also obtained from the second variational derivative of the energy functional.13) For small w\ > 0. small perturbations around </>c(x) do not grow exponentially with t (the zero eigenvalue requires special attention). i.10) Inserting (pc and evaluating the integral we obtain E&c) Am6 3ff 2" (24. (24.t) dt 0 are written (f>(x). Path Integrals and Bounces on R1 Static fields given by d$(x.14) . dx2 + u"(<j>) 5(x y). — XQ) (24.e. which is an important point as we shall see.9) 9 for the doublewell potential given above.8) This equation possesses the kink solution <t>c{x) m tanhm(a. We observe that this kink solution is odd in (x — XQ). (24. h " 52E(<j>) 8(j){x)5(j>{y) +u {Mx)) = ip {x) k wlipk(x). (24. Thus dx J —c 1 k'\2 {4>'cy + c/(0 c (24.7) yields the following equation called stability or small fluctuation equation. t) = <f>c(x) + Y^ iPk{x)ekt (24. so that the EulerLagrange equation becomes the Newtonlike equation <t>"{x) = U'(</>). we see that (24.11) We investigate the classical stability of the configuration by setting in the full t—dependent equation for small ipk: iw $ ( s . i.8).622 CHAPTER 24.e. (24.
We have seen that general considerations tell us that the zero mode is given by the generator of translations applied to the classical. In the case of the double well potential (24. . 22.16) V[<H x= oo 0 d2 .e. (24. the other a positive quantity) and a continuum which are given respectively by w\ — 0.17) . the zero mode).e.1) the fluctuation equation becomes 6m 2 ipk(x) = wkipk(x) (24. eigenvalues w\ > 0 (actually the stability is only local in the sense that E is not minimized in the direction (in Hilbert space) of the eigenfunction associated with the zero eigenvalue.24. 24. underneath the particle's velocity. i. configuration (for our </>4theory). 1 —<pc = ~r~ —tanh mix — xo) oc dx dx [ g cosh m(x — XQ) (24. i. . o 2 + 47TT dx / \ J x=+ oo \ d 4>c/dx 1 ' 0 x = . i.8 we know that this equation possesses two discrete eigenvalues (one being the eigenvalue zero.e.2 The inverted potential. kink.15) cosh mx From our study of the solutions of this equation in Sec. d \m .1 Introductory Remarks 623 For socalled "classical stability" the differential operator has to be semipositive definite. thus the zero mode here is given by d . 3m 2 and w\ > Am2.oo \ x== + CX) Fig.
(24. Before we consider such a theory in more detail we recall the topological aspects of a soliton or instanton.21) The charge is defined as the space integral of the time component of the current. In the pseudoparticle consideration the coordinate x plays the role of time. Then the energy satisfies f°° . Integrating the classical equation (24. We see that this pseudoparticle starts from rest at one peak of — V at x = — oo and in losing potential energy gains just enough kinetic energy to reach the second peak at x = +00.8). The associated zero mode represents the velocity of the pseudoparticle as indicated in Fig. which has the solution 4>c(x).18) This inequality is saturated by the first order BogomoVnyi equation </>'(x) = ±V2U. by Q = \(°° dxf^eoid1^) = l[M°°) ~ M°°)] = 1(2422) .20) The number 1 on the right is the socalled topological charge which is obtained as follows.e. i. 4 TTT.624 CHAPTER 24.e. 24. where it is then again at rest. 4>"{x) = U'(4>). so that its derivative representing the pseudoparticle's velocity is odd.e. Hence in this case there must be a lower state with a negative eigenvalue and an even (ground state) wave function.3 (24. F (24. We define the current k» := e^dvfc. quantum mechanics. and — clearly — we can resort to a classical interpretation of this function. The classical configuration is then called a pseudoparticle configuration or instanton. Below we shall see that the configuration described as a bounce is given by an even function of x. the associated zero mode. In this case the usual Newton equation reappears with an opposite sign of the potential. i. we obtain the BogomoVnyi inequality (<t>'{x) = V2U)2 > 0. (24. The kink solution can now be looked at as the classical solution of a onedimensional field theory.2. i. 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.e. and also represents a nodeless even wave function — in complete agreement with our expectations for a quantum mechanical ground state wave function. i.19) E((f>c) > / J00 dxV2U(f)'c(x x0) = —x &9 1. with Euclidean time.
the kink solution having a monotonic behaviour on the way from one endpoint to the other.* In the following we discuss a simple model which allows an explicit discussion of bounces.A W + £/(A0C (24.• ±00 : U(X(pc) = ^ ( 1 . H. [57] and C.A2)2 ^ 0 except for A2 = 1. T. We are here interested in these nontrivial solutions with Q / 0. M. These are the constant solutions < = ±m/g which are therefore > / called nontopological or are described as topologically trivial. [19]. (4.2 The Bounce in a Simple Example 625 For reasons discussed earlier this charge is called the topological charge. We can introduce the concept of topological stability as follows.^ 24. ^ *C. Coleman [54]. In Eq. . Callan and S.11) of this reference the number 35 of 8/35 is misprinting of 3. W. (24. (22.Q. + H.2 The Bounce in a Simple Example The concept of bounces was introduced by Coleman.24. MullerKirsten [166]. We call this topological stability. See in particular Sees. Bender and T. MullerKirsten.5. We consider the family of configurations {\<fic(x)} where A is some number. Then E[\4>c f dx .24) Thus the energy is finite only for A = ± 1 . Wu [18]. Liang and H. The theory does. 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. G. In the preceding chapter we performed this calculation with the help of the path integral method. J.23) J —( In this expression we have the potential with the behaviour (see Eq. 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. Tchrakian [208]. of course. W.37)) 4 x . We now turn to corresponding considerations for a potential which allows tunneling to infinity and hence the definition of a particle current. 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). possess other solutions which have Q — 0. Jianzu Zhang and D. Coleman [56]. J.5 = 15. 2. 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. The requirement of finite E therefore does not permit \<j)c deformations. Coleman [41]. f J. See also S.1 to 6. S.4 and 6.
We do not carry through the entire calculation to the explicit expression for the decay rate. We therefore proceed immediately to the Euclidean EulerLagrange equation for particle mass TUQ = 1 1 fdq 2\dx = V(q). we can discuss it in the context of quantum mechanics.2 . the formu